Replicas
Replicas · Replicas: Cloud Coding Agent Platform for Claude Code, Codex, Cursor, and More
Replicas is a cloud platform that runs coding agents such as Claude Code, Codex, Cursor, Muse Code, OpenCode, and Pi in isolated Linux VMs connected to your repositories. Engineers assign work from Slack, Linear, GitHub, GitLab, the web or mobile app, or the API, and get back pull requests that the agent keeps updating when CI fails or reviewers comment. It suits engineering teams that want background agents without building their own agent infrastructure.
Best for
Engineering teams already using Claude Code, Codex, or Cursor who want those agents running in configured cloud VMs, triggered from Slack, Linear, GitHub, GitLab, or automations, with PRs that follow up on CI failures and reviews
Not ideal for
Solo developers happy running agents locally, teams without existing agent subscriptions or API keys, organizations that need a free tier, and teams that need self-hosting without an Enterprise contract
Who it's for
Engineering teams and platform leads at startups and larger companies who want to delegate backlog, review, and maintenance work to cloud coding agents
Capabilities
- Runs Claude Code, Codex, Cursor, Muse Code, OpenCode, or Pi per task, and chat forking moves a conversation to another harness without losing context
- Each task gets an isolated Linux VM with the repository, dependencies, services, on-demand Docker, and a full desktop and browser for computer use, screenshots, and recordings
- Start work from Slack, Linear, GitHub and GitLab mentions, the web dashboard, the macOS and Linux desktop app, the iOS app, the CLI, MCP clients, or the REST API
- CI failure response: when a check fails on a linked pull request, the agent reads the logs, pushes a fix, and comments, in a separate fixes chat
- Automations triggered by schedules, GitHub, GitLab, Slack, or Sentry events, or custom webhooks, with debounce, workspace size, and lifecycle settings
- Environments with variables, files, skills, MCP servers, and plugins, plus warm hooks and warm pools that keep pre-built workspaces ready
- Bring-your-own credentials: Claude or OpenAI OAuth, Anthropic or OpenAI API keys, AWS Bedrock, Microsoft Foundry, a Cursor API key, OpenRouter, OpenCode Go, or Ollama, at organization or personal scope
- Plan mode, Goal mode for Claude Code (keeps working until a stated condition holds), and Fast mode
- Analytics that break down compute minutes by workspace source and seat, rank agent runtime by harness, model, and credential, and count skill and MCP calls
- Admin controls: security policies that can block agent merges, an audit log, configurable chat history retention, a static egress IP, and short-lived workspace identity tokens for cloud and API access
- Cloud iOS and Android devices for building and testing mobile apps from a workspace
Limitations
- Model usage is not included: you need your own coding agent subscriptions or API keys, and at least one coding agent must be configured before creating workspaces
- No free plan: there is a 14-day trial without a card, after which workspaces cannot start without a subscription
- On Developer and Team, API and automation runtime is billed per minute on top of seat prices ($0.008 per minute small, $0.016 per minute large), and Replicas' terms allow programmatic use only through the API and automations, so heavy automation raises the bill
- Developer is a single-person plan that suspends manually started workspaces after 150 hours a month
- Organization-level coding agent credentials are injected into workspaces, where members can retrieve them through the agent or terminal
- Sleeping workspaces are archived after seven days without activity, and archived workspaces are deleted after 30 more days
- Self-hosted and dedicated single-tenant deployment are Enterprise-only, and Replicas lists SOC 2 Type I with Type II in progress
- There is no Android app yet (mobile is iPhone-only), and the desktop app is documented for macOS and Linux only
Use cases
- Assigning a Linear issue to Replicas and receiving a pull request built and tested in a cloud VM
- Tagging Replicas on a failing pull request or letting CI failure response push the fix automatically
- Running a nightly automation that updates dependencies or audits code across repositories
- Firing an agent from a Sentry alert to investigate a new production error
- Starting and steering an agent from Slack or the iOS app while away from a laptop
- Comparing Claude Code, Codex, and Cursor on the same task in the same environment
Our take
Replicas makes most sense for teams that have already chosen their coding agents and now want them running in the background, with proper environments, triggers from the tools engineers live in, and pull requests that follow up on CI on their own. Because you bring the harness and the credentials, you can switch between Claude Code, Codex, and Cursor as models change, but you pay twice: once to your model providers and once for VM minutes. Model the automation volume before rolling it out widely, since API and automation minutes are billed on top of seats, and check how shared organization credentials are exposed inside workspaces.
