v1 — public beta

Agents that ship code, on guardrails you control.

Joule Atelier runs local MCP servers for tasks, memory, and policy enforcement, then syncs to a multi-tenant cloud you can audit. Stop wiring bespoke hooks for every agent. Start shipping with confidence.

Start for free Read the docs

Free forever for solo use. No credit card required.

~ jouleatelier doctor

$ jouleatelier doctor
  ✓ node 22.10.0
  ✓ ~/.jouleatelier/config.json
  ✓ keychain reachable
  ✓ org joule-atelier (member)
  ✓ claude-code hooks installed
  ✓ codex hooks installed
  ✓ mcp-tasks      ready in 47 ms
  ✓ mcp-memories   ready in 89 ms
  ✓ mcp-guardrails ready in 32 ms
  ✓ bridge daemon  PONG (3 ms)
  all green — happy hacking.

The boring parts of agent infra, done right.

You shouldn’t need a side project to give your agents a memory, a policy engine, and an audit log. We built the side project so you don’t have to.

Local MCP servers

Tasks, memory, and guardrails ship as native MCP servers your agent talks to over stdio. Your code never leaves the machine until you say so.
Policy hooks for every agent

Pre-tool, post-tool, and stop hooks for Claude Code and Codex. One config, one bridge, identical guardrails across providers.
Cloud sync with RLS isolation

Aurora-backed multi-tenant API with row-level security on every table. Bring your team, audit every action, scope every credential.
BYOK reflections

Run reflection passes against the model and provider you trust. Your API keys stay in your OS keychain — never on disk, never on our servers.
Audit and observability

Every transition, every blocked tool call, every memory write is recorded with a stable correlation id. Export to your SIEM in one click.
Open contracts

Zod schemas at every boundary, OpenAPI for the cloud API, public MCP tool surface. No vendor lock-in beyond your own data.