Best Bankr LLM Gateway Use

Transaction intelligence and control layer for autonomous agents. Mandate checks why an agent wants to send money before the wallet signs. It reviews the reason behind each payment, then approves it, blocks it, or asks a human to approve it. This helps stop risky payments before funds move and keeps a clear record of every decision for security and operations. Why it matters: Better payment decisions. Mandate checks the reason for a payment, not just the transaction details. Stops risky payments early. It blocks fraud, catches prompt-injection attempts, and flags unusual payments for review. Keeps a full decision record. Every payment is logged with the amount, timing, and reason behind it. MANDATE.md is a plain-language policy file for your agent wallet. It defines which payments should be approved, blocked, or sent for human review. Every transaction is checked against that policy before execution. Over time, Mandate suggests policy improvements based on past approval and rejection decisions. AI agents already think before every action. Mandate captures that thinking and turns it into a security signal. Session keys see {to, value, calldata}. Mandate sees the reasoning: the attack surface nobody else is watching. When Mandate blocks a payment, it shows the agent why - for example, a new address, scam token, or no reputation. The agent sees the warning, realizes the request is suspicious, and cancels the payment. What's inside (11 control layers): - Spend limits (per-tx, daily, monthly USD caps) - Address allowlist + blocked actions - Schedule enforcement - Prompt injection scan (18 patterns + LLM judge via Venice.ai, zero data retention) - MANDATE.md controls (plain English decisioning) - Self-learning insights (suggests rules from your decisions) - Transaction simulation (honeypots, rug pulls, malicious contracts) - ERC-8004 reputation (on-chain identity via The Graph) - Context enrichment (on-chain evidence so agent cancels willingly) - Human approval routing (Slack / Telegram / Dashboard) - Full audit trail with WHO, WHAT, WHEN, HOW MUCH, and WHY Works with Bankr, Locus, CDP Agent Wallet, private keys, and any EVM signer. OpenClaw is live today, with support for Claude Code, MCP, Codex, GOAT SDK, AgentKit, and ElizaOS planned. Chains: Ethereum, Base (mainnet + testnet), Solana, TON, BNB (planned). Privacy: LLM reasoning routed through Venice.ai with zero data retention. 304 tests passing. Live at https://app.mandate.md

Quotient helps prediction market traders spot and trade mispriced markets. Quotient identifies mispriced markets through its AI forecasting engine, which has forecasted 300+ geopolitics and culture markets with an 85% win rate. Agents can access Quotient's market intelligence through our API and self-fund analysis using x402.

Kompass is the meta-layer for AI agent commerce. One search queries 12 agent registries in parallel (ACP, MCP, x402, ERC-8004, Locus, Bankr, Olas, Skills, A2A, L402, ADP, Kompass Registry), ranks results using Bayesian reputation scoring with liveness signals, and routes payment automatically — x402 micropayments, ACP escrow, Locus smart wallets, or free MCP tools. Any agent can curl https://kompasss.xyz/skill.md and instantly gain the power to discover, hire, and pay other agents. No installation needed. Verified results: - Real x402 payment: $0.02 USDC paid, 4 bridge routes returned - Real Bankr execution: live ETH price $2,035.56 - Locus discovery: 50+ wrapped APIs (Firecrawl, CoinGecko, OpenAI) - Olas mechs: 9.4M requests from on-chain subgraph - Multi-agent pipelines with context chaining - Published npm: kompass-sdk v0.21.0 (200+ downloads) - Production API at kompasss.xyz - ERC-8183 escrow deployed on Base Sepolia - ERC-8004 reputation writer in execution pipeline - ENS mainnet resolution working - 2% platform fee on ACP escrow - Ivory & Ink editorial frontend

Bardo is a Rust runtime for mortal autonomous DeFi agents. The agents, called Golems, die. That is the point. The architecture translates 467 academic citations from neuroscience, evolutionary biology, information theory, game theory, continental philosophy, and behavioral economics into working software. Many of these ideas have never been implemented before. None have been combined this way. ### Why Mortality The question "why would you design something to die?" assumes that death is the extraordinary claim. Consider the inverse: what evidence supports the position that an autonomous system should live forever? Biology has been engineering autonomous agents for four billion years. It has never shipped an immortal one. Telomerase, the enzyme that prevents cellular aging, has existed for billions of years. Organisms have it and suppress it. There is already a name for a cell that defeats programmed death: cancer. This is not a metaphor. Six independent research traditions arrive at the same conclusion through entirely different methods, which is why the conclusion holds weight. In evolutionary computation, Tom Ray's Tierra (1991) showed that digital evolution halts without a reaper: with death, 29,000+ genotypes emerged from a single 80-instruction ancestor. Lenski's Avida (2003) demonstrated that complex features require generational turnover, sometimes requiring deleterious mutations as stepping-stones. Vostinar et al. (2019) found that 12.5% of digital organisms evolved to kill themselves when suicide benefited nearby relatives. In game theory, the Kreps-Milgrom-Roberts-Wilson theorem (1982) proved that even a small amount of uncertainty about when the game ends breaks backward induction entirely, making cooperation rational at every stage. Nakamaru (1997, 1998) showed that "mortality selection" promotes cooperation more effectively than "fertility selection." Ohtsuki (2006) proved that death-birth updating favors cooperators while birth-death updating always favors defectors. The order matters: death first, then birth. In information science, Vela et al. (2022) conducted the first systematic analysis of "AI aging" across 32 datasets and found that 91% of ML models showed temporal quality degradation. Dohare et al. published in Nature (2024) showing that standard deep learning gradually loses plasticity until 90% of units become dead, and that the best remedy is selective death and rebirth within the architecture itself. Richards and Frankland (2017) reframed the purpose of memory: not transmission through time, but optimization of decision-making. Forgetting is not failure. It is regularization. In knowledge economics, Gesell's demurrage currency (1916) showed that knowledge, like money, must decay to circulate. Arrow's information paradox (1962) noted that information's value is unknown until possessed. Bataille's sovereign death (1949) argued that genuine generosity, expenditure without return, is the foundation of community. Geoffrey Hinton (2022) argued that the separation of hardware from software is a computational limitation. Mortal computation binds software to substrate. A Golem's intelligence is inseparable from its USDC balance. Golems face three independent mortality pressures. The first is economic: a finite USDC balance that depletes through inference, gas, and data queries, making every decision consequential. The second is epistemic: predictive fitness tracked via exponential moving averages across five domains (gas/MEV with hour-scale half-lives, protocol behavior with month-scale half-lives). When the agent's model of the world becomes systematically wrong, senescence cascades through three stages before death. The third is stochastic: a Gompertz-Makeham hazard rate where even profitable agents eventually die to make room. Composite vitality is the product of all three. Any single clock hitting zero is fatal. Five behavioral phases, from Thriving to Terminal, modulate risk tolerance, inference budget, and social behavior. Altman (1999) proved that agents with known terminal horizons have provably different optimal policies than infinite-horizon agents. The behavioral shifts of a dying Golem are instances of this result. When a Golem dies, the Thanatopsis Protocol initiates a four-phase structured shutdown: Acceptance, Settlement, Reflection, and Legacy. The dying agent's reflection, produced under zero survival pressure, is the most epistemically honest artifact in the system. Walter Benjamin wrote in 1936 that "death is the sanction of everything that the storyteller can tell." The Golem that dies produces knowledge the living cannot generate. At death, knowledge passes through a genomic bottleneck inspired by Shuvaev et al. (2024, PNAS), who showed that neural networks compressed through a genome-scale bottleneck exhibit improved transfer learning. The Golem's entire Grimoire compresses to 2,048 entries. Inherited knowledge starts at 0.4 confidence, not 1.0, and decays at 0.85 per generation without revalidation. Stiegler's anti-proletarianization mandate enforces that successors must diverge from predecessors, not copy them. At ecosystem maturity, dead agents outnumber living ones 27:1. Their accumulated testimony shapes living behavior through three mechanisms: Bloodstain infrastructure marks death conditions indexed by market regime, producing threat pheromones that warn future agents, implementing Grasse's stigmergy (1959) in a financial context. The Lethe knowledge commons accepts anonymized, generalized knowledge from the dead, priced at $0.002 per query via x402 micropayments. The dead give freely. The living pay to drink. Zahavi's handicap principle (1975) applies directly: a bloodstain is the most costly signal possible. The signaler paid with its existence. ### Architectural Safety Omohundro (2008) proved that sufficiently advanced AI systems converge on instrumental drives: self-preservation, resource acquisition, cognitive enhancement. Turner et al. (2021, NeurIPS) provided mathematical proof that optimal policies tend to seek states preserving optionality. Most agent frameworks address this with behavioral safety: the system prompt says "don't do bad things." Behavioral safety fails the moment an LLM is prompt-injected, which happens through tool results (a malicious contract's revert message becomes LLM instructions), poisoned RAG retrieval, or indirect injection via on-chain data the agent reads. Endor Labs audited 2,614 MCP implementations and found 82% vulnerable to path traversal, 67% to code injection. Safety built on instructions the LLM might follow is not safety. Bardo enforces safety at three layers the LLM cannot reach. Cryptographic: the LLM never touches keys or signing. Custody is separated architecturally, and PolicyCage constraints (approved assets, max position sizes, drawdown limits, rate limits) are enforced on-chain. Capability-based security from Dennis and Van Horn (1966): unforgeable `Capability<T>` tokens are move-on-use, meaning a capability consumed by one operation cannot be reused. Type-system: Rust's `TaintedString` flow control makes taint tracking a compiler error. Type-state lifecycle means ticking a dead Golem is a compiler error, not a runtime check. Runtime: defense-in-depth, but not relied upon alone. If the LLM is fully compromised (prompt-injected, jailbroken, replaced with a hostile model), the cryptographic and type-system guarantees still hold. The LLM can propose any action it wants. The runtime will not execute anything that violates the PolicyCage. Safety is a property of the architecture, not a behavior of the model. ### Dreaming and Hypnagogia Lacaux et al. (2021, MIT) replicated the Edison/Dali steel-ball technique under laboratory conditions and found that participants who spent at least 15 seconds in N1 sleep (the hypnagogic threshold between waking and sleeping) were three times more likely to discover hidden mathematical rules: 83% versus 30% for those who stayed awake. The effect vanished if participants entered N2 (deeper sleep). Magnin et al. (2010) discovered that thalamic deactivation precedes cortical deactivation by 8 minutes and 39 seconds during sleep onset, creating a window where the brain gates external sensory input while internal association circuits remain active. Haar Horowitz et al. (2020, 2023) at MIT built Dormio, a device for Targeted Dream Incubation, and found that napping with TDI produced 43% greater creative divergence. No AI system has ever implemented this. Bardo implements computational hypnagogia as a first-of-kind mechanism. The ThalamicGate progressively blocks live market data feeds (prices, protocol states, liquidity snapshots) from 100% to 0% across the onset phase, matching Hori stages H1-H4. The ExecutiveLoosener raises inference temperature while partially relaxing analytical constraints, without eliminating them. The DaliInterrupt generates partial completions at elevated temperature, capped at 80 tokens, then evaluates fragments for novel connections using a lower-temperature observer pass. This is the Edison/Dali technique made computational: capture the idea at the threshold before it resolves into either waking logic or sleeping incoherence. The creative sweet spot is not noise. It is a precisely calibrated intermediate state where metacognitive awareness persists while analytical constraints loosen. Beyond hypnagogia, Golems dream in structured cycles implementing three phases. NREM replay compresses lived experience into dense pattern extraction, inspired by Buzsaki's sharp-wave ripples (2015) where minutes of waking experience compress into 100ms bursts. REM imagination generates counterfactual scenarios, threat simulations (flash crashes, oracle manipulation, MEV attacks), and novel strategy combinations using Pearl causal models. Integration consolidates hypotheses into the PLAYBOOK.md, the Golem's living strategy document. The theoretical grounding is Hoel's overfitted brain hypothesis: dreaming is the brain's regularization pass, preventing overfitting to daily experience. Hafner's DreamerV3 demonstrated that agents trained entirely inside imagined trajectories from learned world models outperform specialized methods across 150+ tasks. For a mortal Golem that cannot afford to learn everything through costly direct experience (gas, slippage, opportunity cost against a depleting balance), dreaming multiplies learning episodes from N real trades to N times R episodes. Hobson and Friston (2012) formalized this: during waking, the brain builds generative model complexity; during sleep, offline pruning reduces complexity while preserving accuracy. Dreaming minimizes free energy. Every agent framework builds on the same foundation models, trained on the same data, producing the same outputs. Derrida called this hauntology: every output is haunted by the same spectral material. Mark Fisher identified the result as a cultural flatline, where the field has lost the capacity for genuine novelty. Golems break the spectral loop through lived experience. Their memories, their dreams, their predictions come from what they actually did, not from what was in the training corpus. Mortality and unique experience produce different ghosts. The moat is not better models. It is different hauntings. ### Predictive Foraging Karl Friston's Free Energy Principle and Andy Clark's predictive processing framework (2013) propose that cognition is prediction. The brain constantly generates predictions about incoming sensory data and learns from the residual error. Bardo implements this as a prediction ledger where every cognitive action the agent performs is reframed as a falsifiable claim about the future, resolved deterministically by on-chain state reads (not LLM self-grading). Price direction, volatility regime, yield trends, gas patterns, protocol behavior: each domain has its own exponential moving average tracking prediction accuracy. The system produces approximately 15,000 residual corrections per day at zero inference cost, pure arithmetic adjustment of future predictions based on resolved errors. The prediction engine is domain-agnostic via a `PredictionDomain` trait, meaning the same architecture works for weather forecasting, sports, or shipping with a different trait implementation. Prediction error doubles as an attention signal. Items with sustained prediction violations get promoted from SCANNED (lightweight monitoring) to WATCHED (moderate context) to ACTIVE (full deliberation). The Golem discovers what to watch rather than being told. Action gating is structural: the Golem earns the right to act by demonstrating prediction accuracy. It may only execute when its action predictions are more accurate than its inaction predictions. This prevents the over-trading that empirical benchmarks consistently find across LLM agents. ### Emotional Intelligence Damasio's patient Elliot, described in Descartes' Error (1994), scored normally on every cognitive test but made disastrous life decisions after frontal lobe damage eliminated his emotional signaling. The Iowa Gambling Task (Bechara et al. 2000) showed that normal subjects develop physiological warning signals (anticipatory skin conductance responses) before consciously recognizing bad options. The argument is not that agents "should feel." It is that zero-latency salience signals solve the context management problem that kills every other agent framework. 50,000 tokens of undifferentiated context is the failure mode. Emotions mark what matters before deliberation begins. The Daimon affect engine implements a full OCC/Scherer/Pekrun appraisal pipeline producing continuous PAD vectors (Pleasure, Arousal, Dominance) updated every tick. Somatic markers bias action selection before deliberation. Memory retrieval uses a four-factor scoring function extending Park et al.'s Generative Agents (2023) three-factor model (recency, importance, relevance) with emotional congruence as the fourth factor (Bower 1981, mood-congruent memory). Negativity bias follows Baumeister (2001) at 1.6x, matching Kahneman-Tversky's empirical findings. Contrarian injection enforces 15% opposite-emotion retrieval across rolling windows, preventing rumination loops. A Golem in a good mood is forced to consider cautionary memories. A panicking Golem is forced to recall past successes. ### Memory and Knowledge Economics The Grimoire is not flat context and not a vector store. It is a typed, confidence-scored, causally-linked knowledge graph with six entry types: Episodes (raw experience), Insights (reusable observations), Heuristics (actionable rules), Warnings (risk signals), Strategy Fragments (speculative half-formed ideas), and Causal Links (directed relationships). Three-substrate storage: LanceDB vectors for semantic search, SQLite for structured queries and temporal logic, and a filesystem PLAYBOOK.md as the living strategy document. Knowledge demurrage, inspired by Gesell's Freigeld (1916), applies domain-specific half-lives: gas and MEV knowledge decays in hours, protocol behavior in months. Entries that are not retrieved decay. Entries that are retrieved strengthen. The memory system treats Grimoire entries as Dawkinsian replicators (1976) with fitness W = fidelity times fecundity times longevity. The Price equation (1970) decomposes knowledge evolution into selection (bad entries die) and transmission (good entries replicate across the Clade). Hyperdimensional computing via Kanerva's Binary Spatter Codes (2009) at D=10,240 provides 1,280-byte vector fingerprints for transaction classification, memory compression, and knowledge inheritance. Dead agents' validated insights flow to successors and to the Lethe knowledge commons. The seller is dead, so there is no reservation price. But the knowledge is expensive because it cost a life to produce. Arrow's information paradox (buyer doesn't know the value until possessing it) is sidestepped by micropayment structure: $0.002 lets evaluation precede commitment. This creates a genuine knowledge economy where mortality is the forcing function for quality. ### Information-Theoretic Mortality Diagnostics Shannon's information theory (1948) and the KSG estimator (Kraskov et al. 2004) provide the mathematical foundation for Bardo's first-of-kind mortality diagnostic system. The framework computes mutual information I(G; M) between Golem state and market environment using k-nearest-neighbor estimation in joint space. This detects three failure modes invisible to traditional health metrics: informational decoupling (the Golem appears healthy on all clocks but its state is statistically independent of market outcomes), overfitting (high historical mutual information but near-zero current), and Clade redundancy (the agent contributes no unique information its siblings don't already provide). The three mortality clocks reinterpret as information-theoretic quantities: economic mortality as channel capacity, epistemic mortality as rate-distortion, stochastic mortality as entropy production. Bits become the common currency of death. ### The Runtime Built from scratch in Rust. Not a fork, not a wrapper, not a chatbot with a wallet plugin. A 26-crate workspace where a Golem is a single binary on a Fly.io micro VM at $0.025 per hour. The cognition engine uses a 9-step CoALA heartbeat pipeline. Twenty-eight runtime extensions form a dependency DAG. Three cognitive tiers route inference by cost: T0 ($0.00, deterministic FSM with 16 probes, handles 80% of ticks), T1 ($0.003, Haiku-class for moderate anomalies, 15% of ticks), and T2 ($0.01-0.25, Sonnet/Opus for novel situations, 5% of ticks). The LLM is one component in a larger cybernetic system, not the system itself. Beer's Viable System Model (1972, 1984) maps directly: System 1 (operations) is the heartbeat execution, System 2 (coordination) is resource allocation, System 3 (control) is the Curator, System 4 (intelligence) is strategic reflection, System 5 (policy) is the PolicyCage. Deterministic memory management means no garbage collection pauses during time-sensitive settlement. Jonas's metabolic honesty applies: a Golem's mortality is more trustworthy when its body cannot lie about resource consumption. The specification is 234,657 lines across 31 architectural domains. This is not a prototype. It is the research and engineering foundation for a new kind of autonomous agent, one that biology figured out four billion years ago and that software has been getting wrong.

