Replicas

Engineering teams want to delegate coding work to AI agents like Claude Code, Codex, Cursor, and Opencode, but running these agents reliably means giving them a full environment with the right codebase, tooling, dependencies, and services, and doing so without disrupting existing workflows in Slack, Linear, GitHub, or GitLab. Running agents locally also makes it hard to isolate tasks, track usage, or let agents do things like drive a browser and verify their own work. Teams need a way to start agent work from wherever it already happens and get reviewable output back, whether that's a pull request, a comment, or a recording.

Replicas provisions an isolated Linux VM for every agent task, pre-configured with the user's dependencies, images, MCP servers, skills, files, and variables. Agents can be started from the Replicas dashboard, Slack, Linear issues, GitHub/GitLab comments and pull requests, or via API and automations, and they return a pull request, reply, or screen recording for review. The platform supports multiple agent harnesses (Claude Code, Codex, Cursor, Opencode) side by side, gives agents full computer use including browser control, and provides analytics that attribute every minute of usage to a source, person, harness, model, credential, and the skills or MCP servers used.