SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI agentsauthsome

authsome

authsome is an open-source credential gateway for AI agents, deployed between agents and the external services they call. Users only need to log in once via OAuth2 or API Key, and Authsome encrypts and stores the credentials and injects them at runtime through an HTTP proxy, allowing agents to remain authenticated in headless environments without manual intervention. The agents themselves never see the real credentials, avoiding the risk of credentials being scattered and leaked, while also providing a unified entry point to manage access permissions, rotate keys, and view the call behavior of each agent. The project is maintained by agentrhq, is licensed under MIT, requires Python 3.13+, supports a Docker resident daemon, and provides adapters for agent frameworks such as the OpenAI Agents SDK.

Open website

About this tool

Purpose

authsome is an open-source credential gateway that sits between AI agents and the external services they need to call. The core problem it solves is: agents do not only run in interactive sessions, but also appear in CI, SSH, cron jobs, background workers, and parallel pipelines, scenarios that require API access that survives without manual intervention. Tokens hardcoded in environment variables are prone to leaking or expiring, and writing the authentication flow, token storage, refresh handling, and per-provider configuration into every project amounts to repeatedly rebuilding the same infrastructure. Authsome acts as the credential proxy that agents call at runtime.

Key capabilities

  • No credential sprawl: All providers and all agents share a single encrypted store, centrally managed.
  • Agents never see credentials: Authentication is completed outside the agent process, avoiding leakage risk, and secrets no longer appear in environment variables.
  • No browser needed at runtime: The first setup can use browser PKCE, device code, or a browser bridge to securely enter API Keys; thereafter agents run headlessly.
  • CLI as the agent interface: After setting up once, the latest credentials are injected on demand at tool runtime.
  • Credentials are encrypted at rest and automatically refreshed before expiry.
  • Built-in providers: OAuth2 and API Key providers are supported out of the box, and the official documentation provides a complete list.
  • Agent integrations: Adapters for common agent frameworks and CLIs are included, including the OpenAI Agents SDK; see authsome.ai/docs/integrations for the complete list.

Usage

The basic flow is to first complete authentication with the CLI, after which agents obtain valid credentials on demand. Simply prefix the command to be executed with authsome run --：

authsome run -- <command>

First-time installation and run setup (including identity, claim, and importing an API Key from .env ):

# 具体命令见官方文档

For a remote daemon, pass in --base-urlonce, and that address will be saved to the client configuration for use by subsequent commands.

Run a resident daemon with Docker, with no need for Python on the host:

# 具体命令见官方文档

If using a hosted daemon, before starting the stack set AUTHSOME_BASE_URL Set as a public HTTPS URL. For initial setup, volume backups, TLS termination, and environment variable reference, see the self-hosting guide.

After adding the authsome skill to an agent (claude, codex, cursor, hermes, etc.), you can try an example task that requires access to an external service; the agent will use authsome to log in to the external service and perform the task.

Use Cases

Suitable for agent workflows that run in headless environments and require continuous access to external APIs, such as CI pipelines, SSH sessions, scheduled tasks, background workers, and parallel processing pipelines. Also suitable for teams that want to centrally manage credentials for multiple providers, uniformly rotate keys, and audit agent behavior.

Limitations and Requirements

  • Requires Python 3.13+ (when using Docker daemon mode, the host machine does not need Python).
  • The official materials do not provide specific pricing or billing information.
  • The official materials do not provide performance data, details of the complete list of supported providers, or comparisons with other gateways.
  • Licensed under MIT; see LICENSE for details.

Related tools

Browse category