Replica.dev

Replicas lets you run coding agents inside isolated cloud workspaces that already match your repo setup. It’s for engineering teams that want agents to implement, test, and produce reviewable results without moving their codebase or tooling into a new workflow.

You can hand work to agents from where your team tracks it, then review what comes back.

• Slack + Linear + GitHub/GitLab starts: kick off agent runs from team tools, tickets, or code comments/PRs.
• Multiple coding harnesses: run Claude Code, Codex, Cursor, or Opencode, each in its own cloud Linux VM.
• Review outputs: get changes as a pull request (or other reviewable artifacts like replies, commit, comment, or a recorded session).
• Automations + API: start runs on a schedule, via webhook, or with a single API call.
• Sandboxed environments per task: each task runs in an isolated VM with your repository, tools, dependencies, and services available so tests and verification happen in the same kind of environment.
• Attribution and activity analytics: see minute-by-minute attribution tied to source, person, harness, model, credentials, and the skills/MCP servers the agent used.

What stands out is the combination of cloud execution (separate Linux VMs per task) and work placement. The agents operate on your enabled GitHub or GitLab repositories with your prepared environment, then you review and merge like you would any other change. If you care about knowing what happened during agent work and linking it back to who requested it and what it used, the attribution data is built for that.