Skip to content

Integrations

RLM Code is designed to integrate with the broader AI ecosystem through three integration surfaces: LLM providers for model connectivity, MCP servers for tool interoperability, and framework adapters for plugging into third-party agent frameworks.

Integration Architecture

graph LR
    RLM[RLM Code] --> LLM[LLM Providers]
    RLM --> MCP[MCP Server/Client]
    RLM --> FW[Framework Adapters]

    LLM --> Local[Local Models]
    LLM --> Cloud[Cloud APIs]

    MCP --> Claude[Claude Desktop]
    MCP --> VSCode[VS Code]
    MCP --> ExtMCP[External MCP Servers]

    FW --> PydAI[Pydantic AI]
    FW --> ADK[Google ADK]

LLM Providers

RLM Code supports 20+ model providers across two connection modes:

Mode Description Providers
Local Self-hosted models, no API key required Ollama, LM Studio, vLLM, SGLang, TGI, MLX, llama.cpp, OpenAI-Compatible
BYOK Cloud APIs, bring-your-own-key OpenAI, Anthropic, Gemini, xAI, Mistral, DeepSeek, Moonshot, Alibaba, OpenRouter, OpenCode, Groq, Together, SiliconFlow, Fireworks, Perplexity, Cerebras

All providers are managed through a unified ProviderRegistry with automatic alias resolution and model inference.

See LLM Providers for the complete provider reference.

MCP (Model Context Protocol)

RLM Code implements both an MCP server and an MCP client:

Component Purpose
MCP Server Exposes RLM capabilities as MCP tools for external clients (Claude Desktop, VS Code)
MCP Client Connects to external MCP servers for tool augmentation

Server Tools

The RLM MCP Server exposes five tools:

Tool Description
rlm_execute Execute tasks using RLM paradigms
rlm_query Query large contexts efficiently
rlm_compare Compare paradigms on the same task
rlm_benchmark Run preset benchmark suites
rlm_trajectory View or export execution trajectories

Client Capabilities

The MCP Client Manager supports:

  • Three transports: stdio, SSE, WebSocket
  • Auto-reconnect on closed connections
  • Retry logic with configurable backoff
  • Environment variable resolution in configuration

See MCP Server for the full MCP reference.

Framework Adapters

Framework adapters allow RLM Code to delegate execution to third-party agent frameworks while maintaining RLM trajectory logging and reward tracking.

Framework Adapter Class Install Extra
Pydantic AI PydanticAIFrameworkAdapter rlm-code[pydantic]
Google ADK GoogleADKFrameworkAdapter rlm-code[adk]

All adapters implement the RLMFrameworkAdapter protocol and are registered through the FrameworkAdapterRegistry.

See Framework Adapters for implementation details.

Quick Configuration

# Local (Ollama)
rlm-code
> /connect ollama qwen2.5-coder:7b

# Cloud (OpenAI)
export OPENAI_API_KEY="sk-..."
rlm-code
> /connect openai gpt-4o

{
  "mcpServers": {
    "rlm-code": {
      "command": "python",
      "args": ["-m", "rlm_code.mcp.server"]
    }
  }
}

from rlm_code.rlm.frameworks import FrameworkAdapterRegistry

registry = FrameworkAdapterRegistry.default(workdir="/tmp/work")
adapter = registry.get("pydantic-ai")

Next Steps