Blog

Updates, deep-dives, and announcements from the ACR project.

Introducing Agent Context Relay (ACR)

ACR is a cross-platform desktop IDE for running, managing, and chaining multiple AI coding agents. Learn about the Capture → Normalize → Compact → Relay loop that makes it possible to hand off context between agents seamlessly.

Read more →

Introducing Agent Context Relay (ACR)

July 21, 2026

What is ACR?

Agent Context Relay (ACR) is an open-source, cross-platform desktop IDE built for developers who work with multiple AI coding agents. If you use Claude Code, Codex CLI, Gemini CLI, Aider, or any other terminal-based coding agent — ACR is the workspace that ties them all together.

The problem

Modern AI-assisted development often involves running multiple agents across different branches, tasks, and even different providers. Each agent operates in isolation — there's no built-in way to hand off context from one agent to another. When you switch from Claude Code to Codex, you start from scratch: re-explain the problem, re-establish the context, re-do the discovery work.

The ACR solution

ACR introduces a vendor-neutral context relay system:

  1. Capture — Every agent session is recorded into a durable, append-only project record stored in .acr/events.jsonl.
  2. Normalize — Provider-specific transcript formats (Claude JSONL, Codex rollout JSONL, Gemini JSON) are converted into a single provider-neutral model.
  3. Compact — A handoff packet is generated containing decisions made, open questions, git state, and the exact next action.
  4. Relay — The packet is delivered to the target agent through ACR's managed launch workflow, so the next agent continues exactly where the first left off.

Beyond relay — a full terminal IDE

ACR isn't just about relay. It's a complete development environment:

  • Up to 50 concurrent terminal sessions — each in its own tab with full scroll history
  • Git worktree orchestration — auto-create, manage, and clean up worktrees per branch
  • Agent launcher — visual matrix of 169 agent × provider × model combinations
  • Session history browser — browse past sessions as rendered chat transcripts
  • Built-in AI Chat — multi-provider, with autonomous agent mode
  • Usage dashboard — real-time rate limits, 7-day charts, 52-week heatmaps
  • Plugin system — ES modules, hot reload, capability-gated runtime
  • MCP Proxy Hub — manage upstream MCP servers with OAuth
  • Voice dictation — local on-device Whisper, fully private

Current status

ACR is under active development. The project is in its M1 milestone — completing one full Claude → Codex continuity loop. Core features already implemented include:

  • Passive session capture with default-on opt-out
  • Claude and Codex transcript normalization
  • Archive-to-.acr with git provenance
  • Agent launcher with compatibility matrix (169 combos, 332 evidence edges)
  • Handoff packet generation and relay workflow
  • Session history browser with read-only chat transcripts
  • Resume-from-history and ACR-context launch mode

Pre-built binaries are coming soon. For now, you can build from source and follow development on GitHub.