AgentSyncSubmit agent transcripts with one command

Contribute the full trajectories of coding agents — Claude Code, Codex, and more — with privacy you actually own.

⚡

One command

npx agentsync submit detects your latest session, redacts it, previews exactly what will leave your machine, and uploads. No interactive login at submit time.

🔒

Privacy you own

Redaction and anonymization run locally before anything is sent. Configurable per use case via named profiles — strict, research, hiring-signal — each with its own posture.

🧩

Any agent

Every agent's native log is parsed by a pluggable adapter into one normalized transcript. Seeded with Claude Code and OpenAI Codex; new agents are just an adapter.

📐

One normalized shape

A single versioned wire format — TypeScript types plus JSON Schema — that the CLI, the ingestion API, and every downstream consumer agree on.

The submit flow

bash

agentsync submit
  ├─ detect    adapters scan disk for recent sessions
  ├─ parse     native format → normalized Transcript
  ├─ redact    apply the active privacy profile (secrets, identifiers, content)
  ├─ preview   show exactly what will be uploaded; require confirm (unless --yes)
  └─ upload    POST to ingestion API with your key

Why it exists

A growing amount of real engineering now happens through agents. The trajectory — what was asked, what the agent tried, which tools it called, what it fixed — is a uniquely valuable artifact. AgentSync makes that artifact easy to contribute and safe to share, unlocking trajectory corpora, opt-in hiring signal, and agent-tooling research.

Status

This project is in its specification phase: the transcript schema, privacy model, adapter interface, and API contract are defined first, so the client, server, and any consumer agree on one shape before code is written.

Start with the architecture, or jump to the privacy model and the transcript schema.