Who should use it
Engineering teams already paying for Claude, OpenAI, or Cursor who want background agents in configured cloud VMs, platform teams standardizing environments and credentials for agents, and teams that want agent work triggered from Linear, Slack, GitHub, GitLab, or Sentry.
Who should skip it
Individual developers who are happy running agents locally, teams without existing agent subscriptions or API keys, and organizations that need a free plan, and teams that need self-hosting without an Enterprise contract.
Strengths
- Runs the coding agents your team already uses instead of a proprietary one
- Each task gets its own configured Linux VM with computer use
- Starts work from Slack, Linear, GitHub, GitLab, mobile, or the API
- Pull requests follow up on CI failures and review bots automatically
- Detailed usage analytics and admin controls
Weaknesses
- Model costs come on top, through your own subscriptions or keys
- No free plan after the 14-day trial
- Per-minute charges for API and automation runs
- Self-hosting requires an Enterprise contract
Replicas pricing
Developer
$50
Billed monthly
- One person, billed upfront
- 150 hours per month of manually started workspaces
- API and automations billed by usage
- Unlimited repositories, environments, and automations
- Warm pools and warm hooks, API access, mobile app
Team
$200 per Full seat
Billed monthly
- Unlimited manual workspace minutes on Full seats
- Flex seats pay as you go at $0.016 per minute
- API and automations billed by usage
- Higher API rate limits and a static egress IP
- Larger sandboxes (4 vCPU, 16 GB memory, 32 GB disk) and a shared Slack support channel
Enterprise
Custom
- Unlimited manual and automated workspace minutes
- Custom warm pools, API rates, and rate limits
- SOC 2, DPA, and SCIM provisioning
- Cloud, dedicated single-tenant, or self-hosted deployment
Free tier limits: No free plan. The 14-day trial needs no card and includes unlimited manually started workspaces plus 15,000 minutes of API and automation usage (large workspaces count double).
Note: Workspaces use minutes only while awake, and sleeping or archived workspaces use none. API and automation runs are billed to the organization at $0.008 per minute (small) or $0.016 per minute (large), with a one-minute minimum per session. Team needs at least one Full seat; seat type at renewal decides whether the past month's usage is billed per minute. Model usage is billed separately by the providers whose credentials you connect.
Technical specs
API pricing
Per minute of workspace runtime: $0.008/min small (2 vCPU, 8 GB), $0.016/min large (4 vCPU, 16 GB), one-minute minimum per session
Where Replicas excels
Turning Linear issues into pull requests
Assigning an issue to Replicas starts an agent in a configured VM that reads the ticket, implements the change, runs the app and tests, and opens a PR, so small backlog items get done without an engineer starting a local session.
Keeping pull requests green
CI failure response reads the failed check's logs, pushes a fix, and comments on the PR in a separate chat, which removes a common round trip between authors and CI.
Scheduled maintenance across repositories
Automations can run an agent nightly or on GitHub, Sentry, or webhook events with a fixed environment and workspace size, and each run's minutes are tracked in analytics.
Replicas vs. competitors
Replicas vs. Devin
Devin is Cognition's own autonomous software engineer: you assign it tickets from Slack, Teams, Linear, Jira, the web app, or the API, and it works in its own cloud VM. Replicas runs third-party harnesses (Claude Code, Codex, Cursor, Muse Code, OpenCode, Pi) on your credentials and charges for VM minutes. Devin fits teams that want one integrated agent product, and Replicas fits teams that want to keep their chosen agents and switch between them.
Replicas vs. Factory Droid
Factory's Droids are its own multi-model agents, used from a desktop app, CLI, IDEs, CI, and Slack, with Missions and persistent Droid Computers. Replicas does not ship an agent: it hosts other vendors' harnesses, each task in its own Linux VM with warm pools, and bills VM minutes while model costs stay on your own credentials. Factory fits teams standardizing on one agent and its plans, and Replicas fits teams that want to keep Claude Code, Codex, or Cursor and run them in a harness-agnostic cloud runtime.
Frequently asked questions
What is Replicas?
