Skip to content
← Tools / Skills
ToolHumanLayer/10 trustVerified
HumanLayer logo

HumanLayer

Human-in-the-loop approvals for AI agents

Visit official site →

About

HumanLayer is an open-source API and SDK (Python and TypeScript) that lets AI agents contact humans for approval, feedback, and answers mid-run. Developers annotate risky functions with @require_approval(); when the agent calls one, execution pauses, an approval request goes out over Slack, email, SMS, or Discord, and the human's decision is injected back into the agent's context. A human-as-tool primitive covers cases where the agent needs advice rather than a yes or no.

The company was founded by Dex Horthy (YC Fall 2024) and the repo has grown into a broader agent-harness project around his 12-factor agents work; the team now also builds an agentic IDE for complex codebases. Community docs list support for LangChain, CrewAI, OpenAI, Claude, ControlFlow, Vercel AI SDK, and Mastra, though that compatibility list is community-reported rather than vendor-tested.

What it can do

  • Approval gates on agent function calls
  • Human-as-tool for mid-run questions and feedback
  • Slack, email, SMS, and Discord contact channels
  • Granular routing of approvals to teams or individuals
  • Audit trail with run and call IDs
  • Python and TypeScript SDKs plus REST API

Setup

  1. Install the SDK: pip install humanlayer (Python) or npm i humanlayer (TypeScript)
  2. Sign up at humanlayer.dev and create an API key
  3. Wrap risky functions with @hl.require_approval() or the TypeScript equivalent
  4. Configure a contact channel (Slack, email, SMS, or Discord) for approval requests
  5. Run your agent; approvals pause execution until a human responds

Prerequisites

Python 3.10+ or Node 18+, a HumanLayer API key, and a configured contact channel.

Use cases

human-in-the-loopagent-safetyapproval-workflowsagent-oversight

Works with

Trust score — 8/10

/10 trustVerified

Scored 8/10 (TIER verified, metrics-v1). Evidence: 11,703 GitHub stars (api.github.com, 2026-10-16); pushed 2026-10-15; Apache-2.0 license verified from the LICENSE file; YC F24 company. Community docs corroborate framework support (LangChain, CrewAI, OpenAI, Vercel AI SDK, Mastra). Strengths: large open-source adoption, very active, real production pattern for agent safety. Docked for: (1) the company has pivoted toward an agentic IDE, so the original approval SDK may get less focus; (2) approval latency depends on human response times, a limit the vendor acknowledges. License: Apache-2.0.

★ 11,703 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
9.0/10
Maintenance · recency, releases, issue load
9.0/10
Openness · license + docs
8.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · 11,703 GitHub stars via api.github.com/repos/humanlayer/humanlayer (2026-10-16)
  • · Last push 2026-10-15T19:57:34Z - very active
  • · YC Fall 2024 company founded by Dex Horthy
  • · Company has pivoted toward an agentic IDE per aaif.io (2026-09); original HITL SDK remains open source
  • · Framework support list (LangChain, CrewAI, Vercel AI SDK, Mastra) is community-reported

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.