All tools / HumanLayer

Tool profile

HumanLayer: Shared Coding-Agent Sessions, Plans and Reviews

HumanLayer brings coding-agent sessions, plans and diffs into a shared workspace. Agents run on connected hosts, while the team reviews work through desktop or web interfaces. It is a coordination layer around your coding tools and compute, not a replacement model subscription.

Official access and docs

Checked October 25, 2026 · Official-source review; no hands-on benchmark

From an individual session to a shared task

The current repository describes support for Claude Code, Codex, OpenCode and Pi, with a daemon on each host and synchronized tasks, sessions, plans and reviews. The product page shows shared artifacts and comment-driven design review.

A useful workflow is to attach the research, plan and implementation session to one task, then review the plan before accepting a diff. Ask whether your team needs that shared context; a solo, short-lived coding task may not need another workspace.

Cloud plan and model costs are separate

The current pricing section advertises Starter for up to three team members and 200 sessions per month. Pro is listed at $100 per user per month, with bring-your-own-key support and unlimited tasks and sessions.

Those plan labels do not establish unlimited model tokens or free worker machines. Budget for the chosen agent/model account and compute separately, and confirm the billable seats and session definition before upgrading. We did not create an account or verify checkout.

Source code is available; self-hosting still needs work

The repository README states Apache-2.0 licensing, Bun and Docker prerequisites, with Rust also needed for the native desktop build. It says self-hosting guides are in progress and warns that the Compose starter uses simple/default passwords.

The product website FAQ says the code is not yet open source; the linked repository now contains the application and its license. Use the repository's current license and setup notes when evaluating source access. Do not equate a source build with a complete managed-service deployment.

Evaluate with one disposable project

Try one non-sensitive repository and one host first. Check whether a teammate can understand the plan, inspect the actual changes and see which session is still running. Verify what happens when the host disconnects and how access is revoked; a UI status alone is not proof that work completed.

Retain normal code review and tests before merging. This is an editorial evaluation checklist, not a hands-on result. Start with the source and setup notes or the hosted product.