fleet-rlm 0.4.9

Recursive Language Models with DSPy + Modal and an integrated Web UI for secure long-context code execution

fleet-rlm

Secure, cloud-sandboxed Recursive Language Models (RLM) with DSPy and Modal.

fleet-rlm gives AI agents a secure cloud sandbox for long-context code and document work, with a Web UI-first experience, recursive delegation, and DSPy-aligned tooling.

Paper | Docs | Contributing

---

Quick Start (Web UI First)

Fastest path: install and launch the built-in Web UI.

# Install as a runnable CLI tool
uv tool install fleet-rlm

# Launch the Web UI server
fleet web

Open http://localhost:8000 in your browser.

• Prefer a regular environment install instead of uv tool?

uv pip install fleet-rlm
fleet web

• fleet web is the primary interactive interface.
• Plain fleet-rlm installs are intended to support fleet web.
• Runtime settings (LM / Modal) can be configured from the Web UI Settings surface in local development.
• Full setup for Modal secrets, Neon DB, auth modes, and deployment is linked below.

Why fleet-rlm

• Chat with an RLM-powered agent in the browser (fleet web)
• Run recursive long-context tasks with a secure Modal sandbox
• Analyze documents (including PDF ingestion with MarkItDown/pypdf fallback)
• Stream execution events and trajectories for observability/debugging
• Expose capabilities as an MCP server (fleet-rlm serve-mcp)

Other Ways to Run It

Common commands:

# Standalone terminal chat
fleet-rlm chat --trace-mode compact

# Explicit API server
fleet-rlm serve-api --port 8000

# MCP server
fleet-rlm serve-mcp --transport stdio

# Scaffold assets for Claude Code
fleet-rlm init --list

Terminal chat surfaces

• fleet starts the standalone interactive chat launcher (Ink runtime path).
• fleet-rlm chat starts the in-process terminal chat.
• OpenTUI workflows and setup are documented in the guides (see links below) because they require additional local tooling.

Running From Source (Contributors)

# from repo root
uv sync --extra dev --extra server
uv run fleet web

Frontend build workflow (when validating packaged Web UI assets):

# from repo root
cd src/frontend
bun install --frozen-lockfile
bun run build
cd ../..

Use the full contributor setup (frontend builds, env/bootstrap, quality gates) in AGENTS.md and docs/contributing.md.

Architecture Overview

Read this after the quick start if you want the full system picture (entry points, ReAct orchestration, tools, Modal execution, persistent storage).

graph TB
    subgraph entry ["🚪 Entry Points"]
        CLI["CLI (Typer)"]
        WebUI["Web UI<br/>(React SPA)"]
        API["FastAPI<br/>(WS/REST)"]
        TUI["Ink TUI<br/>(standalone runtime)"]
        MCP["MCP Server"]
    end

    subgraph orchestration ["🧠 Orchestration Layer"]
        Agent["RLMReActChatAgent<br/>(dspy.Module)"]
        History["Chat History"]
        Memory["Core Memory<br/>(Persona/Human/Scratchpad)"]
        DocCache["Document Cache"]
    end

    subgraph tools ["🔧 ReAct Tools"]
        DocTools["📄 load_document<br/>read_file_slice<br/>chunk_by_*"]
        RecursiveTools["🔄 rlm_query<br/>llm_query<br/>(recursive delegation)"]
        ExecTools["⚡ execute_code<br/>edit_file<br/>search_code"]
    end

    subgraph execution ["⚙️ Execution Layer"]
        Interpreter["ModalInterpreter<br/>(JSON protocol)"]
        Profiles["Execution Profiles:<br/>ROOT | DELEGATE | MAINTENANCE"]
    end

    subgraph cloud ["☁️ Modal Cloud"]
        Sandbox["Sandbox Driver<br/>(Python REPL)"]
        Volume[("💾 Persistent Volume<br/>/data/<br/>• workspaces<br/>• artifacts<br/>• memory<br/>• session state")]
    end

    WebUI -->|"REST / WS"| API
    CLI --> Agent
    API --> Agent
    TUI --> Agent
    MCP --> Agent

    Agent --> History
    Agent --> Memory
    Agent --> DocCache

    Agent --> DocTools
    Agent --> RecursiveTools
    Agent --> ExecTools

    DocTools --> Interpreter
    RecursiveTools --> Interpreter
    ExecTools --> Interpreter

    Interpreter --> Profiles
    Interpreter -->|"stdin/stdout<br/>JSON commands"| Sandbox
    Sandbox -->|"read/write"| Volume

    style entry fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
    style orchestration fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
    style tools fill:#fff3e0,stroke:#f57c00,stroke-width:2px
    style execution fill:#e8f5e9,stroke:#388e3c,stroke-width:2px
    style cloud fill:#fce4ec,stroke:#c2185b,stroke-width:2px

Docs and Guides

• Documentation index
• Explanation index
• Quick install + setup
• Configure Modal
• Runtime settings (LM/Modal diagnostics)
• Deploying the server
• Using the MCP server
• CLI reference
• HTTP API reference
• Auth modes
• Database architecture
• Source layout

Advanced Features (Docs-First)

fleet-rlm also supports runtime diagnostics endpoints, WebSocket execution streams (/api/v1/ws/execution), multi-tenant Neon-backed persistence, and opt-in PostHog LLM analytics. Those workflows are documented in the guides/reference docs rather than front-loaded here.

Contributing

Contributions are welcome. Start with CONTRIBUTING.md, then use AGENTS.md for repo-specific commands and quality gates.

License

MIT License — see LICENSE.

Based on Recursive Language Modeling research by Alex L. Zhang (MIT CSAIL), Omar Khattab (Stanford), and Tim Kraska (MIT).