agent-stack

ACP protocol + omp acp agent backend. Powers AI-native editors via omp acp.

Layer	Technology	Status
Protocol	ACP (JSON-RPC 2.0 over stdio)	✅ Stable
Agent backend	omp acp — model routing, auth, MCP	✅ Live
Editor frontend	Zed via HTTP bridge	✅ Working
Legacy frontend	Terax AI (Tauri/Rust)	➡️ Archived

Architecture

Zed (or any OpenAI-compatible client)
  │  POST /v1/chat/completions (HTTP/SSE)
  ▼
omp-acp-bridge    ← this repo
  │  ACP (stdin/stdout JSON-RPC)
  ▼
omp acp           ← agent backend
  │  model routing, auth, tool execution
  ▼
Model API (DeepSeek, Anthropic, local, etc.)

Quick Start

1. Install omp

npm install -g @oh-my-pi/pi-coding-agent
# or: bun install -g @oh-my-pi/pi-coding-agent

2. Configure models

Create ~/.omp/agent/models.yml:

modelRoles:
  default: opencode-go/deepseek-v4-pro
  smol: opencode-zen/deepseek-v4-flash-free
  slow: opencode-go/deepseek-v4-pro
  plan: opencode-zen/deepseek-v4-flash-free
  commit: opencode-go/qwen-3.6-plus

Set your API key:

export DEEPSEEK_API_KEY=sk-...   # or whatever your provider needs

3. Start the bridge

node omp-acp-bridge/bridge.mjs
# Listening on http://127.0.0.1:7654

4. Connect Zed

In Zed settings.json:

{
  "assistant": {
    "provider": "open_ai",
    "model": "acp-agent",
    "open_ai": {
      "api_url": "http://localhost:7654/v1/chat/completions",
      "api_key": "any-value-works"
    }
  }
}

How It Works

ACP Protocol (stdin/stdout)

JSON-RPC 2.0 messages over stdin/stdout:

→ {"jsonrpc":"2.0","method":"initialize","params":{"protocolVersion":1},"id":1}
← {"jsonrpc":"2.0","id":1,"result":{"protocolVersion":1,"agentInfo":{...}}}

→ {"jsonrpc":"2.0","method":"session/new","params":{"cwd":"/","mcpServers":[]},"id":2}
← {"jsonrpc":"2.0","id":2,"result":{"sessionId":"...","availableModes":[...]}}

→ {"jsonrpc":"2.0","method":"session/prompt","params":{"sessionId":"...","prompt":[{"type":"text","text":"message"}]},"id":3}
← {"jsonrpc":"2.0","method":"session/update","params":{"sessionId":"...","update":{"sessionUpdate":"agent_message_chunk","content":{"type":"text","text":"chunk"}}}}
← {"jsonrpc":"2.0","id":3,"result":{"stopReason":"end_turn"}}

Bridge Layer

omp-acp-bridge translates OpenAI-compatible HTTP requests into ACP protocol:

OpenAI (HTTP)	ACP (stdio)
POST /v1/chat/completions	session/prompt
{"messages": [...], "stream": true}	streaming session/update events
SSE data: {...} chunks	agent_message_chunk content
data: [DONE]	stopReason: "end_turn"

Related

• omp — the agent CLI
• Zed — recommended editor frontend
• agent-ui-vscodium — VSCodium fork (legacy, archived)