withHuman
withHuman intercepts consequential AI-agent tool calls, applies approval rules, and routes requests to human reviewers when needed. Engineering, support, and finance teams use it to gate agent actions and retain a decision history.
One of 653 tools in AI Agents
Who it's for
- Engineering teams using coding agents such as Claude Code or Codex that need review gates for consequential tool calls.
- Support teams whose agents issue refunds and need rules to route requests to human reviewers.
- Finance teams using agents for vendor payments who need requests escalated to designated reviewers.
How it fits your workflow
Connect a supported agent with the CLI or MCP gateway, or integrate a custom agent through AAP. The agent's tool call is intercepted before it reaches its target.
Configure rules using the agent, tool, arguments, and context. The pipeline can automatically approve or deny matching calls, use an AI judge for calls rules cannot describe, or send the request to a human.
Route requests to teams or individuals and notify reviewers through supported channels. Reviewers see the tool call and request context; decisions and notes are retained in the audit log.
Pricing
The vendor doesn't publish prices. Check with withHuman directly.
Prices checked on Oct 15, 2026 from the vendor's site. They can change; confirm before you buy.
Key features
- Agent tool-call interception
- Intercepts agent tool calls and holds them for approval before they reach the target tool. The page describes hooks for coding agents and AAP support for custom agents.
- Approval pipelines
- Applies conditions based on the agent, tool, arguments, and agent context to approve or deny routine requests or send them to a human. The pipeline can also use an AI judge for calls that rules cannot describe.
- Escalation paths
- Routes requests to teams or individuals based on rules, supports team escalation order and on-call flows, and can escalate when nobody answers in time.
- Reviewer notifications
- Sends approval requests through Slack, Discord, iPhone and Android apps with push notifications, email, text message, or WhatsApp.
- Audit log
- Keeps an append-only history of requests, decisions, reviewer identities, notes, and the request context.
- Open protocol and self-hosting
- Uses the open Agent Approval Protocol (AAP); the page says the core, SDKs, adapters, and conformance tests are open source. The core can be run on a team's own infrastructure or used as a hosted service.
Works with
- Claude Code, Codex, OpenClaw, Pi, Hermes, and DeepSeek
- MCP gateway
- Agent Approval Protocol (AAP)
- Slack and Discord
- Email, text message, WhatsApp, iPhone, and Android
Alternatives to consider
-
Pushary
Choose it if you want a control panel that delivers agent permission requests to your phone, Slack, and web dashboard, rather than withHuman's described policy pipelines, escalation paths, and append-only audit log.
Getting started
curl -fsSL https://downloads.withhuman.ai/install.sh | sh -s -- onboard- Run the CLI onboarding command to connect an agent on the machine.
- Complete setup in the browser and approve or deny the request that appears.
- For a custom agent, use AAP; the page provides an example API request format.