alh477/dcf-agent repository overview

⁠dcf-agent

LangGraph multi-agent LLM system for the HydraMesh DCF framework.

⁠What this is

A Docker image that runs LLM-powered agents communicating over the HydraMesh DCF mesh in realtime using the Model Context Protocol (MCP). Built with LangGraph state graphs, pluggable LLM backends, and a universal API system.

Encryption-free for export control purposes — agents communicate over the same plaintext DCF transport. No cryptographic layers bundled.

⁠Quick start

# Start the HTTP API server (default)
docker run -p 8000:8000 alh477/dcf-agent

# List available LLM backends
docker run alh477/dcf-agent backends

# One-shot chat with echo backend
docker run alh477/dcf-agent chat "hello mesh" --backend echo

# Interactive TUI
docker run -it alh477/dcf-agent tui

# MCP server (stdio transport)
docker run -i alh477/dcf-agent mcp

⁠Subcommands

⁠HTTP API endpoints

⁠Chat API example

curl -X POST http://localhost:8000/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "hello", "backend": "echo", "graph": "base"}'

⁠LLM backends

⁠Universal provider presets

Any OpenAI-compatible API can be registered at runtime via config:

Pass API keys via environment variables:

docker run -p 8000:8000 -e FIREWORKS_API_KEY=fw_... alh477/dcf-agent

⁠MCP tools

Any MCP client (Hermes, Claude, etc.) can connect to the stdio MCP server:

⁠Docker Compose

services:
  dcf-agent:
    image: alh477/dcf-agent:latest
    ports:
      - "8000:8000"
    environment:
      - FIREWORKS_API_KEY=fw_...
      # - OPENAI_API_KEY=sk-...
      # - XAI_API_KEY=...

⁠Architecture

mesh_recv → receive → route → process (LLM backend) → respond → mesh_send
                                                      ↑
                                         coordinator routes by prefix:
                                         echo:       → echo specialist
                                         summarize:  → summarizer specialist
                                         (default)   → echo

• LangGraph state graphs: receive → route → process → respond
• Pluggable LLM backends: echo, grok, glm5p2, or any OpenAI-compatible API
• MCP tool wrappers: mesh_send, mesh_recv, mesh_status as LangChain tools
• Streaming bridge: UTF-8-safe chunking for DCF-Text frame limits
• Coordinator graph: prefix-based routing to specialist subgraphs

⁠Export compliance

Encryption-free by design. Safe for EAR/ITAR-regulated environments where cryptographic layers are restricted. Any security layer is external and connected through IPC barriers.

⁠Source

• Repo: https://github.com/ALH477/HydraMesh⁠
• Agent source: langgraph_agents/ directory
• License: LGPL-3.0-only
• Built with: Nix + DockerTools (hermetic, reproducible)

⁠Related images

• alh477/hydramesh⁠ — Lisp SDK node with agent DSL integration
• alh477/dcf-go⁠ — Go mesh node
• alh477/dcf-rs⁠ — Rust mesh node
• alh477/dcf-python⁠ — Python mesh node
• alh477/dcf-c⁠ — C mesh node

Developed by DeMoD LLC⁠