Tool-call interception
Intercepts agent tool calls before they reach services such as Stripe, allowing the call to be evaluated before execution.
withHuman
withHuman is an approval layer for AI agents that intercepts consequential tool calls and applies automated rules or human review before execution. It helps support, finance, engineering, and operations teams let agents perform real work while keeping sensitive actions accountable.
Human approval for AI agent actions
withHuman is an approval layer for AI agents. It intercepts consequential tool calls before they reach the connected service, then applies configured rules, automated checks, or human review before allowing the action to proceed.
The product is designed for teams that want agents to perform operational work while retaining control over actions such as refunds, payments, database changes, deployments, and messages. Approved calls continue to the target service, while denied requests remain blocked; decisions and context are recorded in an audit log.
Approval controls for consequential agent actions
Policy-based approval pipelines
Builds approval pipelines with rules based on the agent, tool, arguments, and agent context. Rules can approve or deny routine requests automatically.
AI and webhook policy checks
Uses an AI judge for calls that are difficult to describe with fixed rules, with webhooks available for custom policy services.
Escalation routing
Routes requests to the appropriate team or person, with escalation paths, team escalation policies, and on-call flows for unanswered requests.
Multi-channel reviewer notifications
Sends approval requests through Slack, Discord, email, text message, WhatsApp, and mobile apps, with availability varying by plan.
Audit log
Keeps an append-only record of requests, decisions, notes, reviewer identity, and the associated request context.
Where teams can use withHuman
Customer support refunds
A support agent can identify a duplicate charge and request a Stripe refund. Rules can approve smaller refunds automatically while sending higher-value refunds to the Payments team for review.
Finance and vendor payments
A finance agent can prepare a vendor payment, such as an annual invoice, and pause the action until an authorized reviewer approves or denies it.
Code and infrastructure operations
Coding or SRE agents can request approval for changes such as merging a pull request or rolling back a Kubernetes deployment when an operational threshold is crossed.
Sensitive data and communications actions
Teams can place database changes, outbound messages, and other difficult-to-reverse tool calls behind policies and accountable review before execution.
Pros and Cons
Pros
- Intercepts agent actions before execution rather than relying only on post-action review.
- Supports both automatic policy decisions and escalation to named teams or human reviewers.
- Offers several connection methods for coding agents and custom agent frameworks, including an open Agent Approval Protocol.
- Provides multi-channel notifications and an append-only record of decisions and reviewer identity.
Cons
- Some capabilities, including MCP gateway access, broader reviewer channels, AI judge availability, and administrative controls, vary by plan.
- The fetched product pages show only a limited set of concrete service examples, so the exact integration experience for other systems should be verified.
FAQ
How does withHuman handle an agent action?
withHuman intercepts an agent tool call before it reaches the target service. An approval pipeline can approve or deny the call automatically, apply an AI judge or webhook policy, or send it to a designated human reviewer. Approved calls then continue to the target service.
How can agents connect to withHuman?
The product supports agents through hooks for coding agents, the Agent Approval Protocol API, an MCP server, an MCP gateway, gateway tunnels, and SDKs for frameworks including LangGraph, CrewAI, and Mastra. The site also lists integrations for Claude Code, Codex, and other agent frameworks.
Where do human reviewers receive approval requests?
Reviewers can receive approval requests through Slack, Discord, email, text message, WhatsApp, and the iPhone and Android apps, depending on the plan. The request includes the tool call and relevant context so the reviewer can approve or deny it.
Can routine requests be handled without a human?
Yes. Approval pipelines can use conditions involving the agent, tool, arguments, and agent context. They can automatically approve or deny routine requests and route other requests to teams or individuals.
What deployment and plan options are available?
withHuman is available as open source, a hosted Free plan, Scale, and Enterprise. The pricing page states that Scale is $20 per user per month; Enterprise is offered through an annual contract, and plan capabilities differ.
Quick Facts
- Category
- AI agent governance and approvals
- Deployment options
- Open source, hosted Free, Scale, and Enterprise
- Supported agent connectivity
- Coding-agent hooks, MCP, Agent Approval Protocol API, SDKs, and HTTP API
- Reviewer channels
- Slack, Discord, email, text, WhatsApp, iPhone, and Android; plan-dependent
- Example workflow
- Intercepts a Stripe refund, applies a threshold rule, and routes larger refunds to a human
- Source domain
- withhuman.ai
Альтернативы withHuman
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads inside Firecracker micro-VMs. It is designed for workflows that need machine-level isolation, private networking between sandboxes, and programmatic control through SDK, CLI, or MCP.
GitBot
GitBot is a local browser-based workspace for creating reusable coding bots powered by Claude Code, Codex, or OpenCode. It helps developers run those agents across repository folders with persistent threads and configurable permissions.
Codex Plugins
Codex Plugins bundle reusable skills, app integrations, and MCP servers into workflows you can install in the Codex app or use from Codex CLI. They help extend Codex with connected-service tasks, reusable instructions, and shared team workflows.
Wallie
Wallie is an open-source AI streamer that watches your screen, hears chat, and generates live commentary in a configurable persona. It runs locally on your machine with your own keys and is aimed at faceless content, autonomous streams, and real-time reactions.
AakarDev AI
AakarDev AI helps teams manage AI provider access, project-level setups, logs, and analytics from one dashboard. It supports BYOK workflows and lists providers including OpenAI, Google Gemini, Anthropic, Groq, Mistral AI, and Perplexity AI.
Trigger.dev chat agent
Trigger.dev chat agent is a durable AI chat backend for developers building stateful conversations that can survive refreshes, crashes, and long-running turns. It connects with the AI SDK `useChat` flow and runs on managed infrastructure with no timeout on a turn.