docs/integrations/index.md

# Integrations Overview

ACE provides both runners and step-based integrations for popular agentic
frameworks. Some integrations are full runners, while others are composable
pipeline steps you can drop into a custom `Pipeline`.

## Available Integrations

| Runner | Framework | Input | Insight Level |
|--------|-----------|-------|--------------|
| [`ACELiteLLM`](litellm.md) | LiteLLM (100+ providers) | Questions | Micro |
| [`LangChain`](langchain.md) | LangChain Runnables | Chain inputs | Meso |
| [`BrowserUse`](browser-use.md) | browser-use | Task strings | Meso |
| [`ClaudeCode`](claude-code.md) | Claude Code CLI | Task strings | Meso |
| [`Claude SDK`](claude-sdk.md) | Anthropic Python SDK | Task strings or `ACESample` | Meso |
| [OpenClaw](openclaw.md) | OpenClaw transcripts | JSONL trace files | Meso |
| [MCP Server](mcp.md) | MCP (stdio) | Tool calls | Micro |
| [MCP Client Setup](mcp-client-setup.md) | Claude Code, Cursor, Windsurf | — | Setup Guide |
| [Opik](opik.md) | Opik observability | — | Monitoring |
| [Tracing](tracing.md) | Kayba tracing SDK | `@trace` / `start_span` | Cloud |
| [Hosted API](hosted-api.md) | Kayba hosted API | Trace files | Cloud |

## The Pattern

All integration runners follow the same three-step pattern:

```
1. INJECT   — Add skillbook strategies to the agent's context
2. EXECUTE  — Run the external agent normally
3. LEARN    — Reflector + SkillManager update the skillbook
```

## Quick Construction

Every runner offers a `from_model()` factory that builds ACE roles automatically:

```python
from ace import BrowserUse, LangChain, ClaudeCode

# Browser automation
browser = BrowserUse.from_model(browser_llm=my_llm, ace_model="gpt-4o-mini")

# LangChain chain/agent
chain = LangChain.from_model(my_runnable, ace_model="gpt-4o-mini")

# Claude Code CLI
coder = ClaudeCode.from_model(working_dir="./project", ace_model="gpt-4o-mini")
```

For direct Anthropic API usage without a runner, compose the SDK steps directly:

```python
from ace import Pipeline, Reflector, SkillManager, Skillbook, learning_tail
from ace.integrations import ClaudeSDKExecuteStep, ClaudeSDKToTrace

skillbook = Skillbook()
pipe = Pipeline([
    ClaudeSDKExecuteStep(model="claude-sonnet-4-20250514"),
    ClaudeSDKToTrace(),
    *learning_tail(Reflector("gpt-4o-mini"), SkillManager("gpt-4o-mini"), skillbook),
])
```

## Shared Features

All runners share these capabilities:

- **Skillbook persistence** — `save()` / load via `skillbook_path`
- **Checkpointing** — automatic saves during long runs
- **Deduplication** — prevent duplicate skills
- **Background learning** — `wait=False` for async learning
- **Progress tracking** — `learning_stats` property

## Which Integration Should I Use?

- **Building a Q&A or reasoning agent?** Use [ACELiteLLM](litellm.md)
- **Have an existing LangChain chain or agent?** Use [LangChain](langchain.md)
- **Automating browser tasks?** Use [BrowserUse](browser-use.md)
- **Running coding tasks with Claude Code?** Use [ClaudeCode](claude-code.md)
- **Calling Anthropic directly from your own pipeline?** Use [Claude SDK](claude-sdk.md)
- **Want to monitor costs and traces?** Add [Opik](opik.md)
- **Learning from OpenClaw session transcripts?** Use [OpenClaw](openclaw.md)
- **Exposing ACE as an MCP tool provider?** Use the [MCP Server](mcp.md) and the [MCP Client Setup](mcp-client-setup.md) guide
- **Want to send traces to Kayba from your code?** Use [Tracing](tracing.md)
- **Want to use the hosted API instead of running locally?** Use the [Hosted API](hosted-api.md) CLI
- **Using a different framework?** See the [Integration Guide](../guides/integration.md) to build a custom runner