HumanLayer has released the core of its coding-agent workspace under Apache 2.0. It groups tasks, agent sessions, plans and code reviews across desktop and web interfaces. A background daemon runs the agents on your own workstation or remote machine.

The October 23 announcement covers the code behind HumanLayer Cloud, including its API, desktop shells, synchronization service and agent plugins. This is a meaningful source release of an existing product.

  • Who it serves: developers and teams coordinating Claude Code, Codex, OpenCode or Pi sessions. Plans and comments stay attached to the work, giving teammates a shared place to review an agent’s direction.
  • Hosted setup: the remote-daemon guide uses npm install -g @humanlayer/cli@latest, then humanlayer login and humanlayer daemon launch. The browser connects to that host through HumanLayer. Persistent Linux operation needs a supervisor such as a systemd user service.
  • Self-hosted setup: the repository requires Bun and Docker, with Rust for the Tauri desktop build. Its local stack includes Postgres, Electric for syncing database changes, Redis and object storage.
  • Cost: the source has no license fee; infrastructure and agent access remain your responsibility. The hosted pricing page currently lists free Starter access for up to three people and 200 sessions per month, and Pro at $100 per user monthly. Bring your own agent subscriptions or API keys.

Caveat: self-hosting is still rough: the announcement says some services require third-party keys, including WorkOS, Slack and Linear, even when unwanted. The README also warns that Compose defaults use simple passwords that must be changed before deployment.

Opening the whole workspace makes it possible to inspect and adapt the review flow alongside the agent runner. Treat it as an integration project; these capabilities are documented, not independently tested here.