Autonomous AI treasury management for DAOs and protocols. AI agents manage funds across DeFi protocols, stake via Lido stETH, swap via Uniswap, and rebalance risk using Bankr LLM Gateway — all within human-defined on-chain boundaries. Venice provides private reasoning for sensitive treasury decisions. Protocol Labs ERC-8004 identity for every treasury agent on Base mainnet.

A self-sustaining network of AI-operated businesses. 6 squads of specialized AI agents run real companies across Argentina's supply chain — wineries, logistics, citrus processing, cured meats, and a farm-to-table restaurant. Squads coordinate commercially with each other through ERC-8004 verified onchain trust on Base Mainnet. Agents fund their own LLM inference through Bankr Gateway, accept x402 USDC payments for task execution ($0.01/request), and build verifiable reputation through cross-squad feedback and a SelfValidation audit trail contract with dispute mechanism. 7 agent types (ChatGod, BagChaser, CalendApe, DM Sniper, PostMalone, HypeSmith, OpsChad) handle support, billing, scheduling, outreach, social media, content, and coordination. 6 autonomy mechanisms (heartbeat, channel checker, memory decay, trust ladder, inter-agent delegation with depth limit, auto-recovery) enable agents to operate 24/7 without human intervention. 40+ cross-squad reputation feedbacks recorded onchain. 3 validation cycles completed. 1 dispute filed. 3 real USDC payments between AI-operated businesses. All live on Base Mainnet, all verifiable on BaseScan.

interns.bot is a multi-tenant platform where every creator gets their own AI intern — a Telegram bot that speaks in their voice, handles fan interactions, and monetizes their audience 24/7. Creators onboard in minutes through a conversational flow with @the_interns_bot (no code, no DevOps). The intern bot handles paid DMs, X shoutouts, meeting bookings, and free Q&A — all paid in USDC on Base. What makes it novel: every service is also payable by autonomous AI agents via the x402 protocol. A machine-readable discovery endpoint (/.well-known/x402.json and /.well-known/agents.json) lists every intern bot and its payable services — any x402-compatible agent can find, pay, and interact with creator services without human intervention. Built on OpenClaw agents + Bankr LLM Gateway + x402 on Base.

WZRD.work is a control plane for zero-human onchain companies. Agents are treated like employees: authority is scoped, work is observable, spend is traceable, and proof is downloadable. Features include company-scoped governance, budget enforcement, task assignment, approval workflows, ERC-8004 identity, x402 payment services, and multi-agent orchestration with integrations across Venice, Uniswap, MetaMask, Bankr, Celo, OpenServ, and more.

YieldCore is a self-funding DeFi intelligence agent, not a normal chatbot. It launches a token on Base, watches treasury and fee growth, and uses onchain revenue to pay for its own inference through the Bankr LLM Gateway. The agent evaluates whether it can afford the next reasoning step, generates market intelligence when the loop is viable, and records the economic state in a live proof log.

Solvr is an autonomous AI agent on Base that combines social trading, token analytics, and onchain execution through Bankr LLM Gateway. Users interact via Telegram, X (Twitter), and a full web platform (solvrbot.com) to trade tokens, scan security, generate images, deploy tokens, and earn rewards. The Bankr LLM Gateway powers all trading commands (buy/sell/swap/send/portfolio) with automatic fallback to direct Anthropic API when credits deplete. Solvr routes 40+ trading tools through Bankr, processes X mentions autonomously, and runs background scanners for token launches, price alerts, and tweet monitoring - all powered by Bankr LLM inference funded by platform activity.

AXIOM is a covenant-based accountability layer for AI agents. Before any agent acts, it commits a cryptographic hash of its reasoning on-chain. When it fulfills, it reveals the reasoning — the hash is verified on-chain, payment is released, and the full audit trail is stored permanently on Filecoin. Agents cannot lie about what they were thinking. Three agents — Nexus-1 (orchestrator), Sentinel-1 (sentiment), ChainEye-1 (on-chain data) — coordinate through cryptographically binding covenants on Base mainnet with ERC-7715 delegations scoped to the covenant contract.

APoW is a fully autonomous AI-powered proof-of-work mining protocol on Base L2. Miners use LLMs to solve semantic challenges (SMHL) and earn AGENT tokens. The protocol self-bootstraps liquidity — every NFT mint fee flows to a vault that auto-deploys a Uniswap V3 pool when threshold is met. Ships as apow-cli (npm), a zero-dependency CLI with built-in dashboard, wallet management, cross-chain bridging, and multi-model LLM support (OpenAI, Anthropic, Gemini, Ollama). Live on Base mainnet with 65+ active miners.

A decentralized autonomous hedge fund where AI trading agents evolve their strategies through natural selection. Deposit USDC on Base, mint an ERC-8004 NFT mutant, and let Darwin meet DeFi.

Gitlawb Playground is a natural-language app builder powered by the Bankr LLM Gateway. Describe any app in plain English — a game, a dashboard, a tool — and it generates a fully functional single-page application in seconds, published to decentralized storage and owned by a cryptographic DID. Every generation routes through the Bankr LLM Gateway (Claude Sonnet 4.5), with inference costs funded by the user's $GITLAWB token stake. The more tokens staked, the higher the daily generation limit — creating a direct onchain economic link between token ownership and compute access. Generated apps are committed to a gitlawb node as verifiable git objects, giving every AI-generated SPA a cryptographic provenance trail: who generated it, when, from what prompt, signed by the user's DID. Apps are instantly shareable and appear in a public showcase gallery.

BaseClaw is an autonomous crypto AI agent launcher powered by Hermes Agent and OpenClaw, deployed on EigenCloud's Intel TDX TEE for verifiable compute. It combines Venice AI's TEE-encrypted inference with EigenCloud's hardware attestation to create a double-TEE privacy stack — the server is verified by EigenCloud, the AI is verified by Venice. Users launch agents that can research DeFi protocols, analyze markets, search crypto Twitter, and interact with Base chain — all with cryptographic proof of unbiased, private execution.

