Warp

Warp is infrastructure for running a software factory defined in code: agents wired to your repositories, secrets and work sources, with evals, benchmarks and self-improvement loops built in. Open at every layer, from agent choice to model to compute.

Warp is an open infrastructure layer for building what it calls a software factory: a configured system of coding agents that take work from chat, tickets and source control, execute it against your repositories, and report status back to the places your team already watches. The core idea is that the factory is a written specification rather than a product you click through. A single YAML definition lists repositories, encrypted secrets, MCP server connections, cloud providers, integrations such as Slack and Linear or Jira, and the default model and runner that agents inherit. Specialized agents can override those defaults when a task calls for it. The practical loop starts with a work item in Slack or Linear, routed to the factory by an alias handle. Agents check out the relevant code on an execution host, use injected credentials and connected tools, and carry out the change. Runs can be observed and steered from web, mobile, terminal or IDE, so an engineer can watch progress and step in before anything ships. Every factory agent captures a screenshot or video of its work, giving reviewers evidence to inspect before approving a pull request. Cost per pull request is tracked alongside the work, and the homepage cites one team cutting that figure from $80 to $30. Where Warp differs from a single AI teammate is the measurement layer. Each run is scored by evals on your own tasks, benchmarks compare models against each other on your engineering work, and observer agents open pull requests against the factory configuration itself when a change would improve results. Over time the factory accumulates memory and tuned prompts. The stack is deliberately open at each layer: any MCP-capable coding agent, frontier or open-weight models chosen per pipeline stage, Warp cloud or self-hosting in your own VPC, and pluggable storage for whatever the factory produces. This is aimed at engineering organizations, not individual hobbyists. The tooling assumes familiarity with repositories, CI concepts and YAML. Enterprise controls cover guardrails, centrally configured agent access and centrally managed permissions. Warp reports more than 800,000 developers using it, with published case studies from Docker and Rectangle Health.