keychains.dev - Secure your OpenClaw bot

The second feature group encompasses the security architecture built from the ground up for autonomous agents. Keychains uses SSH Key Identity for machine authentication, meaning every machine that makes requests must authenticate via SSH keypairs, eliminating the need for passwords or API keys in the agent's environment. Additionally, Stateful Fingerprinting ensures that machines exchange cryptographic fingerprints with every API call, so even if a key is leaked, it is invalidated on first use by an unauthorized machine. The system also provides Full Transparency, offering users a complete audit trail showing every permission granted, every agent, and every task performed. Finally, Instant Revocation allows administrators to revoke any machine's access with a single click, with no waiting periods or grace times, giving immediate control over credential exposure.

The third major capability is sub-agent delegation, which is essential for complex multi-agent systems. Keychains allows parent agents to create scoped delegate tokens for sub-agents, forking only the specific permissions needed for a given subtask. The parent agent retains full control and can revoke the delegation at any time. Alternatively, blank tokens can be spawned with zero initial permissions, requiring the sub-agent to prompt the user for approval each time it needs new access. This ensures informed consent for every action. This feature is critical for safely scaling agent teams where different agents have different responsibilities, and it prevents a single compromised agent from gaining access to all credentials.

Overall, Keychains works as a credential proxy that sits between AI agents and their target APIs. The workflow is straightforward: Step 1, the agent uses `keychains curl` with template variables instead of real credentials. Step 2, when a new API scope is needed, the user sees exactly what the agent wants to do and approves with one click. Step 3, Keychains injects the correct credentials server-side, so the agent never touches raw secrets. The architecture is designed for agent-to-API traffic with agent-specific primitives like SSH-based machine identity, user consent flows, and multi-agent delegation. This contrasts with traditional secrets managers that protect at rest and API gateways that handle server-to-server traffic. Keychains focuses on the unique requirements of autonomous agents.

Concrete use cases demonstrate the value. Consider a development team building an AI coding assistant that needs access to GitHub repositories and the OpenAI API. Instead of embedding tokens in the agent's environment, they use keychains curl with template variables. When the agent needs to read a private repo, the user approves the scope, and credentials are injected without exposure. Another scenario is a multi-agent system where a manager agent delegates tasks to specialized sub-agents. The manager creates a scoped delegate token for a sub-agent to access only the Stripe payments API, while retaining other permissions. If the sub-agent is compromised, the manager instantly revokes its access. Security teams benefit from full audit trails, ensuring compliance with data protection regulations.

Keychains targets AI agent developers, platform engineers, and security professionals. It integrates with any programming language or framework that can make HTTP requests, as it works as a command-line proxy. The tech stack is agnostic, but it particularly suits teams using LangChain, AutoGPT, or custom agent frameworks. While pricing details are not explicitly provided, Keychains offers a free tier for exploration with a "Get Started" call to action. The product supports thousands of API providers including GitHub, Google, Slack, Stripe, Notion, and more, handling OAuth 2.0 with PKCE, API keys, basic auth, and custom headers. By securing credential delegation at the proxy level, Keychains enables organizations to deploy AI agents with confidence, knowing credentials remain under user control.