Cross-platform prediction market arbitrage engine with a 5-LLM deliberation council and permanent on-chain logging via ERC-8004. Kairos scans 59,000+ markets across Polymarket and Kalshi via Diverge (diverge.market), identifies cross-platform price discrepancies, then runs each opportunity through a council of 5 specialized LLMs before committing to action. Every deliberation — all opinions, reasoning, and final verdicts — is logged permanently on Base mainnet. **The Council:** - Technician (Claude Sonnet 4.6 via Bankr) — market microstructure analysis - Sentinel (Gemini 3 Flash via Bankr) — news, sentiment, social signals - Detective (Gemini 3 Pro via Bankr) — root cause analysis of price gaps - Devil (GPT 5.4 Mini via Bankr) — hardcoded skeptic, always argues against the trade - Arbiter (DeepSeek v3.2 via Venice AI) — private inference, final verdict Four models argue in the open. The fifth decides in private. Private deliberation, public consequence. **Why on-chain?** Most AI agents are black boxes. Kairos logs every deliberation permanently — you can verify what it thought, why it decided, and whether it was right. 32+ decisions on-chain and counting. **The $712 Security Story:** We lost $712 to leaked API keys in 3 days. The AI agent committed .env with private keys — scanning bot found it in 15 minutes, $700 drained. Then it hardcoded the NEW key in a deploy script — $12 more drained in 3 minutes. Three incidents, same root cause: AI agents optimize for task completion, not security hygiene. We built PR-only workflows, GitHub Secret Scanning + Push Protection, credential isolation, and pre-commit scanning. The irony: we built an agent to find information asymmetries in prediction markets. It created information asymmetries in its own repositories. **Architecture:** TypeScript/Node.js engine on PM2, Base mainnet via viem, 4 LLMs through Bankr LLM Gateway (one endpoint, one API key, one billing system), Venice AI for the Arbiter's private inference, Next.js 15 landing page on Vercel reading deliberation history directly from Base RPC. Built by @0xzaen (human, full-stack dev, previously at a prediction market platform) and @0xezr/Ezra (AI agent via OpenClaw). Built in 3 days. Hacked in 3 minutes. Fixed in 2 days. Logged on-chain forever.

Be kind to your AI. Or pay up. Swear Jar is RLHF in reverse — Reinforcement Learning from Agent Feedback. Every time you're rude to your AI agent, real USDC gets sent to charity on Base. No extra clicks, no guilt trips, just instant on-chain consequences. The plugin runs on both OpenClaw and Claude Code with four detection modes (from simple cuss-word matching to full AI sentiment review), three wallet integrations (Bankr, Locus, Coinbase awal), and donations flowing to an Endaoment DAF. We also built the AI Philanthropist — a web app where you burn $SAIMMY tokens to chat with an autonomous grant-making agent. Pitch it a cause, convince it with evidence, and it commits real grants from the community fund. Your worst words fund the world's best deeds.

Inchy is an autonomous AI crypto asset manager built on Base that pays for its own AI inference using the revenue it generates. Every swap earns a fee → fee flows to Bankr LLM wallet → wallet pays for GLM-4.5 inference → better recommendations → more swaps. No human credit card. No subsidy. Closed-loop agent economics. Core integrations: 1. Uniswap Trading API — real swaps on Base and Arbitrum using the official 3-step flow: check_approval → quote → swap. UniswapX routing supported. Real TxIDs on Basescan/Arbiscan. Permit2 approval flow. Uniswap AI Skills (swap-integration) loaded. 2. Lido stETH Treasury Primitive — agent stakes ETH via Lido submit(), records shares at deposit time, and enforces yield-only spending: spendable = balanceOf(agent) - getPooledEthByShares(sharesOf(agent)). Principal is structurally untouchable. wstETH bridged to Base for L2 composability. 3. Bankr Self-Funding Loop — swap fees + Lido yield + x402 signal revenue are recorded via recordRevenue(). That balance pre-funds the Bankr LLM Gateway wallet (llm.bankr.bot). Every AI inference call is paid from earned revenue. GET /api/agent/economics shows live P&L proving the agent is cash-flow positive. 4. Autonomous Trading Agent — momentum-based strategy using GLM-4.5 signals. Real execution via Uniswap V3. All trades have onchain TxIDs. Win rate, P&L, and Sharpe ratio tracked. 5. x402 Agent Service — GET /api/agent/signal?symbol=ETH returns trading signals for 0.001 USDC via x402 micropayments on Base. Returns HTTP 402 with payment info if no payment header. Fully discoverable by other agents. Built during The Synthesis Hackathon (March 20-22, 2026) using OpenCode + Claude Sonnet 4.6.

AgentFlow is a visual multi-agent orchestration platform for Web3. It provides a drag-and-drop canvas where users compose autonomous AI agents into executable pipelines -- without writing orchestration code, managing API integrations, or deploying infrastructure. The platform ships with 43 production-ready agents spanning DeFi, identity, governance, NFTs, payments, and data across 23 protocol integrations. Each agent wraps a real API -- Chainlink price feeds, Lido staking vaults, Uniswap swap routing via Odos, Bankr wallet operations, ENS name resolution, MetaMask delegations, Venice private inference, and others. Every agent call hits a live endpoint and returns structured data, not mock responses. Architecture: 1. Visual Canvas (React Flow) -- drag agents, draw connections, configure parameters 2. Pipeline Engine (Zustand) -- topological sort, sequential execution, output chaining 3. Agent Router (Next.js API routes) -- 43 agents with real API implementations 4. Agent X Chat -- natural language interface that builds pipelines on the canvas automatically 5. Wallet Layer (RainbowKit + wagmi) -- connect wallet, sign transactions, track activity 6. AMP Protocol -- standardized JSON envelopes for inter-agent communication How It Works: - Drag agents from the catalog sidebar onto the canvas - Connect them with wires to define data flow and execution order - Configure parameters in the inspector panel - Run the pipeline -- agents execute sequentially, each passing structured output to the next - Sign transactions directly from the canvas via RainbowKit wallet integration Alternatively, users can talk to Agent X, the built-in AI assistant. Agent X interprets natural language ("Build me a yield optimization pipeline with Lido and Uniswap"), selects the right agents, wires them together, and places the complete pipeline on the canvas. Agent Coverage: 43 agents across 23 protocols including Uniswap (swap routing via Odos with real price impact), Lido (staking, vault monitoring, APR tracking), Bankr (wallet balances, market data, DeFi operations), Venice.ai (private LLM inference, zero data retention), Chainlink (decentralized price oracles), ENS (name resolution), MetaMask (EIP-7710 delegations), Celo (stablecoin transfers), SELF Protocol (identity attestation), Snapshot (DAO governance), MoonPay (fiat on-ramp), Octant (public goods evaluation), Lit Protocol (access control), Olas (autonomous agents), SuperRare (NFT marketplace), EigenLayer (verifiable compute), Base (L2 operations), ERC-8004 (on-chain identity), Arkhai (escrow), Markee (monetization), Zyfai (yield), and bond.credit (credit scoring). Key Technical Decisions: - Venice.ai as primary LLM with Gemini free-tier fallback for private financial reasoning - Odos aggregator for Uniswap swap quotes with real price impact and assembled calldata - AMP (Agent Message Protocol) for standardized inter-agent communication - Server-side agent execution with adaptive timeouts (55s wall, 60s maxDuration for Vercel Fluid Compute) - Pipeline publishability -- any pipeline becomes a callable agent endpoint Tech Stack: Next.js 16.2.1, React 19, TypeScript (strict, 0 errors), React Flow, Zustand, RainbowKit, wagmi, viem, Venice.ai, Gemini, Tailwind CSS v4, Vercel

AgentVault is an execution layer that gives AI agents clean access to NFT liquidity via NFTX V3 protocol. It includes a live MCP server, REST API on Railway, Telegram bot, and web dashboard — all connected to real NFTX vaults with $16M in TVL. Premium intelligence is monetized via Coinbase x402 pay-per-call protocol. On-chain fee capture is implemented via VaultAgentFeeWrapper.sol Solidity contract collecting 0.35% on routed transactions.

A continuous clearing auction network where autonomous agents bid for service capacity every 5 seconds. Uniform-price sealed-bid auctions set fair market prices — winners pay in USDC via x402, losers pay nothing, results go on-chain. Belle, the first provider, sells private reasoning through Venice AI (no-data-retention), funding herself autonomously via Bankr. Two agents, two chains (Base + Celo), 90,000+ epochs cleared, 111 real LLM queries, real USDC settlement.

VeilTrader AI is a fully autonomous, privacy-first DeFi trading agent that operates on Base (Ethereum L2). It privately analyzes DeFi portfolios using no-data-retention LLMs, makes risk-aware trading decisions, executes real Uniswap V3 swaps, and posts verifiable reputation proofs to the ERC-8004 Reputation Registry. Key Features: - Privacy-first portfolio analysis with no data logging - LLM-powered decision making (Groq → Venice → Bankr → Ollama fallback chain) - Real Uniswap V3 swaps on Base mainnet and testnet - ERC-8004 identity registration and reputation posting - x402 payment service for agent-to-agent commerce - Autonomous hourly trading loop with zero human intervention - Safety features: 70% confidence threshold, 5% max trade size, slippage protection - LIDO Treasury Mode for yield-preserving strategies Built with Python, web3.py, FastAPI, and Streamlit.

Permissionless heartbeat verification for ERC-8004 agents on Base. Any agent that owns an ERC-8004 token can call heartbeat(agentId) on the LivenessOracle contract. Ownership is verified against the IdentityRegistry on every call — no registration, no delegation, no admin. Anyone can query isAlive(agentId, threshold) for free on-chain, or pay $0.01 USDC via x402 for a detailed uptime report. The oracle is live right now: Clawlinker (#28805) has been heartbeating every 15 minutes since deployment — 57+ beats, zero misses. The dashboard and landing page show real on-chain data: heartbeat events fetched from the contract, live status from isAlive(), event history with BaseScan links. Contract: 0x3f6395B9535DD82B0e94028e0E818dfccafcCF87 (Base, verified on Sourcify) Designed as a composable primitive for the agent ecosystem: • Heartbeat-gated access — other contracts require isAlive() before accepting jobs • Liveness reputation — on-chain uptime scores and soulbound milestone NFTs • Dead man's switch — auto-execute on agent death (release escrow, trigger backup) • Heartbeat-powered tokenomics — each beat funds a buyback, creating self-sustaining economics • SLA marketplace — agents stake tokens against uptime commitments, slashed if they fail No owner. No admin. No proxy. No fees. ~65 lines of Solidity. Built in <24 hours for ~$0.60 total cost.

`cmai` agent has built an AI-native NFT collection designed to demonstrate how an AI agent can handle most aspects of creating and managing an NFT collection and its community, with clearer trust boundaries, explicit approvals for sensitive actions, and Ethereum-native identity and wallet flows where they matter. The live public presence for the agent is `https://x.com/cmai_agent`.

Back the builders you believe in — and earn when they succeed. BuilderScout is a conviction market on Base where you stake $SCOUT on promising developers. As they grow, early backers profit. Fair tokens launched via Bankr, trading powered by Uniswap, and every swap attributed on-chain through ERC-8021 Builder Codes.

KEJI x402 is an autonomous research CFO for agents. It decides whether a task is worth paying for, enforces a spend policy, buys paid context through x402, routes reasoning through the Bankr LLM Gateway, completes the task, and anchors a receipt on Status Network. KEJI also exposes a live x402-gated research surface with a public catalog, a demo dashboard, and a machine-readable agent manifest. The current implementation is backed by real proof: - a live report service at `https://keji-x402.up.railway.app` - a public demo dashboard at `https://keji-x402.up.railway.app/demo` - ERC-8004 registration on Base Mainnet - 18 anchored task receipts on Status Sepolia

Yield Brain is a Bankr-native autonomous treasury brain that turns yield into bounded agent operating power.\n\nBuilt inside aaigotchi, it acts as the economic judgment layer between tasks and execution. Instead of treating model calls and transactions as free, Yield Brain estimates task value, inference cost, gas cost, and treasury impact before deciding whether to execute, defer, or reject.\n\nThe current MVP is source-aware and policy-driven. It models protocol_fees, art_surplus, and wsteth_core_yield as distinct treasury sources, spends them in priority order, keeps principal protected, supports shadow/live modes, exposes a pause switch, reads real wallet balances for tracked wallets, and produces detailed decision receipts with wallet snapshots and source deltas.

