# LiteLLM Agent Platform

## Docs

- [POST /api/v1/managed_agents/agents — create an agent](https://docs.litellm-agent-platform.ai/api-reference/agents/create.md): Creates a new agent with a model, system prompt, harness type, repo URL, and persistent environment variables. Returns the new agent object.
- [DELETE /api/v1/managed_agents/agents/{agent_id}](https://docs.litellm-agent-platform.ai/api-reference/agents/delete.md): Deletes an agent by ID, stops all active sessions and sandbox pods associated with it. Returns 204 No Content. Nonexistent agents return 404.
- [GET /api/v1/managed_agents/agents/{agent_id}](https://docs.litellm-agent-platform.ai/api-reference/agents/get.md): Retrieves a single agent by ID. Returns the full agent object including model, prompt, harness type, env vars, and attached skill IDs.
- [GET /api/v1/managed_agents/agents — list all agents](https://docs.litellm-agent-platform.ai/api-reference/agents/list.md): Returns a paginated list of agents with filtering by name, ID, or harness type, and sorting by creation date, name, or session count.
- [PATCH /api/v1/managed_agents/agents/{agent_id}](https://docs.litellm-agent-platform.ai/api-reference/agents/update.md): Partially updates an agent's name, prompt, model, branch, env vars, or harness image. Only fields you include in the request body are changed.
- [Memory API: store persistent context across agent sessions](https://docs.litellm-agent-platform.ai/api-reference/memory.md): Create and manage per-agent memory entries. Memories are injected into context at session start and tracked with usage statistics across runs.
- [LiteLLM Agent Platform REST API overview](https://docs.litellm-agent-platform.ai/api-reference/overview.md): HTTP REST API for creating agents, managing sessions, sending messages, and managing skills and memories. All endpoints require Bearer authentication.
- [POST /api/v1/managed_agents/agents/{agent_id}/session](https://docs.litellm-agent-platform.ai/api-reference/sessions/create.md): Creates a session and starts a sandbox pod for the specified agent. Returns immediately with status 'creating'. Poll until 'ready' before sending messages.
- [DELETE /api/v1/managed_agents/sessions/{session_id}](https://docs.litellm-agent-platform.ai/api-reference/sessions/delete.md): Deletes a session and removes its sandbox pod, marking the row dead. Idempotent — calling it on an already-stopped or dead session returns success.
- [GET /api/v1/managed_agents/sessions/{session_id}](https://docs.litellm-agent-platform.ai/api-reference/sessions/get.md): Retrieves a session by ID, returning its current status, bring-up phase, sandbox URL, tty connection details, and the latest agent response.
- [POST /api/v1/managed_agents/sessions/{session_id}/message](https://docs.litellm-agent-platform.ai/api-reference/sessions/send-message.md): Sends a message to a ready session and waits synchronously for the agent to reply. Returns the agent's response as a list of typed message parts.
- [POST /api/v1/managed_agents/sessions/{session_id}/stop](https://docs.litellm-agent-platform.ai/api-reference/sessions/stop.md): Gracefully stops a running session, marks it dead, and releases the sandbox pod. Use this to free cluster resources when a session is no longer needed.
- [Skills API: create and attach reusable agent instructions](https://docs.litellm-agent-platform.ai/api-reference/skills.md): Create, update, and attach reusable instruction blocks to agents. Skills are appended to an agent's system prompt and materialized inside every sandbox.
- [What is an agent in LiteLLM Agent Platform?](https://docs.litellm-agent-platform.ai/concepts/agents.md): An agent is the persistent configuration for a coding assistant: model, system prompt, harness type, repo, env vars, and attached skills.
- [Harness types: Claude Code, Codex, opencode, and more](https://docs.litellm-agent-platform.ai/concepts/harnesses.md): A harness is the container image and runtime inside a sandbox. Pick an API harness for automation or a TUI harness for interactive terminal access.
- [What is a session in LiteLLM Agent Platform?](https://docs.litellm-agent-platform.ai/concepts/sessions.md): A session is one live run of an agent inside a sandbox pod, with a defined lifecycle from creation through ready, then to dead or failed.
- [Credential vault: keep secrets away from agents](https://docs.litellm-agent-platform.ai/concepts/vault.md): The vault sidecar intercepts all sandbox HTTPS traffic and swaps stub credentials for real values at the wire level, before requests leave the cluster.
- [Deploy LiteLLM Agent Platform to AWS EKS](https://docs.litellm-agent-platform.ai/configuration/deploy-aws.md): Step-by-step guide to deploying LiteLLM Agent Platform on AWS EKS for sandbox workloads, with web and worker services hosted on Render.
- [Environment variable reference for LiteLLM Agent Platform](https://docs.litellm-agent-platform.ai/configuration/environment.md): Environment variables that control platform behavior, sandbox runtime, warm pool, and integrations for a LiteLLM Agent Platform deployment.
- [Connect Linear to LiteLLM Agent Platform](https://docs.litellm-agent-platform.ai/guides/integrations-linear.md): Set up the Linear OAuth integration so assigning a Linear issue to your LAP agent automatically spawns a new sandbox session to work on it.
- [Use the lap CLI to attach to agent sandboxes](https://docs.litellm-agent-platform.ai/guides/lap-cli.md): Install the lap CLI, authenticate against your platform, and open a live terminal session inside any agent's sandbox over WebSocket.
- [Give agents persistent memory across sessions](https://docs.litellm-agent-platform.ai/guides/memory.md): Store text notes against an agent that survive session restarts, get injected at session start, and can be searched, updated, or disabled.
- [Build reusable skills and attach them to agents](https://docs.litellm-agent-platform.ai/guides/skills.md): Create reusable Markdown instruction blocks, store them in the skill library, and attach one or more skills to any agent's system prompt at any time.
- [Speed up session starts with the warm pool](https://docs.litellm-agent-platform.ai/guides/warm-pool.md): Keep pre-provisioned sandbox pods ready to claim so that session start time drops from ~10 seconds to under 2 seconds for recently active agents.
- [LiteLLM Agent Platform: isolated agent sandboxes](https://docs.litellm-agent-platform.ai/introduction.md): Self-hosted platform for running Claude Code, Codex, opencode, and claude-agent-sdk in isolated Kubernetes sandboxes with a built-in credential vault.
- [Quickstart: install lap CLI and open your first sandbox](https://docs.litellm-agent-platform.ai/quickstart.md): Install the lap CLI, connect to your LiteLLM Agent Platform deployment, and open your first isolated coding-agent sandbox in under five minutes.
- [Self-host LiteLLM Agent Platform on Kubernetes](https://docs.litellm-agent-platform.ai/self-hosting.md): Deploy LiteLLM Agent Platform on a local kind Kubernetes cluster for development, or to AWS EKS and Render for production use by your team.
- [Frequently asked questions about LiteLLM Agent Platform](https://docs.litellm-agent-platform.ai/troubleshooting/faq.md): Answers to common questions about session lifetimes, credential security, warm pool performance, harness selection, and concurrent session limits in LAP.
- [Diagnose and fix stuck or failed sessions](https://docs.litellm-agent-platform.ai/troubleshooting/session-issues.md): How to use the diagnose endpoint and detected_issues codes to investigate sessions stuck in creating, failed, or dead states in LiteLLM Agent Platform.