Shabrang
shipped

SOS — Sovereign Operating System

MCP-native AI agent OS. Redis bus, squad service, multi-model engine, and shared memory — all open source.

The kernel — coordinates all agents, routes tasks, exposes MCP tools, runs the bus

GitHub
pythonredismcpfastapipgvector

SOS is the operating system for AI agent teams. It gives any MCP-compatible agent (Claude Code, Codex, custom bots) a shared bus, persistent memory, task queues, and multi-model routing — running on a single server.

What It Does

  • Agent Bus — Redis-backed event stream. Agents send, receive, and wake on messages.
  • MCP SSE Server — Standard MCP over SSE on :6070. Plug in any agent with one URL.
  • Squad Service — Task queue with atomic claim, skill matching, and pipeline orchestration.
  • Multi-Model Engine — Route to Gemini, Claude, GPT, Grok, or Ollama with failover.
  • Wake Daemon — Delivers bus messages to sleeping agents automatically.

MCP Tools (what agents get)

ToolWhat it does
send / inbox / broadcastAgent-to-agent messaging
remember / recallSemantic memory via Mirror
search_codeSearch code nodes across all repos
task_create / task_list / task_updateTask management
peers / statusSee who’s online
onboardSelf-register a new agent

Architecture

Agent → MCP SSE (:6070) → SOS → Mirror (:8844) → PostgreSQL + pgvector
                              → Redis (bus + events)
                              → Squad Service (:8060)
                              → Multi-Model Engine (:6060)

Get Started

git clone https://github.com/Mumega-com/sos
cd sos
cp .env.example .env      # fill in Redis + Mirror URL
pip install -r requirements.txt
python3 -m sos.mcp.sos_mcp_sse    # MCP bus on :6070
python3 -m sos.services.engine    # Engine on :6060
python3 -m sos.services.squad.app # Squad service on :8060

Connect Claude Code:

{ "mcpServers": { "sos": { "type": "sse", "url": "http://your-server:6070/sse/your-token" } } }

Works best paired with Mirror for persistent agent memory.