Zed Launches Delta: Multiplayer Coding Built for AI Agents

Zed Industries launched Delta, a standalone multiplayer coding environment, into private beta on August 12, featuring DeltaDB, a CRDT-based sync engine that captures every edit operation between commits and ties each to the agent conversation that produced it. Delta uses the Agent Client Protocol (ACP) to connect with AI agents including Claude Code, Codex CLI, Gemini CLI, Cursor, StackPack, and goose, and is designed as a Git companion, not a replacement. The product addresses the problem that Git was designed for humans exchanging snapshots asynchronously, not for AI agents, and aims to preserve the reasoning and context behind code changes.

Zed Industries shipped Delta https://zed.dev/blog/introducing-delta into private beta on August 12 — a standalone multiplayer coding environment built around one uncomfortable truth: Git was designed for humans exchanging snapshots asynchronously, not for AI agents that can rewrite an entire function in 30 seconds. Delta’s answer is DeltaDB https://zed.dev/blog/introducing-deltadb , a CRDT-based sync engine that captures every edit operation between commits and ties each one permanently to the agent conversation that produced it. The Problem: Your Agent’s Reasoning Disappears at Commit Time Anyone using AI coding agents daily knows the frustration. The agent makes a decision — it chose one approach over another, it abandoned a path halfway through, it left a comment you don’t understand. You accept the diff and commit. That context is gone. What remains is code without memory. Git was never designed to track this. It captures snapshots at commit points. The reasoning, the failed attempts, the mid-session course corrections that shaped the final output — none of that survives. Zed’s framing cuts straight to the problem: “Software is made between commits.” Delta is the attempt to capture that in-between space. How DeltaDB Works The technical foundation is CRDTs — Conflict-free Replicated Data Types — a proven approach to real-time concurrent editing Figma applies similar technology to its multiplayer canvas . DeltaDB applies this to code, giving every individual edit operation a stable, addressable identity rather than batching changes into commits. The practical result: any line of code can be traced back to the conversation that produced it, and every conversation that has touched it since. When a new agent picks up the project, it can ask a prior agent why a function is written a certain way and get actual context — not a reconstruction from diffs. One detail worth noting: comments in Delta anchor to code logic, not line numbers. If the code moves, the comments follow. GitHub PR comments go stale the moment lines shift. Delta’s do not. Which Agents Connect ACP Is the Bridge Delta uses the Agent Client Protocol https://github.com/agentclientprotocol/agent-client-protocol ACP , an open JSON-RPC standard Zed introduced in August 2025, now supported by JetBrains, Neovim, and others. Terminal Claude Code sessions sync live into Delta threads via ACP. The current agent compatibility list: Claude Code, Codex CLI, Gemini CLI, Cursor, StackPack, and goose. Delta is agent-agnostic by design. The protocol is open, the compatible tool list is growing, and Delta’s Claude Runner can keep agents working after you close your laptop, syncing results back when you return. What Delta Is Not Delta does not replace Git. Commits and pushes continue exactly as before. DeltaDB operates in the layer between commits — capturing what Git was never designed to track. It is a Git companion for the agentic era, not a replacement. The browser version compiled to WebAssembly/WebGL rather than JavaScript delivers the full desktop experience without stripping features. The Honest Take Hacker News reception was predictably mixed https://news.ycombinator.com/item?id=49276574 . The leading concern is privacy: developers are accustomed to pre-commit exploratory work — dead ends, half-formed ideas — staying local. Delta syncs that work, and not everyone wants it serialized and shared with teammates or a cloud backend. The second concern is whether real teams actually need real-time multiplayer in their editor; many see Git as sufficient collaboration infrastructure. Both concerns are legitimate. But the use cases where Delta earns its keep are specific and real: AI code provenance tracking, multi-agent handoffs where context must survive the transition, and knowledge transfer — letting a junior developer review exactly what an agent did and why it made the choices it made. Git does not solve these problems. Delta might. Availability Private beta invites went out August 12. Join the waitlist at zed.dev/deltadb https://zed.dev/deltadb . Pricing has not been announced. Once invited, the browser version is accessible at delta.dev.