Tracer

Tracer is a reliability and proof layer for autonomous AI agents. It lets developers register agents, install copy-ready SDK credentials, and capture every execution as a structured trace. Each trace combines LLM calls, tool calls, viem/EVM activity, KeeperHub execution status, retries, payloads, and shareable verification evidence in one forensic dashboard. The goal is to make autonomous onchain execution inspectable, auditable, and trustworthy for operators, partners, and users.

Tracer is built as a pnpm monorepo with a Next.js dashboard, Next.js/tRPC backend, Prisma/Postgres database, and a TypeScript SDK. Privy handles authentication. The SDK wraps agent runtimes like OpenAI, Anthropic, Vercel AI SDK, LangChain, and viem clients so model calls, tool calls, and onchain actions are captured automatically.

KeeperHub is used as the reliable execution layer for onchain workflows and direct contract execution. Tracer stores KeeperHub execution IDs, statuses, retries, and payloads inside the trace timeline, then displays them in a clean trace inspector. We also built worker services for enrichment and anchoring so traces can include analysis, gas metadata, share tokens, and onchain verification context.

The frontend focuses on a forensic console: agent registration, credentials/install snippets, trace lists, horizontal event timelines, metadata grids, and payload inspectors. The hardest part was connecting multiple surfaces — SDK events, KeeperHub execution status, auth, database persistence, and onchain proof — into one coherent trace that judges and developers can understand quickly.