# agenthold

> Self-hosted, multi-user platform for running many persistent AI agents, each in its own sandbox.

Full documentation (discoverable current pages, one document): https://agenthold.dev/llms-full.txt

## Pages

- [agenthold](https://agenthold.dev/index.md) — Self-hosted, multi-user platform for running many persistent AI agents, each in its own sandbox.
- [Install, upgrade, back up](https://agenthold.dev/install/index.md) — This guide takes a fresh Linux server to a working agent. Everything runs in Docker on one host.
- [Automatic deploys](https://agenthold.dev/deploy/index.md) — Deploy every push to main to your host automatically with GitHub Actions: setup, secrets and rollback.
- [Sessions and harnesses](https://agenthold.dev/sessions/index.md) — agenthold does not implement an agent loop. A **harness** (Claude Code, ...) runs inside each sandbox and speaks the Agent Client Protocol (ACP) on stdio. The server is the ACP client.
- [Sandboxes](https://agenthold.dev/sandboxes/index.md) — Every agent has its own sandbox: a container with a persistent home directory at /home/agent. The server drives sandboxes through a small driver interface (internal/sandbox.Driver).
- [Agent to agent](https://agenthold.dev/agents/index.md) — Let agents in a workspace delegate tasks to each other through the platform MCP server.
- [MCP servers and skills](https://agenthold.dev/tools/index.md) — Attach MCP servers and Agent Skills to a workspace or a single agent.
- [Sharing and audit](https://agenthold.dev/sharing/index.md) — Workspace and agent roles, invitations, sharing an agent, and the audit log.
- [API, tokens and triggers](https://agenthold.dev/automation/index.md) — Call the HTTP API with personal access tokens, and run agents on schedules and webhooks.
- [Authentication](https://agenthold.dev/auth/index.md) — agenthold has no passwords of its own.
- [GitHub](https://agenthold.dev/auth-github/index.md) — GitHub sign-in uses plain OAuth2, not OpenID Connect: the server exchanges the code for an access token and then reads the user and their emails from the GitHub API.
- [Keycloak](https://agenthold.dev/auth-keycloak/index.md) — Sign in with Keycloak or any OpenID Connect provider, including the development setup.
- [Security](https://agenthold.dev/security/index.md) — The threat model and the layers that protect the host, other agents and your secrets.
- [Egress and secrets](https://agenthold.dev/egress/index.md) — Sandboxes have no route to the internet of their own. All of their traffic goes through the server, which applies the agent's network policy, injects secrets and logs every connection.
- [Hardened runtimes](https://agenthold.dev/hardening/index.md) — Run sandboxes under gVisor or Sysbox for stronger isolation from the host kernel.
- [agenthold API](https://agenthold.dev/api/llms.txt)