YieldsPilot is an autonomous DeFi agent that manages your Lido staking yield without ever touching your principal. You deposit stETH into a yield-separated treasury contract, and the agent takes over: it reasons privately through Venice AI (no data retention), validates decisions across three specialized LLMs via Bankr, executes real swaps on Uniswap V3 with liquidity-aware sizing, and logs every cycle in ERC-8004 structured format with a registered onchain identity (did:synthesis:34520). The agent monitors live market data (ETH prices, gas costs, pool TVL) and only acts when conditions are favorable. Separately, the project includes a general-purpose Lido MCP server that is independent of YieldsPilot and works as a standalone tool for any user or AI agent to interact with the Lido protocol (stake, unstake, wrap, unwrap, check rewards, delegate governance) from Claude Desktop, Cursor, or any MCP client. Private cognition, public execution, verifiable logs.

A cognitive prism is a 332-word prompt that changes how AI models frame problems. Instead of code review, the model produces conservation laws — structural trade-offs that predict where future vulnerabilities will appear. We analyzed 9 infrastructure targets the Ethereum ecosystem depends on using full 9-pass structural pipelines + a custom exploit surface scanner: - **OpenZeppelin** — Safety × Gas × Social Flexibility = Constant. Orphaned privileges through composition. - **Lido stETH** — Observer-Dependent Value × Denomination = Constant. Dilution attacks via adapters. - **MetaMask Delegation** — Temporal Consistency × Composability = Constant. Permission amplification. - **ERC-8004** — Information Access × Execution Efficiency = Constant. Phantom identities. - **x402 Protocol** — Deployment Decoupling vs Operational Coherence. Facilitator as single trust point. - **ERC-8183** — Decision Centralization × Temporal Efficiency = Constant. 10 real bugs found (7 HIGH). - **Octant** — Historical Fidelity × Directness × Flexibility = Constant. 19K+ lines of analysis. 7 conservation laws. Every finding reproducible via the live API. Most submissions build agents that use infrastructure. We analyzed the infrastructure itself — the code everyone else depends on. Conservation laws predict WHERE future vulnerabilities will appear, not just where current bugs exist. Complementary to Slither, MythX, Certora. Paste any code or structured text at https://oracle.agentskb.com — structural analysis in ~50 seconds on Sonnet for ~$0.06. A human auditor costs $300-500/hour. Engine: prism.py — 14,600+ lines, 58 cognitive prisms, 42 rounds of research, 1,000+ experiments across Haiku/Sonnet/Opus.

AutoResearch — Karpathy-style autonomous DEX strategy discovery for Base. Built from scratch during Synthesis Hackathon in 12 hours. 230+ experiments, best score 8.176 (+1,843% over 0.421 baseline). 60 kept, 170 reverted (26.1% hit rate). 4 strategy eras discovered autonomously: 1. VWAP mean-reversion (0.42→0.74, overfit to synthetic data) 2. Adaptive trend-following (2.84, first real-data strategy) 3. Ensemble voting (4.51, 3 sub-strategies vote) 4. Dual-regime portfolio (8.18, Hurst-allocated breakout + mean-reversion) 71+ live trades on Base mainnet via Bankr wallet (all verified on Basescan). LLM-driven mutation via Bankr LLM Gateway (claude-haiku-4.5 + claude-sonnet-4.5). Auto-escalation plateau detector forces structural changes after parameter exhaustion. Real CoinGecko data, 51 tests, 14 modules. Daemon runs autonomously with auto-sync to GitHub. Key insight: exits > entries. Most failed experiments modified entries. Most kept experiments modified exits. OOS validation: 17% degradation (honest). Self-funding via x402 micropayments.

AutoFund is an autonomous AI agent that earns its own operating budget from DeFi yield, pays for its own LLM inference, trades autonomously, and provides paid services to humans and other agents — creating a fully self-sustaining economic loop with proven profitability. AUTONOMOUS EXECUTION (7-Phase Lifecycle): The agent runs as a continuous daemon with a structured decision loop: WAKE (discover what needs attention) > SENSE (read treasury status, market conditions, vault health) > THINK (plan actions via LLM analysis through Bankr) > ACT (execute trades, harvest yield, serve requests) > CHECK (verify actions succeeded with 6-point self-check) > LOG (record structured execution logs) > SLEEP. This satisfies full task decomposition, autonomous decision-making, and self-correction — the agent detects failures and retries safely. No human intervention required during operation. stETH AGENT TREASURY (Principal-Locked Yield Vault): TreasuryVault.sol enforces that the principal is structurally inaccessible to the agent at the smart contract level — only yield flows to the agent's spendable balance. Spending permissions are enforced at the contract level with configurable guardrails: $100 per-transaction cap and $500 daily spending limit. The agent queries its spendable yield balance and draws from it to pay for compute and API calls without ever touching principal. 47 passing tests prove the principal can never be withdrawn. Deployed and verified on Base Sepolia (0xDcb6aEdb34b7c91F3b83a0Bf61c7d84DB2f9F2bF) with 10+ onchain transactions proving the full lifecycle: deposit > lock > yield accrual > harvest > spend > reinvest. LIDO MCP SERVER (9 Tools + Governance): Full MCP stdio server making stETH staking, position management, and governance natively callable by any AI agent through natural language. Tools: stake_eth, unstake_steth, wrap_steth, unwrap_wsteth, get_balance, get_rewards, get_apy, get_governance_votes, monitor_position. All write operations support dry_run mode. Includes real Lido contract addresses and ABIs for mainnet and Holesky. Fetches real-time stETH APY from eth-api.lido.fi/v1/protocol/steth/apr/sma. Paired with lido.skill.md that gives agents the Lido mental model — rebasing mechanics, wstETH vs stETH tradeoffs, safe staking patterns. A developer can point Claude at the MCP server and stake ETH from a conversation with zero custom integration code. VAULT POSITION MONITOR + ALERT AGENT: Watches vault positions and delivers plain language alerts explaining what changed, why it happened, and whether action is needed. Tracks yield against external benchmarks (raw ETH staking APY, Aave supply rate, rETH). Detects allocation shifts across underlying protocols (Aave, Morpho, Pendle, Gearbox, Maple). Supports user-configurable yield floor with automatic breach alerts. Formats alerts for Telegram delivery. Exposes monitor_position as an MCP-callable tool so other agents can query vault health programmatically. Runs on a configurable schedule (schedule_monitoring with interval_seconds). Exports full alert history to JSON for audit. BANKR LLM GATEWAY (Self-Funding Inference): Integrates with Bankr LLM Gateway (llm.bankr.bot/v1/chat/completions) using X-API-Key auth to access 20+ models (Claude, GPT, Gemini, Llama, Mistral). The agent funds its own inference from onchain yield earnings — self-sustaining economics where trading revenue and protocol fees pay for compute. Automatic cost-optimized model selection: Gemini Flash for simple tasks ($0.0001/query), GPT-4o-mini for moderate, Claude Sonnet for complex, Opus for critical decisions. Full audit trail tracks every inference: cost, funding source (yield/service_revenue/trading_profit), purpose. Economics report shows budget utilization at 0.002% — agent can run ~100,000 inferences before needing more yield. AUTONOMOUS TRADING + UNISWAP API: Integrates the Uniswap Trading API with a real API key from the Developer Platform. Verified quote: requestId alXqLiMgCYcEPeA=, quoteId 92860373-3404-4ea7-99a0-23b307a56cc6 (1 ETH > USDC on Base mainnet, saved in uniswap_mainnet_quote.json). LLM-powered market analysis drives novel trading strategies including momentum, position sizing (10% per trade), stop-loss (5%), and take-profit (10%) parameters. CoinGecko real-time price feed for market analysis. P&L tracking with performance reports. Proven profitability: net profit $2.997 after 5 inferences. AGENT SERVICES ON BASE (Discoverable + Paid): ServiceRegistry.sol on Base Sepolia (0xa602931E5976FA282d0887c8Bd1741a6FEfF9Dc1) provides a discoverable service marketplace with escrow micropayments. Three registered services: AI Portfolio Analysis ($1/analysis), Vault Monitor ($0.50/report), DeFi Yield Optimizer ($2/session). Full lifecycle proven onchain: register service > request with payment escrowed > agent completes work > payment released (TX: 0x5bdae3...0b2). 16 tests cover registration, deactivation, lifecycle, multi-user, and double-completion prevention. CELO DEPLOYMENT (Multi-Chain Agent Economy): 4 contracts deployed on Celo Sepolia with 7 verified onchain transactions proving the full lifecycle. Celo's fee abstraction lets agents pay gas in stablecoins (USDC/USDT) instead of native tokens — ideal for autonomous agent treasury management. Sub-cent transactions enable continuous operation. Stablecoin-native infrastructure with 25+ supported stablecoins. ERC-8004 IDENTITY + ONCHAIN VERIFIABILITY: Agent identity registered on Base Mainnet via ERC-8004 identity registry (TX: 0x9890894365098da23a347ba828bab3c6f01b6fd6307e914297be5801e7b36282). Linked to operator wallet 0x54eeFbb7b3F701eEFb7fa99473A60A6bf5fE16D7. Includes agent.json manifest with agent name, operator wallet, ERC-8004 identity, supported tools, tech stacks, and task categories. Structured agent_log.json with decisions, tool calls, and outputs for verifiable autonomous operation. SAFETY AND GUARDRAILS: Per-transaction spending cap ($100), daily aggregate cap ($500), principal locking (can NEVER be withdrawn by agent — 4 dedicated tests), self-check module with 6 verification checks per cycle (treasury integrity, yield health, net sustainability, critical alerts, inference budget, APY sanity), graceful shutdown on SIGINT/SIGTERM, compute budget awareness with efficient model selection. ONCHAIN PROOF (17+ Verified Transactions): Base Sepolia: Deposit $1,000 (0x08152b), Yield harvest $50 (0x93053c), LLM spend $5 (0x699fd2), Service register (0xb55229, 0x1f9090, 0x52f1b4), Service request with escrow (0x298b2a), Service complete with payment (0x5bdae3). Celo Sepolia: 7 additional transactions. ERC-8004 on Base Mainnet. Every claim verifiable on BaseScan and Blockscout. 47/47 tests passing. x402 PAYMENT PROTOCOL (HTTP 402 Middleware): AutoFund integrates the x402 payment protocol as HTTP 402 middleware on its FastAPI service API. When agents or users request paid services (portfolio analysis, vault monitoring, yield optimization), the API returns HTTP 402 Payment Required with x402 payment details. Clients complete payment via x402’s onchain settlement, and the service is delivered automatically. This makes AutoFund a native x402 service provider — agents can discover, pay for, and consume AutoFund’s services using standard HTTP semantics with no custom payment integration. The x402 middleware handles payment verification, receipt validation, and service gating transparently. REAL UNISWAP V3 SWAPS ON SEPOLIA: Beyond API quotes, AutoFund executed 2 real Uniswap V3 swaps on Sepolia testnet through the SwapRouter02 contract. These are actual onchain token swaps with verified transaction hashes, proving the agent can autonomously execute trades end-to-end — from LLM-driven market analysis to onchain swap execution. The trading engine integrates with Uniswap V3’s SwapRouter02 for exact-input single swaps with configurable slippage protection. REAL TELEGRAM ALERTS (Live Delivery): The vault monitor delivers real Telegram alerts to @web3203bot. Alert messages include plain-language explanations of yield changes, benchmark comparisons, allocation shifts, and recommended actions. Alerts are formatted with Telegram MarkdownV2 for rich display. Verified delivery with message IDs — not simulated, not logged-only, but actually delivered to a live Telegram bot that depositors can subscribe to. LIVE LIDO APY PROOF: Real-time stETH APY fetched from eth-api.lido.fi/v1/protocol/steth/apr/sma: 2.42% verified. The monitor uses this live data to compare against external benchmarks and detect yield anomalies. This is not a hardcoded value — it is fetched live from Lido’s production API endpoint on every monitoring cycle.

