Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.litellm-agent-platform.ai/llms.txt

Use this file to discover all available pages before exploring further.

LiteLLM Agent Platform (LAP) gives you a self-hosted environment for running coding agents — Claude Code, Codex, opencode, and others — inside isolated Kubernetes sandboxes. A built-in credential vault proxy ensures your real API keys and tokens are never exposed to the agents themselves, while the lap CLI lets you attach a local terminal directly to any running sandbox.

Quick Start

Install the lap CLI, connect to your platform, and open your first sandbox in under five minutes.

Self-Hosting

Deploy LiteLLM Agent Platform on Kubernetes locally with kind, or on AWS EKS for production.

API Reference

Manage agents and sessions programmatically with the REST API.

Core Concepts

Understand agents, sessions, the vault proxy, and harness types.

How it works

1

Create an agent

Define your coding agent — choose a model, write a system prompt, pick a harness (Claude Code, Codex, opencode, or claude-agent-sdk), and optionally point it at a Git repo.
2

Open a session

Start a sandbox with lap <agent-name> from the CLI, via the web UI, or by calling POST /api/v1/managed_agents/agents/{id}/session. A fresh Kubernetes pod spins up in seconds.
3

Work securely

The vault proxy intercepts all outbound HTTPS from the sandbox and swaps stub credentials for real ones at the wire level. The agent process never sees your actual keys.
4

Send messages or attach a terminal

For API harnesses (opencode, claude-agent-sdk), send tasks via POST /sessions/{id}/message. For TUI harnesses (Claude Code, Codex), attach your local terminal over WebSocket with lap.

Key features

Credential Vault

Agents receive stub tokens. The vault sidecar swaps them for real credentials on every outbound TLS connection.

lap CLI

Attach your local terminal to any sandbox over WebSocket. Same feel as SSH, but sandboxed.

Skills Library

Package reusable instructions as skills and attach them to any agent at creation time.

Linear Integration

Delegate Linear issues to agents automatically via OAuth webhooks.