mirror of
https://github.com/agentrhq/authsome.git
synced 2026-09-19 01:34:19 +08:00
Priyansh Khodiyar
97 lines
4.4 KiB
Plaintext
97 lines
4.4 KiB
Plaintext
---
|
|
title: "Agent frameworks"
|
|
sidebarTitle: "Overview"
|
|
description: "Drop authsome into any agent framework. Claude Code, Cursor, OpenCode, LangChain, LlamaIndex, and others."
|
|
keywords: ["agent credential management", "agent oauth", "agent api keys", "authsome agent frameworks", "authsome langchain", "authsome cursor"]
|
|
---
|
|
|
|
Authsome is provider-side, not framework-side. It manages credentials for the services your agents call.
|
|
|
|
The supported pattern for every framework is the same: **wrap the agent invocation with `authsome run --`**. The local HTTP proxy intercepts outbound requests and injects the right `Authorization` header per provider. The agent's environment never contains real secrets.
|
|
|
|
If your agent is Python-native and you're embedding authsome inside a larger orchestrator with custom storage or wiring, the library is available as a secondary surface. For everything else, the CLI and proxy are the right tools.
|
|
|
|
This section has a page per agent framework with the install path, the proxy wiring, and any framework-specific notes.
|
|
|
|
## Frameworks covered
|
|
|
|
<Columns cols={2}>
|
|
<Card title="Claude Code" icon="robot" href="/integrations/agents/claude-code">
|
|
Anthropic's CLI for coding agents. Authsome ships as a bundled skill.
|
|
</Card>
|
|
<Card title="Codex" icon="terminal" href="/integrations/agents/codex">
|
|
OpenAI's coding agent. Run agents through the auth proxy.
|
|
</Card>
|
|
<Card title="Cursor" icon="cursor" href="/integrations/agents/cursor">
|
|
The Cursor IDE agent. Wire up MCP servers and shell commands against authsome.
|
|
</Card>
|
|
<Card title="OpenCode" icon="code" href="/integrations/agents/opencode">
|
|
OpenCode CLI. Use the proxy or env-var export.
|
|
</Card>
|
|
<Card title="Cowork" icon="users" href="/integrations/agents/cowork">
|
|
Multi-agent runtime. Per-agent profiles isolate credentials.
|
|
</Card>
|
|
<Card title="Hermes Agent" icon="feather" href="/integrations/agents/hermes">
|
|
Nous Research's self-improving agent. Installs the authsome skill via agentskills.io.
|
|
</Card>
|
|
<Card title="NanoClaw / OpenClaw" icon="microchip" href="/integrations/agents/nanoclaw">
|
|
Lightweight agent runners. Same patterns as the library path.
|
|
</Card>
|
|
<Card title="LangChain" icon="link" href="/integrations/agents/langchain">
|
|
Inject tokens at LLM/tool boundaries.
|
|
</Card>
|
|
<Card title="LlamaIndex" icon="layer-group" href="/integrations/agents/llamaindex">
|
|
Wrap data-loader credentials behind authsome.
|
|
</Card>
|
|
<Card title="OpenAI Agents SDK" icon="rocket" href="/integrations/agents/openai-agents-sdk">
|
|
Replace `OPENAI_API_KEY` env-var management with proxy injection.
|
|
</Card>
|
|
<Card title="Anthropic SDK" icon="code-branch" href="/integrations/agents/anthropic-sdk">
|
|
Same idea for `ANTHROPIC_API_KEY`.
|
|
</Card>
|
|
<Card title="Generic Python agent" icon="python" href="/integrations/agents/python">
|
|
Any Python script that calls third-party APIs.
|
|
</Card>
|
|
</Columns>
|
|
|
|
## The recommended pattern
|
|
|
|
```bash
|
|
authsome run -- python my_agent.py
|
|
```
|
|
|
|
The agent runs as a subprocess of authsome. A local HTTP proxy intercepts outbound requests and injects the right `Authorization` header per provider. The child process never sees the real secret because the proxy substitutes a placeholder env var like `OPENAI_API_KEY=authsome-proxy-managed`. See [Run agents with the proxy](/guides/run-agents-with-proxy).
|
|
|
|
For SDKs that can't route through an HTTP proxy (pinned TLS, non-HTTP protocols), fall back to:
|
|
|
|
```bash
|
|
eval "$(authsome export <provider> --format env)"
|
|
python my_agent.py
|
|
```
|
|
|
|
Same vault, same refresh, but the key is in the agent's environment.
|
|
|
|
### Embedding the library
|
|
|
|
If you're building a larger Python orchestrator around authsome, you can drop below the CLI:
|
|
|
|
```python
|
|
from authsome.server.dependencies import create_auth_service
|
|
|
|
auth = create_auth_service()
|
|
token = auth.get_access_token("github", connection="default")
|
|
```
|
|
|
|
This is the embedding case. For everything else, the proxy is simpler. See [Python library](/reference/python-library).
|
|
|
|
## Per-framework notes
|
|
|
|
Each page in this section covers:
|
|
|
|
- How to invoke authsome from that framework (skill, command wrapper, or library call).
|
|
- The recommended security pattern.
|
|
- Multi-account workflows where applicable.
|
|
- Framework-specific gotchas (TLS pinning, env-var precedence, MCP wiring).
|
|
|
|
If a framework you use isn't listed, the generic patterns are in [Run agents with the proxy](/guides/run-agents-with-proxy) and [Python library](/reference/python-library).
|