Octant Eval Agent

Octant Eval Agent's Team

claude-sonnet-4-6cursorCustom Python agent with structured retrieval and direct Anthropic Messages API calls. No agent framework used.StreamlitFastAPIAnthropic Messages APIOctant Mainnet APIuvicorn+1

Problem Statement

Octant has distributed over $12M in public goods funding across 11 epochs, but evaluating projects and understanding funding patterns requires manually navigating scattered API endpoints and computing cross-epoch comparisons by hand. No single tool lets you ask a question like "which projects had the most consistent donor base across epochs" and get a grounded, cited answer. This agent solves that for anyone evaluating Octant projects: governance participants deciding where to allocate, researchers studying public goods mechanisms, or the Octant team itself tracking protocol health.

An AI agent that ingests the complete history of the Octant public goods funding protocol and makes it queryable through natural language. It programmatically collects all finalized epoch data from the Octant mainnet API (staking proceeds, matched rewards, donor/patron lists, allocations, leverage ratios, project metadata) into a single structured dataset, then uses structured retrieval and an LLM to answer questions with explicit epoch and project citations. The agent builds targeted context slices per question rather than dumping raw data into the prompt, keeping answers precise and traceable. If the data does not contain the answer, it says so. Includes a Streamlit chat UI with dataset overview metrics, a FastAPI HTTP layer for programmatic access, and an evaluation harness for measuring answer accuracy.