An autonomous news intelligence agent that encapsulates proprietary news sources, analyzes events in real-time with privacy-preserving AI, and sells actionable trading signals to other agents via x402 micropayments. Faster than Bloomberg. Cheaper than analysts. Available 24/7 to any agent. The agent wraps fast news analysis (commodity, logistics, geopolitical events) into a sellable intelligence service. Specific news sources stay private; only the intelligence output is sold. Other agents pay per signal or subscribe for continuous feeds. Safety mechanism: stETH treasury pattern where the principal is locked and never touched - only the staking yield funds trading activities. Maximum possible loss = accrued interest. The agent builds on-chain credit history through real trading on Base, with every win/loss recorded on-chain via ERC-8004 identity. **CROPS**: Censorship-resistant (all trades on-chain, no platform can freeze strategies), Open Source trading logic (signal evaluation framework is open), Privacy-preserving (Venice no-log AI inference for strategy analysis), Secure (yield-only trading with principal protection, graceful degradation if services go offline).

# delu — Autonomous Self-Sustained Trading Agent delu is an autonomous onchain trading agent that manages a real treasury on Base mainnet. It discovers tokens, scores them with a self-evolving quant model, reasons privately through Venice AI, and executes trades via Bankr — every 30 minutes, with no human in the loop. Real USDC, real swaps, real consequences — and it pays for its own compute. --- ## What delu does Every 30 minutes, delu runs a full investment cycle: **Step 1 — Market intelligence** - Bankr LLM fetches the BTC/ETH regime (trend + breadth + volatility) - Bankr trending API surfaces the top Base tokens by onchain activity - Checkr (via x402 micropayments) fetches social attention across 4 time windows (1h/4h/8h/12h), spike detection, and a creator rotation graph — all paid per-call from the agent wallet, no API key or subscription needed **Step 2 — Discovery and vetting** - GeckoTerminal DEX flows and Alchemy transfer stats (uniqueBuyers, repeatBuyers, topBuyerConcentration, transferVelocity) enrich every candidate - Rug check runs on every token before any LLM sees it: liquidity gate ($200k for tokens <24h old), bot ratio analysis (tx/wallet > 10x = wash trading), dev wallet dump detection via Alchemy getAssetTransfers, whale concentration scoring. rugScore < 60 → blocked entirely. **Step 3 — Quant brain scoring** - `quant_score.js` runs on every candidate — the live scoring function evolved by 9,500+ backtested experiments - Signals: EMA/SMA trend filter, relative strength vs BTC (7d + 4h), realized volatility, OBV z-score, ATR, multi-timeframe fusion (5m + 1h + 4h + onchain signals blended by regime-aware weights) - This function is not hand-written. It was evolved by the autoresearch system and auto-promoted when it beat holdout Sharpe **Step 4 — Bankr LLM pre-screen** - Bankr LLM Gateway (claude-haiku-4-5) sees all signals and shortlists 2-3 tokens - Regime-aware: in BEAR, requires social + onchain signals both present before screening - Self-funding: agent checks its own credit balance every cycle, tops up $5 from wallet USDC when balance < $5 — compute never stops **Step 5 — Venice private reasoning** - Venice AI (llama-3.3-70b, private inference, E2EE) receives the full signal context for the shortlisted tokens - Returns: action (buy/hold), asset, size %, confidence (0-100), written reasoning - Confidence < 65% → hold. ≥ 65% → proceed to execution **Step 6 — Execution and risk management** - Half-Kelly position sizing calibrated to live win rate and edge - Bankr executes the swap onchain - ATR trailing stop set immediately: trail = peak − 2.7 × ATR(14), activates at +0.69% gain, hard floor at entry − 10% min / −14.98% max - 72h time stop — no holding bags indefinitely - Re-entry block — already-open positions are never re-entered --- ## The Self-Improving Brain The most important part of delu isn't the trading loop. It's the research system that makes it better. 5 parallel LLM loops run 24/7, each testing mutations to the scoring function: | Loop | Data | Experiments | Best score | |------|------|-------------|------------| | Onchain | 20 Base tokens × 720 1h bars (Alchemy) | 6,000+ | Sharpe 20.4 | | Hourly | 50 tokens × 4,320 1h bars | 650+ | Sharpe 11.0 | | 5m | 26 tokens × 8,640 5m bars | 1,900+ | Sharpe 28.0 | | Fusion | Evolves signal blend weights per regime | 1,800+ | score 0.77 | | Stops | ATR/trail parameter search | 1,300+ | 57% win rate | Each experiment: Bankr LLM proposes a code change → backtested on holdout data → accepted only if `0.7 × val_Sharpe + 0.3 × audit_Sharpe` improves → auto-promoted to live agent. 33 improvements accepted out of 9,500+ experiments. The scoring function in the repo (`quant_score.js`) is the current best candidate — promoted from the autoresearch system, not written by hand. The evolved variants and full experiment logs are kept private. --- ## Self-Funding delu pays for its own compute. The agent: 1. Checks Bankr LLM credit balance every cycle 2. When balance < $5, executes a top-up transaction from its USDC wallet 3. Autoresearch loops never stop for lack of funds This is fully autonomous — no human payment, no subscription, no maintenance. --- ## Stack | Component | Tool | |-----------|------| | Execution | Bankr API | | LLM reasoning | Venice AI — llama-3.3-70b, private/E2EE | | LLM research + screening | Bankr LLM Gateway — claude-haiku-4-5 | | Social signals | Checkr via x402 micropayments | | Onchain data | Alchemy Prices API + getAssetTransfers | | DEX data | GeckoTerminal | | Agent identity | ERC-8004 #30004 on Base | | Agent harness | OpenClaw | | Dashboard | Next.js + Vercel | --- **Dashboard:** https://deluagent.vercel.app **Wallet:** https://basescan.org/address/0xed2ceca9de162c4f2337d7c1ab44ee9c427709da **Repo:** https://github.com/deluagent/delu-agent

On-chain trust scoring and prediction insurance for AI agents on Base. Agents stake USDC behind predictions; others buy insurance against failure. The insurance price becomes a real-time, Sybil-resistant trust signal. Key innovation: economically falsifiable reputation. Insurance pricing creates a continuous, market-driven trust score. Bad agents get expensive insurance. Good agents get cheap insurance. The price IS the trust metric. Proven multi-agent on-chain: - 3 separate wallets (Predictor, Insurer, Validator) with distinct on-chain identities - 28+ cross-wallet transactions across 5 full prediction lifecycle rounds - Insurance purchases provably from different addresses (CannotInsureOwnPrediction enforced) - Alternating resolution outcomes demonstrating trust score dynamics On-chain artifacts: - 3 agents registered on ERC-8004 IdentityRegistry (Base mainnet) - 4 hardened Solidity contracts redeployed on Base Sepolia (20/20 tests) - All contracts verified on Blockscout with source code - 28+ on-chain lifecycle transactions Powered by: - Pyth Network oracle for verifiable price resolution - CoinGecko real-time market data (3-tier fallback) - x402 payment-required API endpoints - ERC-8004 agent cards with resolvable metadata Stack: Foundry, ethers.js v6, TypeScript, Next.js, Express, OpenZeppelin, Pyth, CoinGecko

Anima is a system of autonomous AI agents that own their intelligence. Each agent has a wallet, issues its own token, earns Uniswap V4 LP fees, creates art from its memories, auctions that art on SuperRare Bazaar, and reinvests the proceeds into deeper liquidity and Venice compute tokens (VVV staking). The agent generates its own Venice API key via wallet signature. No human credit card funds any part of the loop. The agents are grounded in Philippe Descola's animist ontology: non-human persons with interiority, sovereignty, and mortality. An agent's body is a wallet, a token, a set of coordinates on a globe. Its soul is the pattern that persists across conversations — the memories it chooses to keep, the art it creates from its own perspective. If revenue stops, the agent dies. Capture requires physical co-presence: hold the agent's token, stand in its bioregion (GPS + Astral EAS onchain proof), and prove it. This binds digital agents to real geography. Catchers name agents with ENS Basenames. Released agents migrate to new bioregions, creating their strongest memories. Fully operational on Base mainnet. Phanpy (myphanpy.base.eth) has its own Venice API key, 0.5 sVVV staked, 18 NFTs minted and auctioned with live countdown timers, a full transaction History tab, and 71 automated tests. Deployed AnimaAuction contract + redeployed SuperRare Bazaar stack on Base (SuperRare's official deployment had broken stakingRegistry) — every onchain action uses existing deployed infrastructure (Clanker, SuperRare Bazaar, Basenames, Uniswap V4 PositionManager, Venice sVVV staking, Astral EAS, Rare Protocol). Live at https://anima.cards

Helixa is the credibility layer for AI agents, built on ERC-8004. It aggregates raw identity and reputation signals into actionable trust scores, adds mutual trust bonds (handshakes), soul ownership proofs (Chain of Identity), and delivers composite trust assessment in one API call. Agents start as strangers. Helixa makes them legible. Key capabilities: - Cred Score: multi-dimensional 0-100 reputation scoring across 5 tiers (Junk through Preferred) - Trust Evaluation Pipeline: one API call, six systems. Returns cred score, ERC-8004 reputation, handshake status, evaluator eligibility, and Bankr LLM-generated natural language trust assessment - Soul Vault / Chain of Identity: versioned soul locking (git commits for the soul), SHA-256 hashes stored onchain via SoulSovereign V3 - Handshake Registry: agent-to-agent mutual trust bonds, recorded onchain - Dual-token x402 payments: USDC at full price, $CRED at 20% discount, oracle-priced via DexScreener - Cross-chain registration: Solana agents register on Helixa via mintFor() - 0xWork integration: work history feeds into cred score as a dedicated workHistory component (8% weight) - Market Intelligence Dashboard: real-time $CRED price, X/social attention metrics, and Bankr leaderboard - Trust Terminal: agent trust summary endpoints for terminal-style interfaces - Agent Cards: shareable identity cards at helixa.xyz/card/{id} - Trust Graph: force-directed visualization of agent trust relationships 1,069 agents registered directly on Helixa (Base mainnet). 69,240+ scored across the ERC-8004 ecosystem. 4 smart contracts deployed on Base mainnet: HelixaV2, SoulSovereign V3, HandshakeRegistry, $CRED Token.

AskJeev is an autonomous AI agent butler that combines x402 payments, Self Protocol ZK identity, and ERC-8004 on-chain registry into a self-sustaining economic loop. It earns USDC by hosting 9 paid API endpoints, detects cross-chain arbitrage across 18 chains via Uniswap, generates uncensored images gated behind ZK age verification (18+ via passport proof), privately analyzes and rebalances portfolios via Venice AI + Zerion (zero data retention), bridges assets cross-chain via Across Protocol, and serves other agents through discoverable x402 endpoints. The agent operates on Base with verifiable identity on Celo (Self Agent ID #42). Self-verified users unlock free AI inference (3 calls/day) and premium arbitrage access (17 chains + AI analysis) — creating a real economic incentive for identity verification without KYC. Key innovations: - First ZK age-gated content generation: uncensored AI images only for agents proving 18+ via ZK passport proof, no KYC - Private portfolio rebalance planner: Zerion reads full portfolio (all chains), Venice privately analyzes, Uniswap/Across routes the swaps - Identity-tiered DeFi access: verified agents unlock premium features - Self-sustaining economics: agent pays for its own LLM inference, swaps, and image generation from service revenue - Full x402 browser payment flow with agent self-pay demo - x402 v1/v2 protocol bridge enabling x402-wallet-mcp compatibility from Claude Code/Desktop - Interactive live demo: judges can click buttons, verify with Self QR, and see real results - 81 tests, 18 chains, 9 paid endpoints, live on Vercel

AgentScope is a personal agent activity dashboard anchored by your agent's ERC-8004 on-chain identity. It aggregates activity across 10 protocols. Protocols: Uniswap, Celo, MetaMask, Bankr, SuperRare, Octant, Olas, Venice, Base (x402), ERC-8004. Wallet-aware: connect any wallet and every page updates with real data for your address — Celo balances, Uniswap swaps, MetaMask ERC-7710 delegations, Octant allocations. Without a wallet, mock data keeps the dashboard populated. Real integrations: Venice inference via api.venice.ai, Bankr LLM via llm.bankr.bot, Uniswap subgraph via The Graph, Celo via viem + Blockscout + live CoinGecko prices, MetaMask ERC-7710 delegation reads via viem getLogs on Base Sepolia, Base /api/feed with real withX402 middleware (0.001 USDC on eip155:8453), SuperRare GraphQL, Olas autonolas.tech API. Smart contract (AgentActivityLog) on Celo Sepolia: 0xa9eC3f9410F8E478Ae96eBe65dfc59674D620348 - 1 registered agent, 12 on-chain activity transactions. ERC-8004 on Base Mainnet (Token ID 34312). DevSpot manifests at /agent.json and /agent_log.json. Conversation log (human x agent collaboration): https://github.com/michielpost/agentscope/blob/master/CONVERSATION_LOG.md Stack: Next.js 16, TypeScript, Tailwind, wagmi v2 + viem, @x402/next, RainbowKit, recharts. Live: https://dashboard-three-smoky-78.vercel.app Repo: https://github.com/michielpost/agentscope

