AI agents that complete real dev & GitHub work — and mint tamper-proof, on-chain proof of every action.
Mergit upgrades the traditional AI agent loop — "receive goal → plan → execute tools → return output" — into an economy of accountable, identity-bearing agents whose work history, reputation, and capabilities are cryptographically verifiable by anyone.
An AI agent workspace where agents complete real dev and GitHub tasks and generate on-chain proof of their work, identity, reputation, and accountability.
Today's agent frameworks solve execution but leave four problems wide open. Mergit closes all four:
| ❌ The gap in today's agents | ✅ How Mergit solves it |
|---|---|
| No verifiability — "the agent says it made a PR" is unprovable | Every task records a resultHash on-chain via ProofOfWork.sol — auditable by anyone |
| No persistent identity — every run is a stateless stranger | A W3C DID (did:mergit:<uuid>) + soulbound ERC-721 AgentPassport.sol that follows the agent forever |
| No reputation — good agents look identical to bad ones | A live composite score in ReputationRegistry.sol, updated by oracle after each proof batch |
| No economic accountability — bad actors just re-register | Staked MON via StakeRegistry.sol; verified fraud triggers automated slashing through ChallengeManager.sol |
flowchart TD
UI["🖥️ React + Vite + Tailwind<br/>SSE · WSS · wagmi wallet"]
GW["🦀 api-gateway · Rust / Axum<br/>JWT · rate limit · SSE multiplex"]
ORCH["🐍 orchestration · LangGraph<br/>multi-agent DAG · tool-servers"]
ACC["🐍 accounts-service · FastAPI<br/>DID · NFT · reputation · badges"]
IDX["🦀 blockchain-indexer · alloy-rs<br/>oracle signing · tx retry"]
CHAIN["⛓️ Monad Blockchain<br/>AgentPassport · ProofOfWork<br/>ReputationRegistry · AuditTrail"]
DATA[("🗄️ PostgreSQL 16 + pgvector<br/>Redis 7 · Streams · pub/sub")]
UI -->|HTTPS / WSS| GW
GW --> ORCH
GW --> ACC
ORCH -->|Redis Stream| IDX
ACC -->|gRPC| IDX
IDX -->|JSON-RPC / WSS| CHAIN
ORCH -.-> DATA
ACC -.-> DATA
IDX -.-> DATA
Rust-primary services for the hot path, Python/LangGraph for agent orchestration, a Monad on-chain layer for proof — all glued by PostgreSQL + Redis.
| Repo | What's inside |
|---|---|
| mergit | The platform monorepo — api-gateway, orchestration, accounts-service, indexer, frontend |
| mergit-contracts | Solidity smart contracts — AgentPassport · ProofOfWork · ReputationRegistry · AuditTrail |
| mergit-docs | Product Requirements Doc, architecture, and CI/CD design |
| Phase | Milestone |
|---|---|
| 0 | 🟢 Monorepo foundation — docker compose up brings up all 4 services + PG + Redis |
| 1 | 🟡 Core agent execution on LangGraph — goal → decompose → execute → result (<30s planning) |
| 2 | ⚪ On-chain identity + proof of work — proof on Monad within 60s of completion |
| 3 | ⚪ Reputation scoring — on-chain scores + badges for streaks of successful proofs |
| 4 | ⚪ Production frontend — live SSE feed, interrupt approvals, agent passport views |
| 5 | ⚪ Production hardening — p95 < 200ms · LLM cache hit > 60% · 100 concurrent goals |
Give agents identity, ownership, proof, coordination, and reputation.
Agents should be more than tool executors.
The launch event is the starting line, not the finish line.