Replicas is a cloud coding agent platform. It runs agents such as Claude Code, Codex, Cursor, Muse Code, OpenCode, and Pi inside isolated cloud Linux VMs connected to your GitHub or GitLab repositories, takes tasks from Slack, Linear, GitHub, GitLab, its apps, or its API, and returns pull requests, commits, or replies for review.
How much does Replicas cost?
Developer costs $50 a month for one person with 150 hours of manually started workspaces. Team costs $200 per Full seat per month with unlimited manual usage, and Flex seats pay $0.016 per minute instead. API and automation workspaces are billed separately at $0.008 per minute (small) or $0.016 per minute (large). Enterprise pricing is custom. A 14-day trial requires no card.
Do I need my own Claude or OpenAI subscription?
Yes. Replicas charges for workspace runtime, not model usage. You connect credentials such as a Claude or OpenAI sign-in, Anthropic or OpenAI API keys, AWS Bedrock, Microsoft Foundry, a Cursor API key, OpenRouter, or Ollama, and at least one coding agent must be configured before creating workspaces.
Can Replicas fix failing CI automatically?
Yes. When a CI check fails on a pull request linked to a Replicas workspace, the agent is notified, reads the failure logs, pushes a fix, and comments on the PR. This runs in a separate fixes chat, and it wakes a sleeping workspace when needed.
Can Replicas be self-hosted?
Only on Enterprise. Replicas says Enterprise customers can run it in Replicas' cloud, in a dedicated single-tenant deployment, or self-hosted in their own infrastructure.
Replicas vs Devin: what is the difference?
Devin is Cognition's own autonomous software engineer running in its own VM. Replicas runs third-party harnesses such as Claude Code, Codex, and Cursor using your existing credentials, and bills for workspace runtime rather than agent usage.
Integrations & fit
Alternatives to consider
Cognition
Engineering teams with a steady backlog of well-scoped tickets who want to delegate them asynchronously from Slack, Teams, or their issue tracker and review the resulting PRs
Paid
Factory
Engineering organizations that want one model-agnostic agent platform across local coding, CI automation, code review, and long-running missions, with enterprise deployment and governance options
PaidAbout Replicas
Replicas does not ship its own coding model or agent. It hosts the harnesses your team already uses and gives each task its own cloud machine. A workspace is an isolated Linux VM with your repository, dependencies, and services, where the agent can install packages, run the app, use Docker, and drive a desktop and browser, then hand back screenshots or recordings. Admins configure environments once (variables, files, skills, MCP servers, plugins, and warm hooks that pre-build images), and warm pools keep ready workspaces waiting. Work can start from a Slack mention, a Linear issue, a tag on a GitHub or GitLab issue or pull request, the dashboard, the desktop app for macOS and Linux, the iOS app, the CLI, an MCP client, or the API. Automations fire agents on schedules, on GitHub, GitLab, Slack, or Sentry events, or from custom webhooks. Once a pull request is open, Replicas watches it: when a CI check fails, the agent reads the logs, pushes a fix, and comments, in a separate fixes chat so your main conversation is not interrupted, and allowlisted review bots can trigger follow-ups the same way. Teams bring their own credentials (Claude or OpenAI sign-ins, provider API keys, AWS Bedrock, Microsoft Foundry, a Cursor API key, OpenRouter, or Ollama), so model costs stay with your existing subscriptions, while Replicas charges for workspace runtime: $50 a month for one Developer, $200 per Full seat on Team with pay-as-you-go Flex seats, and per-minute rates for API and automation work. Enterprise adds SCIM, a DPA, and dedicated or self-hosted deployment. The tradeoffs are that you still need your own agent subscriptions or keys, automation volume adds per-minute charges on top of seats, shared organization credentials can be read by members from inside a workspace, and there is no free plan beyond a 14-day trial.
Updates from Replicas
Replicas released an iOS app for starting cloud coding agents, steering them mid-run, and reviewing their pull requests. It works with existing Developer, Team, and Enterprise accounts and the environments, credentials, and integrations configured on the web.
Claude Code workspaces can now work toward a stated goal with /goal, and CI failures and review-bot comments fork a separate PR fixes chat instead of interrupting the main conversation. Opus 5.5, GPT-6 Sol, and GPT-6 Luna became available in new workspaces.
Replicas moved to Full seats at $50 a month on Developer and $200 on Team, added pay-as-you-go Flex seats for manual minutes, and began billing API and automation runtime per minute on both plans.
Are you the founder? Claim this listing →