# withHuman withHuman intercepts consequential AI-agent tool calls and applies policy rules or routes them to a human reviewer before execution. It supports agent teams that need review and an audit trail for sensitive actions. - Findings24: - Website: - Pricing: Freemium - Categories: AI Agent Infrastructure, AI Gateway And Routing ## What is withHuman? withHuman is an approval layer for AI agents. It intercepts consequential tool calls before they reach the target service, then applies configured rules to approve or deny routine actions or hold requests for human review. Reviewers receive the request context, decide whether to approve it, and the agent continues after an approved call runs. The product is framework-agnostic and offers agent integrations, review channels, escalation paths, and an append-only decision history. ## What can withHuman do? ### Intercept agent tool calls The Agent Approval Protocol (AAP) sends tool calls to withHuman before execution. The site also lists SDKs for agent frameworks and hooks for coding agents. ### Policy-based approval pipelines Rules can evaluate the agent, tool, arguments, and agent context to automatically approve or deny calls, or request human review. Webhooks can connect a policy service; an AI judge is also available for cases rules cannot describe. ### Route reviews to teams and on-call people Escalation paths assign requests to a team or person, support team escalation order and on-call flows, and can escalate when nobody responds in time. ### Notify reviewers across channels Approval requests can be delivered through Slack, Discord, email, text message, WhatsApp, and iPhone or Android apps, depending on the plan. ### Keep an approval audit trail The append-only audit log records requests, decisions, reviewer identity, notes, and request details. ## Use Cases ### Review sensitive customer refunds Hold refunds above a configured threshold for a support or payments reviewer while allowing policy-defined routine refunds to proceed automatically. ### Control production infrastructure changes Require human review for consequential operations such as Kubernetes changes or rolling back a service, with requests directed to the responsible team. ### Gate database and code operations Use policies to review potentially destructive database tool calls or coding-agent actions before they reach their target. ### Approve finance transactions Route an agent's payment or invoice action to an appropriate reviewer, with the tool call and its context available for a decision. ## Frequently Asked Questions ### How does an approval affect an agent's tool call? withHuman intercepts the call before it reaches the target service. A pipeline can approve or deny it automatically or hold it for a human decision; an approved call then runs and the agent can continue. ### Can routine calls avoid human review? Yes. Approval rules can automatically approve or deny calls based on the agent, tool, arguments, and context. The product also lists an AI judge for calls that rules cannot describe. ### How can teams connect agents to withHuman? The site describes the open Agent Approval Protocol, SDKs for frameworks including LangGraph, CrewAI, and Mastra, and hooks for coding agents such as Claude Code and Codex. ### Where do reviewers receive requests? Listed destinations include Slack, Discord, email, text message, WhatsApp, and iPhone and Android apps. Availability varies by plan. ### What does the audit log retain? The audit log keeps an append-only history of requests, decisions, reviewer identity, notes, and the request associated with each decision. ## Quick Facts - **Product type**: Approval and policy layer for AI-agent actions - **Integration approach**: Agent Approval Protocol, framework SDKs, and coding-agent hooks - **Review workflow**: Automatic policy decisions or human approval before execution - **Hosting options**: withHuman cloud; Scale also lists on-premises or bring-your-own-cloud hosting - **Open-source option**: Open-source version available on GitHub