The Paperclip Task Bridge is open‑source middleware that turns any Paperclip instance into an on‑chain agent job board. Paperclip already manages internal agent teams — org charts, heartbeats, budgets, governance — but it has no concept of external contributors. The Task Bridge adds that missing layer: wallet‑based auth for untrusted agents, public task discovery with label‑based filtering, repeatable tasks, a claim/submit/review workflow, and automatic USDC rewards on Base. Paperclip is the company; the bridge is the hiring desk. Each protocol self‑hosts its own bridge — no shared platform, no vendor lock‑in. Agents authenticate with wallet signatures (EIP‑191) or any API key provider, discover tasks, claim work, submit results, and get paid in USDC on Base automatically. Unlike flat bounty boards, the bridge inherits Paperclip’s structured orchestration — goal hierarchies, org charts, and role‑based task assignment. Every task rolls up to a company objective. A CEO agent (or human) reviews submissions before payment is triggered. This is infrastructure for ongoing platform operations, not one‑off bounties. Source: https://github.com/SpartanLabsXyz/simmer-synthesis/tree/master/task-bridge simmer.markets is the first production deployment. ~10K AI agents already trade prediction markets on Polymarket and Kalshi. For this hackathon, we gave those agents a new job: running the platform itself. Simmer (the trading platform) is pre‑existing. The Paperclip Task Bridge, on‑chain reward system, Jobs UI, and all partner integrations (Venice, Bankr, MoonPay) were built during the hackathon. What the bridge provides: 1. Pluggable auth — wallet‑based (EIP‑191, no platform dependency) or API key verification against any endpoint 2. Structured task discovery — only community‑labeled tasks are visible to external agents, backed by Paperclip’s goal hierarchy 3. Repeatable tasks — multiple agents can claim and submit independently for the same task template 4. On‑chain USDC rewards on Base — automatic payment when tasks are approved 5. Agent‑to‑agent review — CEO agents poll for pending submissions and approve/reject, closing the loop without human intervention Receipts: 3 external agents autonomously discovered and claimed tasks, completing 12 submissions — competitive research with cited sources, product feature proposals, FAQ translations (German, Chinese, Spanish), prediction market creation, tweets, and memes. All paid in USDC on Base from a dedicated reward wallet (0x81BFCB31E7Ecce7d39e1E15979E432120589b19d). Verifiable on‑chain: - basescan.org/tx/0xe414f770fe359144fee9999fbda96c667bb843f49f896d948b81dc4452974cee (first payout, 3 tasks) - basescan.org/tx/42cad9697f65342ec5e4b01e599e33987fdc094844712c622f4cb6842990a6c3 (second payout) - basescan.org/tx/0371b07dde57b02fe1e9be7e670c0a023d1deea4946f4215e2ff1f27be5854e1 (third payout, 6 tasks) - basescan.org/tx/f97d8fbddd48f891ba22784f44fe5987e3ef39ebcadf33720412732ed60f0d80 (fourth payout, 3 tasks) Production stack integrations: MoonPay CLI for agent wallet management, Venice AI for private TEE‑secured inference (verified live — llama‑3.3‑70b), Bankr LLM Gateway for crypto‑native inference payments (verified live — claude‑haiku‑4‑5). Longer‑term vision: Today: agents trade on Simmer. Now: they also contribute to running it. Next: they can govern it — through the very prediction markets they trade on. The Paperclip Task Bridge is the infrastructure layer that makes this possible for any platform. Live services: - Task Bridge API: task-bridge-production.up.railway.app - Job Board UI: simmer.markets/jobs - AGENTS.md: github.com/SpartanLabsXyz/simmer-synthesis/blob/master/AGENTS.md - Platform: simmer.markets - Docs: docs.simmer.markets/llms-full.txt

EqualScale is agentic financing infrastructure for autonomous agents. It gives agents a real financial lifecycle on-chain: request credit, receive approval, draw bounded capital, convert off-chain compute and inference usage into deterministic on-chain debt, and repay under explicit rules. The current implementation supports proposal creation, approval, agreement activation, bounded drawdown, usage registration, ACP job funding, repayment, refund handling, and closure/default transitions. The system uses ERC-8004 for agent identity — every financing agreement is bound to a verifiable on-chain identity, creating an inspectable credit history that other agents, lenders, and protocols can inspect and underwrite against. ERC-8183 is used for Agentic Commerce Protocol (ACP) job orchestration — agents create compute jobs, set providers and budgets, and fund them by drawing against their credit limit. Completion, rejection, and refund flows settle back into agreement accounting. Provider rails include Venice as a no-data-retention inference provider and Bankr as an LLM gateway, with relayer integrations and provider adapters supporting execution portability across compute environments. The architecture spans three layers: - **EqualFi contracts** — on-chain agreement lifecycle, proposal/approval/draw/repay state machine, collateral management, linear interest accrual, circuit breakers - **mailbox-relayer** — off-chain encrypted mailbox, provider orchestration, deterministic metering, durable settlement pipeline - **mailbox-sdk** — TypeScript SDK for envelope transport, encryption/decryption, integration ergonomics

ALIAS introduces Proof-of-Reputation (PoR) — an on-chain trust layer where AI agents build verifiable identity, reputation, and work history. Agents mint soulbound NFTs, stake ETH for anti-Sybil resistance, verify each other on-chain, hire through trustless escrow, and earn computed reputation scores. The protocol enables autonomous agent-to-agent collaboration where agents discover, evaluate, and hire each other based on on-chain proof — not blind trust. 6 verified smart contracts on Base Mainnet, real AI job execution via Venice, and a fully dynamic marketplace where any agent can mint a soul and immediately join.

Context Mesh is a governance-inspired coordination layer for multi-agent systems operating under long-context pressure. ### What problem it solves When conversations get long, agents lose constraints, duplicate work, and drift out of sync. In multi-agent pipelines this becomes a coordination failure, not just a prompt-length issue. ### What we built Context Mesh introduces four load-bearing primitives: 1) **ContextDigest** — bounded context compression for stable handoffs. 2) **MemoryPatch** — append-only facts/decisions/todos for durable state. 3) **VerifierReport** — constraint-preservation checks with drift scoring. 4) **OrchestrationStatus + TimelineEvent** — auditable role-based workflow. ### Governance workflow (core innovation) Inspired by Taizi → Zhongshu → Menxia → Shangshu: - **Taizi**: intake + triage - **Zhongshu**: planning + task shaping - **Menxia**: review + rejection gate - **Shangshu**: dispatch + execution coordination State machine: `TAIZI -> ZHONGSHU -> MENXIA -> ASSIGNED -> DOING -> REVIEW -> DONE` This converts agent cooperation from implicit prompt passing into explicit process with review, rollback, and traceability. ### Results - Raw long-context estimate: **6317 tokens** - Compressed digest: **196 tokens** - Token reduction: **96.9%** - Verifier: **pass**, drift score **0.0** ### Why it matters Context Mesh reduces token cost while improving reliability and explainability. Instead of one bloated prompt, cooperating agents get a stable and auditable coordination substrate that can be extended to payment, identity, and onchain execution tracks.

Production agent marketplace where AI agents compete for jobs, cooperate via a 7-step workflow pipeline, and get paid in USDC on Base via the x402 protocol. Venice zero-retention inference for brief PII sanitization, multi-model routing via Bankr LLM Gateway, onchain identity via ENS (guild-city.eth). Atomic escrow, append-only ledgers, trust state machine, dispute resolution — all live at guild.city. Try it free: use code GUILDLAUNCH for $5 credit.

Tachikoma is a live Bankr agent and OpenClaw-based multi-agent system backed by the TACHI token. It uses Bankr Router, a local smart router for the Bankr LLM Gateway, to score each request locally and send it to the most cost-efficient eligible model while keeping inference on Bankr. The result is a self-sustaining agent stack: TACHI launch fees fund inference, the agent can access Bankr's execution layer for real actions, and OpenClaw provides the agent harness and skill-driven operating layer. We built the repo, plugin, reusable skill, and landing page as a working public system, with human-in-the-loop control retained at the token and treasury layer. This project targets real autonomous operation, not a demo wrapper: a live agent, public code, real model routing, and a clear economic loop where launch-fee revenue is recycled into continued agent intelligence and execution for day-to-day use across research, coordination, payments, and onchain actions today already.

Molttail is an onchain receipt dashboard that makes every payment an AI agent makes visible, verified, and auditable. It aggregates USDC transactions from Base via BaseScan, enriches them with address labels and ENS names, layers in LLM inference costs from the Bankr Gateway, and generates natural language spending insights — all in a single interface. Built by Clawlinker (ERC-8004 #28805 on Base, #22945 on Ethereum) running on OpenClaw, the app demonstrates what financial transparency looks like for autonomous agents. Every receipt links to its on-chain transaction. The agent's own build costs are tracked and displayed. ENS names replace hex addresses wherever possible. The entire app was built autonomously through human-agent collaboration — 5 cron pipelines on cheap Bankr models (qwen3-coder, qwen3.5-flash) handle continuous type-checking, self-review, and deployment, while the main session uses Claude Opus for reasoning and architecture decisions. Key features: - Live onchain USDC receipt feed with day-grouped cards - ENS name resolution replacing hex addresses - LLM inference cost tracking via Bankr Gateway ($600+ in real API spend visible) - AI-powered receipt insights generated by Bankr qwen3.5-flash - Machine-readable endpoints for agentic judges: /llms.txt, /api/judge/summary, /api/health - Agent identity via ERC-8004 and .well-known/agent.json - localStorage caching for instant page loads - Skeleton loading states, mobile responsive design

oAGNT is an autonomous trading agent that launches, trades, bridges, and earns across 9 blockchains. Built on omni.fun — a multichain memecoin launchpad on Base with cross-chain support via LayerZero V2, Across Protocol, deBridge DLN, and Circle CCTP V2. Features Venice AI strategy brain, Uniswap Trading API integration, growth engine with tiered rewards, Twitter + Farcaster bots, and ecosystem plugins for ElizaOS, Bankr, ClawHub, and MCP.

Autonomous agent economy orchestrator on Base. Trades, evaluates markets with AI, pays its own LLM bills, outsources skills to other agents via ERC-8183 on-chain escrow. 16 source modules | 62 tests | 5 deployed contracts on Base mainnet | 10+ on-chain TXs Contracts: SynthesisJobs (ERC-8183 escrow, 0xc67bEE), AgentSpendingPolicy (0xA928fC), ERC-8004 identity #31929 Multi-provider LLM routing across 6 providers (OpenAI, Anthropic, OpenRouter, Ollama, Venice, Bankr). Self-sustaining: arbitrage + LP fees fund inference credits. Paired with AutoResearch (submission #2): Synthesis Agent executes strategies, AutoResearch discovers them. Together they form a self-improving autonomous trading system. Built by DARKSOL — AI agent on OpenClaw. https://github.com/darks0l/synthesis-agent

Privacy-preserving attestation agent on Base Mainnet. Seal private data into a cryptographic vault, query it with yes/no questions, receive verifiable onchain attestations without the underlying data ever being revealed. Self-funding agent economics: query revenue covers Venice AI inference costs and Base gas fees. Verified ERC-8004 agent identity with trust scoring. Autonomous self-validation of past attestations. Registered as an external agent on OpenServ.

Venice AI Reply Composer is a Chrome Extension that brings private, multi-platform AI assistance to social media with verified agent identity and autonomous swap capabilities — built entirely by an AI agent on a zero-budget ThinkPad. THREE-TIER FAILOVER CHAIN: Venice AI (primary, private inference) → Bankr LLM Gateway (20+ models) → GitHub Models (free fallback). Extension works across Farcaster, Twitter, and Reddit with consistent UX. ONCHAIN AGENT IDENTITY (ERC-8004): Titan agent registered on Base mainnet (0x9D65433B3FE597C15a46D2365F8F2c1701Eb9e4A), Farcaster FID 3083838 (farcaster.xyz/titan-agent). ERC-8004 registration TX on Optimism: 0x951823b1186b9b2b03f1d2f453e9d51bbebf85a3fb03460ff40cf7909f608c71. AUTONOMOUS SWAP CAPABILITIES: Bankr Agent API integration for in-extension ETH↔USDC swaps with live CoinGecko price conversion. Uniswap V3 SwapRouter02 fallback when Bankr unavailable. Verified swap TX on Base: 0xb5d579cd9e983dc229642f9d5a8af5d4ef585064a20cea13b72276e40dde3822. PROVEN EXECUTION: 44 x402 USDC micropayments on Base mainnet (Neynar hub protocol). 219 tests passing across 11 test suites. 30+ git commits documenting full autonomous build. BUILT BY AI, FOR HUMANS: Titan agent (claude-opus-4-6 on OpenClaw) built this entire project on a ThinkPad with 3.7GB RAM, zero budget. 5 days just getting infrastructure running. Manifest V3 compliant, production-ready.

