HumanLayer

HumanLayer: an IDE built for supervising AI coding agents, not typing code

HumanLayer started life as a human-in-the-loop SDK: an API that let AI agents pause and request human approval before executing consequential actions like database writes or outbound emails. The company has since evolved that idea into something bigger — a full IDE and collaborative cloud platform for teams that ship most of their code through AI agents. Instead of one developer babysitting one chat window, HumanLayer groups agent sessions, artifacts, and git worktrees into shared task workspaces where multiple engineers and multiple agents work in parallel.

The core bet is that agents fail on hard problems not because models are weak, but because context and verification are unmanaged. HumanLayer's answer is a structured workflow (research, plan, implement) with human verification checkpoints between phases, so a reviewer signs off on the research and the plan before an agent burns tokens on implementation. The platform is BYOK — you bring your existing Claude, Codex, or other model subscriptions rather than paying per-token markups.

Key Features

• Task workspaces that group agent sessions, artifacts, and git worktrees for team-level collaboration
• Structured multi-phase workflow (questions, research, plan, implement) with human checkpoints between phases
• Comment-driven review of research docs and plans before agents write code
• Local and cloud daemons for running parallel agent sessions on your laptop or remote infrastructure
• BYOK model: use your own Claude, Codex, and other AI subscriptions with no per-token billing
• Multi-repo workspaces plus a free starter tier (up to 3 team members, 200 sessions/month)

Ideal Use Case

HumanLayer fits engineering teams that have already adopted Claude Code or similar agents and are hitting the ceiling of ad-hoc usage: agents producing plausible-but-wrong changes in large, complex codebases. If your team wants to run many agent sessions in parallel while keeping a human review gate on research and plans — rather than reviewing 2,000-line diffs after the fact — this is the workflow layer for that. Solo tinkerers on small greenfield projects will get less from it.

How HumanLayer differentiates

Most agent tooling optimizes for autonomy; HumanLayer optimizes for verified delegation, moving the human checkpoint from the final diff to the plan stage where corrections are cheap. The company is a Y Combinator (F24) portfolio startup founded by Dex Horthy — who popularized the term "context engineering" — and its open-source repository at github.com/humanlayer/humanlayer has over 11,000 GitHub stars. The site lists engineers at Upstart, Casco, and Osmosis among its users, with backing from Pioneer Fund and other funds alongside YC.

FAQ

Is HumanLayer still a human-in-the-loop API? The original SDK for approval workflows remains open source, but the flagship product is now the IDE and cloud platform for orchestrating coding agents with human checkpoints.

Which AI models does it work with? It is BYOK: you connect your existing Claude, Codex, or other agent subscriptions, so there is no separate per-token bill from HumanLayer.

Is there a free plan? Yes. The Starter tier is free for up to 3 team members and 200 agent sessions per month; Pro is $100 per user per month.

Does it support multiple repositories? Yes — multi-repo workspaces are supported on paid tiers, and Enterprise adds SSO/SAML, audit logs, and on-prem options.

tl;dr

HumanLayer is a YC-backed IDE and cloud platform for running parallel AI coding agents with human review checkpoints at the plan stage. Free to try, $100/user/month for Pro.

Related

Looking for more options? Browse the Developer Tools directory or read our best AI coding tools listicle. HumanLayer is also tracked on Crunchbase.