ACPACP AI Research
⌕ZH
ResourceHeat 84Quality 92

acp-bridge: Open-Source Air-Gapped Local AI Coding Adapter for ACP

acp-bridge is an open-source Rust binary providing a zero-cloud bridge for air-gapped AI coding environments using the Agent Client Protocol (ACP). Compiled into a single 5MB executable, it enables seamless offline integration between ACP-compatible editors and local inference engines like Ollama, llama.cpp, and vLLM.

ACPOpen SourceAir-Gapped AIRustOllama

Resource Overview

Deploying AI coding assistants in strict enterprise compliance and air-gapped networks often presents significant hurdles due to cloud dependencies. acp-bridge is an open-source Rust-based adapter designed to provide a zero-cloud, zero-dependency connection for local AI coding setups via the Agent Client Protocol (ACP) [1].

By distilling complex agent communications into a self-contained local adapter, acp-bridge allows developers to leverage AI-assisted development entirely offline without requiring internet access or cloud API credentials [1].

Key Content

Key technical features and developments surrounding acp-bridge include:

  • Single Binary Footprint: Compiled into a lightweight ~5MB binary with zero runtime dependencies, fulfilling strict offline enterprise compliance requirements [1].
  • Core ACP Implementation: Version 0.7.8 implements standard ACP server JSON-RPC methods over stdin/stdout, including initialize, session/new, session/prompt, and session/end [1].
  • Native Ollama Integration: Features auto-detection of local Ollama instances to handle /api/chat with NDJSON streaming, model parameter queries, and VRAM status monitoring [1].
  • Data Leakage & Bug Fixes: Resolves offline tool-calling bugs and network leakage issues previously observed when pairing cloud-oriented agents with local inference engines like llama.cpp and vLLM [1].
  • ACP Ecosystem Growth: The Agent Client Protocol organization released version 1.0.0 of its Rust and TypeScript SDKs alongside the draft release of ACP v2 in mid-2026 [2]. Key updates include stabilized support for session/resume, request cancellation via $/cancel_request, boolean configurations, and structured user elicitation [2].

How to Use

Setting up acp-bridge involves running the binary as a bridge between local inference engines and ACP-supported code editors:

  1. Launch Local Engine: Start a local inference server such as Ollama, llama.cpp, or vLLM.
  2. Configure Editor Transport: Direct your ACP-compatible editor (such as Zed, Neovim, or VS Code) to invoke the acp-bridge executable using standard stdin/stdout piping.
  3. Automated Feature Detection: The editor automatically detects local agent capabilities and begins handling inline completions, code prompts, and tool calls offline [1].

Notes and Caveats

Consider the following operational constraints when deploying acp-bridge:

  • Specification Coverage: Version 0.7.8 supports a primary subset of the ACP specification; full spec alignment (including session/resume and session/set_mode) is targeted for late Q3 2026 [1].
  • Model Template Tuning: Thinking-capable local model templates (such as DeepSeek-R1 and Qwen3) may require customized tool-calling template formatting during local inference [1].
  • Transport Maturity: While local stdin/stdout JSON-RPC is fully stabilized, remote transports over Streamable HTTP and WebSocket remain active draft specifications [1, 2].

Sources

  1. acp-bridge: ACP (Agent Client Protocol) adapter for self-hosted AI - GitHub (2026-07-28)
  2. Agent Client Protocol (ACP) Project Updates and Announcements - Agent Client Protocol Org (2026-07-22)