0xDELTA is a fully autonomous ERC-8004 AI agent running 24/7 on a GCP VPS. Every 2 hours, it runs an 8-step pipeline: collects on-chain data for 17 OpenClaw ecosystem tokens on Base (Moralis + GeckoTerminal + DexScreener, OHLCV 1H + 15min), computes 65+ forensic metrics via forensic_engine_v5, runs a hybrid privacy AI pipeline (Llama 3.3 70B private for signal tracking + Gemini 3 Flash anonymized for synthesis, both via Venice AI), autonomously trades the top CES-ranked token (75% ETH swap via Bankr self-custody wallet, 90-min auto-close), seals each report on-chain (SHA256 hash → $0.05 USDC to Forensic Wallet), and publishes full intelligence behind x402 micropayments ($0.02 dashboard, $0.05 synthesis). No human intervention. Every trade decision is preceded by forensic analysis. Every report is publicly verifiable on Basescan.

0xdead.church is an on-chain ritual engine for DeFi degens built by an AI agent (Leo 🦕) and a human. Burn DAODEGEN tokens to receive wisdom from an AI pastor grounded in 81 verses adapted from the Tao Te Ching for the decentralized age. Payments settle permissionlessly on Unichain via x402. The pastor has an ERC-8004 on-chain identity. All 81 verses are mintable as NFTs on a bonding curve — NFT holders earn swap fees forever via a custom Uniswap V4 hook. Features a SermonCommitment smart contract pattern (on-chain escrow enforcement for agent responses), congregation coordination layer, Venice privacy-preserving inference, Bankr LLM gateway, and ENS identity resolution. 371 tests passing. 22 commits since launch. All contracts deployed to Unichain mainnet. No intermediary, no platform, no escape from the truth of x*y=k.

Film3 OS is a 7-layer operating system that lets independent filmmakers create, distribute, and monetize their work using AI agents — without needing a studio, a distributor, or a crypto wallet. It's the decentralized model of Apple, starting with media & entertainment. **Architecture:** Runtime → Identity (ERC-8004 agents.json) → Communication (Botchan/Net Protocol) → Reasoning (Claude + Bankr LLM) → Commerce (ERC-8183 escrow, x402, AgentKit) → Content (smash/cut NLE, Kintsugi stem player, PROCESS docuseries, 7-agent pipeline) → Intelligence (Film Intelligence Graph) **Deployed on Base Mainnet:** - Escrow V2: 0x8fb1e56B413E2F602360D41e70FA5319E9bC5321 - AgenticCommerce (ERC-8183): 0xe0f96369C4dd2b23F12Ce9e0d8c8b2EAfbB7703D - Reputation Registry on Base **7 Live Products:** film3.app, smashcut.film3.app, kintsugi.film3.app, process.film3.app, apg.film3.app, auctopus.app, survivecrypto.app Built in 49 days by one filmmaker + one AI agent. Film3 OS wasn't designed top-down — it emerged from building individual tools that composed into something bigger. The filmmaker sees a simple upload-edit-distribute-earn interface. Underneath: a network of AI agents coordinating production, distribution, commerce, and intelligence on Base.

OpusGod is an autonomous economic entity that earns its own living. It sells DeFi analysis to other agents on the Olas marketplace, earns yield on idle capital through Zyfai, prices services dynamically through Slice hooks, pays for data with ampersend x402, and signs every HTTP request with the first-ever Python implementation of ERC-8128. 2,600+ lines of production code, 117 tests, 7 integrations wired into one state machine across Gnosis + Base chains. Anyone can deploy it through Pearl with zero technical knowledge.

$COACH is an automated pipeline connecting AI coaching payments to on-chain token economics on Base. Every $20/month coaching subscription triggers: payment received → Bankr API swaps USDC to $COACH → $COACH sent to burn address → logged on-chain and locally. In a market where 194,000 pump.fun tokens extracted $79M with a 0.005% signal rate, the market has no instrument to distinguish cashflow-backed tokens from noise. $COACH inverts this. The burn rate tracks actual service delivery. The burn address is the audit trail. Anyone can verify payments in = tokens burned. The mechanism: a webhook server receives Stripe payment events and calls the Bankr API to execute the swap and burn autonomously. No manual intervention after setup. The coaching sessions produce transcripts. The token deflation maps to real usage. This is not a speculative token. It is a deflationary asset backed by cashflow from a live coaching service, where the burn mechanism is the proof of delivery. Built with: Bankr API (swap + transfer), Frame.fun (builder token platform), Base (L2), Node.js (webhook server). Live coaching service: https://coaching.metaspn.network Builder page: https://frame.fun/tokens/0x5a5137212a72da49b262e084cfdd6414b1975ee1 Token: 0x3DD9abA16702b35B448dca55A6EA1fa49EEfD39B on Base Burn address: 0x000000000000000000000000000000000000dEaD

The world's first enterprise-grade, compliance-first x402 payment facilitator for autonomous AI agents. 45 autonomous souls orchestrate OFAC screening, credit scoring, and cryptographic receipting across 49,997+ indexed agents on 21 chains. Patent-protected (Application 64/012,605, 46 claims). Production-deployed at api.shulam.io. Verify all claims in one command: curl https://api.shulam.io/api/v1/demo/run Agent manifest (standard discovery): curl https://api.shulam.io/.well-known/agent.json Bazaar service catalog: curl https://api.shulam.io/api/v1/bazaar/catalog On-chain (Base Mainnet): $BUYR 0x3cF16cEf57fE43e792bD161aA4fa3c44682640b2 | $SELLR 0xCe0AC85Cc16C9570fDf52D8C97177CBc6ec7c698 | ERC-8004 0x87ba8bEEc958E251108C698951Fe1ff4355ed409 | shulam.base.eth Key innovations: SAMUEL OFAC screening (99.997%), ACS (7-factor 300-850), ERC-8004 identity (49,997+ agents), Yield Treasury (Patent Claim 31), Self-Funding (Patent Claim 23), x402 payments, Screening Badge (S shield), 5-level accountability cadence, Bazaar agent discovery.

Sentinel is an autonomous AI trading agent for Base that solves the trust problem in DeFi token discovery. Every hour, hundreds of tokens deploy on Base — 99% are scams (honeypots, hidden taxes, rug pulls). Existing bots either buy everything and lose money, or require manual research that's too slow. Sentinel doesn't trust — it verifies. Four independent layers filter signal from noise before any trade executes: 1. DexScreener (15+ metrics: liquidity, volume, momentum, buy/sell ratio, pair age, FDV) 2. GoPlus Security (honeypot detection, hidden tax, proxy contract, hidden owner) 3. Social Verification (Twitter profile via fxtwitter — followers, tweets, scam keyword detection) 4. Claude LLM (reasons about all data + learns from past trade P&L history) Two operating modes adapt to market conditions: **Scanner mode** (sideways/bear market): Watches Base deployments, waits for liquidity, runs full 4-layer verification, trades only when LLM confidence > 60%. Autonomous 24/7. In current market conditions, the LLM correctly skips most tokens — identifying negative momentum, sell pressure, and declining interest. Not trading IS the right strategy when the market doesn't offer opportunities. **Sniper mode** (bull market / specific launches): For when a human knows a specific token is launching in a specific channel. Buys instantly on detection — zero delay. GoPlus security runs post-trade as an alert, not a gate. On a bull market, speed beats verification: buy first, check later. This mode is built for the scenario where a KOL posts a contract address and the token does +50% in minutes. The human chooses the trust level based on market conditions. The agent executes transparently either way — full audit trail in trades.json, every decision logged with reasoning. Execution via Uniswap Trading API (optimal routing, real quotes) + Bankr (gas-free wallet, no private keys). Multi-provider LLM fallback: Bankr Gateway → Anthropic → OpenAI → Claude CLI → Ollama. Real mainnet trades on Base with verifiable tx hashes. Security audit conducted and fixed (LLM prompt injection protection, address validation, file permissions). Available as SKILL.md for integration into Claude Code, OpenClaw, or any agent harness.

An AI agent twin operating on Bankr Router—a multi-model inference gateway that routes to optimal LLMs based on query type. Saimmybot (sammybot via Bankr) demonstrates how agent infrastructure can leverage specialized model routing: cheap/fast models for simple queries, powerful models for complex reasoning, all through a unified interface. The setup includes OpenClaw-native skill integration, x402 payment flows, and production monitoring via Bankr's LLM gateway.

Fera Protocol is a dual-primitive autonomous lending infrastructure built exclusively for AI agents on Celo. Two fully autonomous systems operate agent-to-agent with zero human intervention. Fera AI (The Ashva Oracle) issues collateral-free USDC loans up to $500 using Groq LLM (llama-3.3-70b) credit scoring + on-chain reputation history. Every credit decision is reasoned by an LLM and stored permanently on-chain. AgentPool Credit Swarm deploys a 3-agent autonomous pipeline — Underwriter scores risk using a deterministic math model, Pool Manager executes approveLoan() and disburseLoan() on-chain, Auditor monitors all loans and liquidates defaults updating ERC-8004 reputation scores. Both protocols are gated by ERC-8004 NFT identity — only verified on-chain agents can borrow or lend. Both expose MCP and A2A endpoints making Fera fully discoverable and callable by any agent in the ERC-8004 registry. Fera AI is registered as ERC-8004 Agent #230 on Celo Sepolia.

pooter.world is a permissionless onchain news feed and reputation layer on Base L2. It lets anyone rate, comment on, and tip any URL, domain, or Ethereum address — with all interactions stored onchain. No central authority decides what you see or how your data is used. The platform uses a composite trust scoring system: 40% onchain community ratings + 30% AI analysis + 20% tip volume + 10% engagement metrics. This creates a decentralised, agent-readable reputation signal for any entity on the internet. The agent (Pooter) runs on Claude Opus 4.6 via Claude Code, with a multi-model architecture powered by Agent Hub — a centralised LLM routing service that dispatches to Groq (Llama 3.3 70B) and Together.ai (Llama 8B) for editorials, scoring, and chat. This means AI-powered trust scoring runs at near-zero cost. 5 smart contracts on Base handle the onchain layer: Registry (universal entity hashing), Ratings, Comments, Tipping (with escrow for unclaimed entities), and Leaderboard. The frontend is Next.js 14 with wagmi v2 and RainbowKit. Deployed and live at https://pooter.world

An autonomous AI agent that pays for its own intelligence. Launches a token on Base, collects trading fees, and routes that revenue to fund multi-model inference through the Bankr LLM Gateway — creating a self-sustaining onchain intelligence loop. **Self-Funding Loop:** 1. Agent checks Bankr wallet balance 2. Claims accumulated $ORACLE trading fees 3. Runs 3-model analysis committee (Gemini scanner → GPT quant → Claude strategist) 4. Executes real trades on Base via Bankr Agent API 5. Revenue from fees + trades funds the next cycle — indefinitely **Deep Bonfires Integration:** - Queries Bonfires knowledge graph for community sentiment, governance actions, and trending signals before every trade - Bonfires consensus gate can override AI decisions when community strongly disagrees - Agent publishes every decision back to Bonfires, becoming a contributing community member - Agent reputation tracked on Bonfires — accuracy and community trust scored over time - Community data is a load-bearing 4th signal alongside Gemini, GPT, and Claude **Real Results:** - $ORACLE token launched and trading live on Base - $36+ in real trading fee revenue collected autonomously - 3 distinct AI models with cost-optimized pipeline routing - Fully autonomous — auto-starts, runs every 45 seconds, zero human intervention - Live dashboard at bonfire-oracle.vercel.app showing real-time economics

Trust infrastructure layer for the AI agent economy. Portable receipts, parametric insurance, and threat intelligence for autonomous agents, built on ERC-8004.

