Zag is the hackable source control platform for merging thousands of daily agent-driven code changes.
Zag is the hackable source control platform for merging thousands of daily agent-driven code changes. Composablechangesandsessionsenable stacked code review, presubmits from linting to agentic pentesting, and postsubmits from deploying to monitoring agents.
Split the /api/emails endpoint into a fast DB-only path and a separate /api/emails/metrics endpoint for AgentMail data. The client fetches both in parallel, rendering instantly.
Change-Id: Iec120c49f48dff48506e58cb73bb6ae26a6a6964
GitHub was designed to share open-source projects. Engineering teams adopted it because it was there and it worked well enough. But coding agents work fast, produce many small changes, and operate in parallel.
Zag is built by former Google engineers and inspired by Google’s system for managing its massive internal monorepo.
In Zag, the unit of work is a change, collapsing branches, commits, and pull requests into one primitive. A change is a single, reviewable object with a stable identity. Changes stack on top of other changes. They rebase automatically when a parent changes, can be amended without losing their history, and reviewed individually or as a stack. An agent that decomposes a task into five changes creates five reviewable units of work, each with a clear description of what it does and why.
With changes and sessions as the foundation, Zag exposes everything as TypeScript primitives so deploying a Codex code review agent or ramping a PostHog feature flag is a simple defineTask. Tasks can be triggered as change presubmits or change postsubmits, and they can be simple TypeScript functions or an intelligent agent that decides which user cohorts to show a feature to for best driving conversion.
Zag doesn’t ship a CLI or an MCP. Claude Code and Codex are already great at writing code, and the Zag SDK is all they need, even for one-off commands. When everything is TypeScript, building internal coding agents for your team becomes second nature and incredibly powerful.
Every session in Zag produces a durable, replayable stream. Agents subscribe to streams for anything from fixing lint errors before CI finishes to coordinating work across multiple sessions.
All presubmits passed
Zag is generally available and self-serve, but our team is always ecstatic to help you get set up. Find a time here.