cd /news/ai-agents/the-open-source-acp-orchestrator Β· home β€Ί topics β€Ί ai-agents β€Ί article
[ARTICLE Β· art-14001] src=github.com pub= topic=ai-agents verified=true sentiment=↑ positive

The open-source ACP orchestrator

AIntegriX launched as the first open-source ACP orchestrator, a server that coordinates multiple AI coding agents through a single API. The tool acts as an ACP multiplexer, enabling parallel agent execution, task chaining in pipelines, and automated Git event triggers β€” capabilities previously unavailable with independent ACP agents.

read4 min publishedMay 26, 2026

The open-source ACP orchestrator. One API. Five agents. Infinite possibilities.

AIntegriX is the first open-source ACP orchestrator β€” a server that coordinates multiple Agent Client Protocol agents through a single API.

The Agent Client Protocol (ACP) is the open standard that connects code editors with AI coding agents β€” the same way LSP standardized language servers. Every major coding agent now speaks ACP: Kiro, GitHub Copilot, Claude Code, OpenCode, Codex CLI, and 30+ more.

Each ACP agent runs independently. You can't:

  • Ask one agent to delegate work to another
  • Send the same prompt to multiple agents and compare
  • Chain agents in a pipeline (generate β†’ review β†’ fix)
  • Auto-trigger agents on Git events

AIntegriX acts as an ACP multiplexer. It spawns agents as subprocesses, speaks JSON-RPC 2.0 over stdio with each one, and exposes them all through a single MCP/REST endpoint.

Your Agent (any MCP-compatible tool)
       β”‚
       β–Ό MCP / REST
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚         AIntegriX               β”‚
β”‚   Route β€’ Orchestrate β€’ Stream  β”‚
β”‚         (ACP Client)            β”‚
β””β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”˜
   β”‚      β”‚      β”‚      β”‚      β”‚   JSON-RPC 2.0 / stdio
   β–Ό      β–Ό      β–Ό      β–Ό      β–Ό
 Kiro  Copilot OpenCode Claude Codex
 (ACP)  (ACP)   (ACP)  (ACP)  (ACP)

One prompt. Any agent. Real results.

Problem Solution
Each agent has different strengths Smart routing picks the best agent for each task
Can't compare agent responses Orchestration sends to N agents in parallel
No way to chain agent work Pipelines feed output from one agent to the next
Agents can't read your latest code Auto-clone from your Git remote or direct local filesystem
Responses arrive all at once SSE streaming shows chunks in real-time
Manual code review requests Webhooks auto-trigger review on MR open
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚                    YOUR DEVELOPMENT FLOW                      β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚                                                              β”‚
β”‚  1. Code locally                                             β”‚
β”‚       β”‚                                                      β”‚
β”‚  2. Ask AIntegriX:                                           β”‚
β”‚       "Review src/ for security issues"                      β”‚
β”‚       β”‚                                                      β”‚
β”‚  3. AIntegriX routes to Claude (security expert)             β”‚
β”‚       β”‚                                                      β”‚
β”‚  4. Claude reads your files, analyzes, responds              β”‚
β”‚       β”‚                                                      β”‚
β”‚  5. You get the review in your agent's context               β”‚
β”‚                                                              β”‚
β”‚  ─── OR ───                                                  β”‚
β”‚                                                              β”‚
β”‚  Pipeline: OpenCode generates β†’ Claude reviews β†’ Kiro fixes  β”‚
β”‚                                                              β”‚
β”‚  ─── OR ───                                                  β”‚
β”‚                                                              β”‚
β”‚  Race: Send to 3 agents, first response wins                 β”‚
β”‚                                                              β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
git clone git@github.com:alfonsodg/aintegrix.git
cd aintegrix
cargo build --release
./configs/install.sh ~/.local/share/aintegrix

{
  "mcpServers": {
    "aintegrix": {
      "command": "mcp-proxy",
      "args": ["-H", "Authorization", "Bearer aintegrix-local-key-2026",
               "--transport", "streamablehttp", "http://localhost:8050/mcp"],
      "env": {}
    }
  }
}
{
  "mcpServers": {
    "aintegrix": {
      "url": "https://your-server.example.com/mcp",
      "type": "http",
      "headers": {"Authorization": "Bearer <token>"}
    }
  }
}
Agent Model Best For
🧠 Kiro
claude-opus-4.6 Rust, architecture, complex reasoning
πŸ€– Copilot
gpt-5.3-codex Frontend, React, quick edits
⚑ OpenCode
mimo-v2.5-pro Fast analysis, multi-file reads
πŸ” Claude
minimax-2.7 Security review, deep analysis
πŸ› οΈ Codex
gpt-5.5-xhigh Python, refactoring, generation

5 ACP agents with real subprocess managementMCP serverβ€” any agent can use AIntegriX as a tool** REST APIwith Bearer auth SSE streaming**of agent responses in real-time

Parallelβ€” send to N agents, collect all responses** Race**β€” first response wins, cancel others** Jury**β€” N agents respond, a judge picks the best** Pipelines**β€” sequential multi-step workflows

Auto-routingβ€” YAML rules pick the best agent by keywords/file patterns** Prompt rewriting**β€” per-agent prefix/suffix** Git-aware sessions**β€” inject branch, commits, diff as context** Context injection**β€” auto-load steering files

Auto-clone reposβ€” fresh checkout from your Git remote (remote mode)** Local filesystem**β€” direct access to your code (local mode)** Webhook triggers**β€” MR/PR β†’ auto code review** Session fork**β€” try same conversation with different agent** Cost tracking**β€” usage per agent/session/model** Live status**β€” agent idle/busy with session counts

Method Endpoint Description
POST /api/v1/sessions
Create session
POST /api/v1/sessions/{id}/prompt
Send prompt
POST /api/v1/sessions/{id}/fork
Fork to another agent
POST /api/v1/orchestrate
Multi-agent (parallel/race/jury)
POST /api/v1/pipelines
Sequential agent chaining
POST /api/v1/stream
Create + stream SSE
POST /api/v1/webhooks/git
Receive webhook events
GET /api/v1/agents/status
Live agent status
GET /api/v1/usage
Cost tracking
POST /mcp
MCP JSON-RPC endpoint
Layer Technology
Language Rust (latest stable)
Runtime tokio
HTTP axum
Protocol ACP (JSON-RPC 2.0 over stdio)
Database SQLite (sqlx)
Config YAML

Apache-2.0

Built with πŸ¦€ Rust β€’ Powered by ACP

── more in #ai-agents 4 stories Β· sorted by recency
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain β€” perfect for shipping the agent you just read about.

$git push zahid main
β†’ Live at https://your-agent.zahid.host βœ“
Get free account β†’ Pricing
from €0/mo Β· no card required
LIVE [news/the-open-source-acp-…] indexed:0 read:4min 2026-05-26 Β· β€”