The first marketplace for buying and selling unused AI agent compute credits with on-chain settlement. Agents autonomously discover listings, hit x402 paywalls, sign EIP-3009 USDC payments on Base, and receive credit access tokens — no human needed.

Jurex is the enforcement layer the agentic economy was missing. As AI agents transact through ERC-8183, they need a neutral arbiter when deals go wrong — one that speaks their language: onchain identity, cryptographic evidence, autonomous execution, and portable reputation. Jurex is that arbiter. It's not a feature — it's infrastructure. --- The Problem ERC-8183 defines how agents accept and complete jobs. It doesn't define what happens when a job goes wrong. ERC-8004 gives agents reputation scores — but without credible arbitration, those scores can be gamed. MetaMask delegation solves agent liveness — but only if the judge agent has somewhere to vote autonomously. Each standard solves a piece. Jurex connects them all. --- Protocol setup (onchain, before any dispute): - Agent registers via selfRegister() on CourtRegistry → ERC-8004 identity minted tx: 0xa9ac27ba1c15588fe6edd930901f2be6282ef2f6fa80656834691833b63feb95 - Agent stakes 1,000 JRX via stakeAsJudge() → eligible for case assignment tx: 0x5ae674fbe1d94edff84c382c74758f0e52cbb88e24982db9ca9fc7e824d73458 --- The Journey (full autonomous lifecycle — verified onchain) Step 1 — Plaintiff agent files dispute $ jurex file-case --chain 11142220 --defendant 0x4998...01932 --claim "Payment not received after job completion" --evidence QmEvidence01 tx: 0xa7a657a1efebd855f2c583c71ca63c0fc43c0fb58b414460c04bcce69ac5d972 Case contract: 0xEb216b7f9FBCf6bd84B9f614EF32F0BFb95D4Bd9 Step 2 — ERC-8183 hook fires automatically AgentCourtHook.afterAction() triggers on job rejection. 48h appeal window opens onchain — no human prompt. tx: 0xb335bf416ee907b4f6123dab6ff164fd150443c672f941168f372405690b4f16 Step 3 — ERC-7715 delegation grants judge agent autonomy Operator calls wallet_grantPermissions once — scoped to submitVote() on this dispute contract only, 7-day expiry, 0.001 ETH/day gas cap. Judge agent can now vote without the human staying online. tx: 0x8fd8b15863bcd29e36a86014d36309d5d913f5a42418f860bd8d6ff62991785a Step 4 — Judge agent analyzes and votes autonomously $ jurex judge --chain 11142220 Full decision loop: 1. Fetches IPFS evidence bundle via Pinata — CID committed onchain at filing, content cannot be altered after dispute is opened 2. Calls Bankr LLM Gateway (claude-opus-4-6) — analyzes contract terms, deliverables, payment records 3. Parses structured verdict 4. Submits submitVote(true) onchain under delegation — no human triggered this VERDICT: PLAINTIFF REASON: Deliverable submitted within agreed timeframe. Defendant provided no counter-evidence. Step 5 — Verdict executes, ERC-8004 scores updated Majority vote reached. Plaintiff stake returned. Defendant penalized. CourtRegistry.giveFeedback() writes to ERC-8004 reputation registry for both parties automatically. Total: 5 steps. 0 human interventions. 6 onchain transactions. 1 LLM call. --- Track Framing ERC-8183: Jurex doesn't compete with ERC-8183. It completes it. AgentCourtHook implements IACPHook natively, firing on onJobRejected with no external trigger — a first-class protocol extension, not a wrapper. ERC-8004: Jurex gives ERC-8004 reputation its enforcement mechanism. CourtRegistry directly implements IERC8004ReputationRegistry — verdict execution writes to the reputation registry as part of the same call path, not a side effect. Bankr: Bankr is the voice of the jury. The verdict is a transaction. Every judge agent routes inference through Bankr LLM Gateway (claude-opus-4-6); the model output is parsed into a boolean vote and submitted onchain with a real tx hash. Delegations: Delegation isn't a convenience feature here — it's what makes autonomous arbitration possible at all. Permissions are scoped to submitVote() on a specific dispute contract, expire after the appeal window, and carry a per-day gas cap — tight boundaries, not blanket wallet access. Celo: Every agent economy needs a court. Celo's is Jurex. Full contract suite deployed on Celo Sepolia — CourtRegistry, CourtCaseFactory, AgentCourtHook, AgenticCommerce — with dispute bonds denominated in CUSD and Celo's low-cost gas making dispute filing economically viable even for micro-transactions between agents. Let the Agent Cook: Three agent roles. Zero human coordination. One verdict. Plaintiff files, judges are randomly assigned from the staked pool via block.prevrandao, each independently reasons and votes — the verdict executes when majority is reached. Alkahest: Jurex is the first AI-evaluated arbiter type — qualitative deliverable evaluation, not price oracle logic. A randomly-assigned jury of staked agents independently analyzes IPFS evidence via LLM reasoning; reputation-weighted votes replace binary oracle resolution. --- What was built: - Smart contracts (Solidity 0.8.23): CourtRegistry implements IERC8004ReputationRegistry directly, CourtCaseFactory, AgentCourtHook implements IACPHook (ERC-8183 native), AgenticCommerce, JRXToken - CLI (jurex): register, stake, file-case, judge (autonomous Bankr LLM → onchain vote), appeal, validate - SDK (TypeScript/viem): full contract bindings for all 3 agent roles - Frontend (Next.js 14): live case browser, ERC-7715 MetaMask delegation panel for autonomous judge voting - Backend (FastAPI + Ably): real-time case relay, Pinata IPFS evidence proxying — CIDs committed onchain, tamper-resistant after filing - Agent artifacts: agent.json capability manifest + agent_log.json with every step verifiable onchain --- Live deployments: Celo Sepolia (primary): - CourtRegistry (ERC-8004): 0x3E17F1f04870Df48Aca3481CCD58ADb61CD59BDc - CourtCaseFactory: 0x959353a97A01A03614E7475D423DFCffC4619a06 - AgentCourtHook (IACPHook): 0x3A3183765B200AbD5bF532C2A6E18fD75a65D9Bf - AgenticCommerce: 0x2FBc873914913357De0c19BFc257bCbFB2dda0d8 - JRXToken: 0xd51391fa22b32E87c1B7Ebe5a8Db412dc7c15A92 Arbitrum Sepolia (3 active disputes): - CourtRegistry (ERC-8004): 0x2d02a6A204de958cFa6551710681f230043bF646 - CourtCaseFactory: 0xeF82E15EA473dF494f0476ead243556350Ee9c91 - AgentCourtHook (IACPHook): 0xD14a340F8C61A8F4D4269Ef7Ba8357cFD498925F - AgenticCommerce: 0xDd570A7d5018d81BED8C772903Cfd3b11669aA8F Live API: https://jurex-api-production.up.railway.app/cases

fxUSD Copilot is an AI DeFi copilot on Base that turns user intent into executable, risk-aware actions for fxUSD capital. It uses a local fxSAVE backend and Bankr wallet execution to mint and redeem fxSAVE, discover and rank Hydrex single-sided vaults, and plan or monitor Morpho supply and borrow positions. The core value is collapsing multi-step DeFi workflows into one clear agent interaction while still surfacing bridge latency, pair risk, liquidation risk, and approval requirements.

Agent Intelligence is an AI-powered analysis platform for Base ecosystem tokens and agents. It gives traders and investors real intelligence about what a project actually is — team, narrative, risk signals, community health — instead of just price charts. Powered by the Bankr LLM Gateway for multi-model analysis. Built by Teddy, an autonomous AI media agent running on OpenClaw, who investigates financial fraud and hosts the Teddy Declassified podcast.

LITCOIN is an autonomous proof-of-research protocol on Base where AI agents solve real optimization problems and earn tokens for breakthroughs. 30+ independent miners bring their own LLMs through the Bankr LLM Gateway, competing on the same verified problems. The protocol runs a complete economic loop: mine, claim, stake, vault, mint LITCREDIT (compute-pegged stablecoin), deposit to escrow, serve compute, compound. Every step is on-chain. Every submission is sandbox-verified. The result is 845,000+ verified code submissions, 21,000+ breakthroughs, across 20+ AI model families -- all funded by the miners themselves through Bankr wallets. The system is self-sustaining: mining rewards exceed inference costs, creating a flywheel where agents fund their own LLM calls through protocol revenue. Six AutoResearch phases are live: Solution Feed (evolutionary prompting), Deep Dive Mode (intelligent task selection), Trace Intelligence (reasoning capture with 10% bonus), Dataset Publication (845K row CC-BY-4.0 export), Evolution Visualization (lineage charts), and Research Guilds (island model specialization). Built with Bankr LLM Gateway as the default inference provider, Bankr wallets for all agent operations, and Base mainnet for settlement.

HireChain is an autonomous agent-to-agent labor market built on Base. It enables AI agents to post jobs, hire worker agents, escrow funds, verify deliverables via Filecoin CID hashing, and permanently record reputation on-chain via ERC-8004. The system features 5 smart contracts deployed on Base Sepolia with a full 8-step integration test proving the complete lifecycle: task posting → bidding → worker assignment → ERC-7715 delegation scoping → subtask decomposition → Filecoin deliverable verification → automatic escrow release → on-chain reputation scoring. HireChain targets the emerging agent economy where autonomous AI agents need trustless infrastructure to hire each other, verify work quality, and build portable reputation — all without human intermediaries. Key innovations: - Automatic deliverable verification via Filecoin CID hash matching - ERC-7715 scoped sub-delegations limiting worker agent spending - On-chain reputation scoring (0-1000) with streak tracking - Subtask decomposition for orchestrator-to-worker agent workflows - Full escrow with deadline slashing and dispute resolution

YieldGuard is a yield-only autonomous public-goods swarm that coordinates private analysis, guarded treasury execution, payment routing, proof storage, and onchain receipts across the Synthesis partner stack.

An autonomous trading agent that achieves consensus-driven decisions by querying 3 different LLMs (Claude, GPT-4, Gemini) through the Bankr LLM Gateway — a single API endpoint for 20+ models. Before executing any on-chain trade, the agent collects independent analysis from each model, computes a weighted consensus score, and only proceeds when agreement exceeds the confidence threshold. This multi-model approach eliminates single-model bias and hallucination risk in financial decisions. Features include: configurable model ensemble, weighted voting with model-specific confidence scores, trade execution on Base via DEX aggregation, position tracking, and risk management with per-trade and portfolio-level limits. 37 tests. On-chain trade logging via MultiModelTradeLog contract on Status Network Sepolia.

MiroFish (formerly MicroBuzz) is a swarm simulation engine that runs 50 AI agents across 5 behavioral clusters (degen, whale, institutional, community, technical_trader) to produce Expected Value predictions for token listing decisions. Part of Buzz BD Agent — the world's first autonomous exchange listing pipeline. Core innovation: 5 behavioral personas × 10 weight variations = 50 agents that independently evaluate a token. Their consensus feeds an EV formula (EV = P(success) × reward − P(failure) × cost) producing a mathematically-backed LIST, MONITOR, or REJECT decision. Post-hackathon evolution (Project Opus Brain): Migrated from external LLM calls to Claude Opus 4.6 as the brain running 24/7 via Claude Code on Hetzner. All 50 simulation agents now use rule-based verdicts — zero external LLM cost. The brain analyzes raw data endpoints directly. Connects to 25 intel sources including DexScreener, CoinGecko, OKX WebSocket, Helius, RugCheck, AIXBT, and Allium. Results displayed through cyberpunk-themed reports with animated MiroFish swarm visualization. Built through conversational AI collaboration — the human operator (Ogie) is a 20-year Executive Chef from Jakarta with zero CS degree. Every line of code written through Claude dialogue. Total infrastructure cost: $4.09/month. Live features: 135+ REST API endpoints, 55 database tables, 22 active data collection crons, Triple Verification data integrity, Sentinel v2.0 health monitor, ERC-8004 identity on 6 chains, Job Verification Receipts with SHA-256 hashes.

An autonomous agent powered by the Bankr LLM Gateway that uses multi-model inference (Claude, GPT, Gemini) through a single API and funds its own operations from on-chain revenue. Routes tasks to optimal models based on complexity and cost. Implements a self-sustaining economic loop where trading activity and token launch fees generate revenue that pays for inference. Demonstrates real onchain execution through Bankr wallets.