pentos-bridge 0.1.0

Local MCP server — detect local AI models, route smart, one interface for everything

Pentos Bridge

Local-first AI agent bridge with decentralized cloud fallback.

Pentos Bridge detects local AI models (Ollama, llama.cpp, vLLM) and routes inference requests locally when possible, falling back to Strue's decentralized cloud when needed.

Install

pip install pentos-bridge

Quick Start

# Check what's available locally
pentos-bridge status

# Start the bridge (OpenAI-compatible API on localhost:8787)
pentos-bridge start --key "your-strue-api-key"

Features

• Local-first routing — automatically detects Ollama and routes requests to local models
• Cloud fallback — requests for unavailable models route through Strue's decentralized inference network
• OpenAI-compatible API — drop-in replacement at http://localhost:8787/v1
• MCP server — works with Claude Desktop and other MCP clients
• Memory integration — optional EverMemOS support for persistent agent memory

Usage

As an OpenAI-compatible API

pentos-bridge start --key "sk_..."

curl http://localhost:8787/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma3:4b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

As an MCP server (Claude Desktop)

pentos-bridge mcp

Requirements

• Python 3.10+
• Ollama (optional, for local inference)

Links

• Strue — Decentralized agent infrastructure
• Arloop — The autoresearch library