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.

The lap CLI detects the claude-agent-sdk harness automatically and switches to chat REPL mode — no WebSocket TTY, just a simple > prompt that sends each line to the agent and prints the response.
First time? Install the CLI and log in:
git clone https://github.com/BerriAI/litellm-agent-platform.git
cd litellm-agent-platform/cli && npm install
ln -sf "$PWD/bin/lap.mjs" ~/.local/bin/lap
lap login   # paste your LAP URL + MASTER_KEY

​
1. Create an agent

In the UI choose claude-agent-sdk from the Harness picker, or via API: 
curl -X POST $LAP_URL/api/v1/managed_agents/agents \
  -H "Authorization: Bearer $MASTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "my-sdk-agent",
    "harness_id": "claude-agent-sdk",
    "model": "anthropic/claude-sonnet-4-5",
    "prompt": "You are a helpful assistant.",
    "repo_url": "https://github.com/your-org/your-repo"
  }'

​
2. Open a session

lap my-sdk-agent

  ✓ agent my-sdk-agent (0f21c021, harness=claude-agent-sdk)
  ✓ session c3970704
  waiting for sandbox. ready
  Chat mode — Ctrl-D to exit

  >
The sandbox spins up, clones the repo, and drops you into a chat prompt. Type a message and press Enter. The agent responds inline. 
  > Summarise the top-level README in three bullets.
  
  Here's the README in three bullets:
  - LiteLLM Agent Platform runs coding agents in sandboxed Kubernetes pods...
  ...

  >
Press Ctrl-D to end the session.

​
3. How it works

Each line you type is sent to POST /sessions/:id/message. The harness runs a full Claude Agent SDK turn — tool calls, file edits, multi-step reasoning — and returns when the turn is complete. The response text is printed and the prompt returns. The pod’s env contains only stub credentials. The vault proxy swaps them for real keys on every outbound TLS connection, so the agent never holds real API keys.

​
Creating agents via API (full options)

curl -X POST $LAP_URL/api/v1/managed_agents/agents \
  -H "Authorization: Bearer $MASTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "my-sdk-agent",
    "harness_id": "claude-agent-sdk",
    "model": "anthropic/claude-sonnet-4-5",
    "prompt": "You are a senior engineer. Think step by step.",
    "repo_url": "https://github.com/your-org/your-repo",
    "branch": "main"
  }'

​
Local dev (no Kubernetes)

You can run against a local harness without spinning up a kind cluster. Point .env at the local harness and the platform skips K8s entirely: 
# .env additions
LOCAL_SANDBOX_URL=http://localhost:4096
WARM_POOL_SIZE=0

# Start the harness
cd harnesses/claude-agent-sdk && npm run build
REPO_DIR=/path/to/local/repo \
LITELLM_API_BASE=https://gateway.litellm.ai/ \
LITELLM_API_KEY=sk-... \
node dist/server.js

# Start the platform
npm run dev

# Point lap at it
lap login   # URL: http://localhost:3000
lap my-sdk-agent
Sessions reach ready in under 2 seconds with LOCAL_SANDBOX_URL. See CONTRIBUTING.md for the full local dev guide.