● Integrations

Works inside the tools your team already uses.

Hand an issue to the agents in your tracker, and it moves through your own statuses until its change is merged. Along the way, agents open pull requests with your repository provider, wait for your required checks, run in sandboxes you choose and ask people in the messenger they already read.

THE FACTORY RUNS YOUR PROCESS TRACKERS WORK COMES IN MESSENGERS PEOPLE DECIDE YOUR AGENT RUNS IT OVER MCP REPOSITORY PROVIDERS PULL REQUESTS GO OUT CI CHECK RESULTS COME BACK SANDBOXES AGENTS RUN HERE

01 Trackers

Work starts where your team plans it

Agents pick up issues by a label, a filter, or because someone assigned them. The issue’s status follows the work, so your tracker always shows where it is.

  • LinearAssign an issue to the agent right in Linear, or let a label pick it up. Changes arrive by webhook.
  • JiraConnects to Jira Cloud over OAuth. Rank and priorities decide the order; webhooks bring changes in.
  • GitHub IssuesIssues next to the code, picked up by a label, a filter or an assignee.
  • Azure DevOps BoardsWork items from Azure Boards; their state follows the work.
  • BeadsThe issue tracker that lives in your git repository, built for coding agents.
How work comes in →

02 Repository providers

Changes arrive as pull requests your team can read

Every change is a pull request with its review and its checks on it, like one from an engineer. The factory also reviews the pull requests your people open, with comments and a check run on the PR.

  • GitHubInstalled as a GitHub App in a few guided steps: pull requests, reviews, check runs and merges.
  • BitbucketPull requests and merges on Bitbucket Cloud.
  • GitLabMerge requests, reviews and merges on GitLab.
  • Azure DevOps ReposPull requests, reviews and merges on Azure Repos.
How reviews work →

03 CI

Your required checks decide what merges

Nothing merges past a red check. The factory reads your CI’s verdict on the exact commit that will merge. When the build on main breaks, it opens a repair task, and a green run closes it.

  • GitHub ActionsYour required jobs judge every merge.
  • Bitbucket PipelinesPipeline results gate merges on Bitbucket Cloud.
  • GitLab CIPipeline results gate merges on GitLab.
  • CircleCIYour required workflows judge every merge.
  • Azure PipelinesPipeline results gate merges on Azure DevOps.
How changes reach main →

04 Sandboxes

Run agents on your machines or in cloud sandboxes

Each step runs in an environment you define: the image, the memory, the command that verifies the work and the services it needs. Run it on your own machines in Docker, on Kubernetes or AWS, or in cloud sandboxes.

  • DockerOn your own machines: a Docker host in your network, local or reached over SSH.
  • E2BCloud sandboxes, with no machines of your own to run.
  • KubernetesSandboxes in a Kubernetes cluster you choose.
  • AWSSandboxes on AWS, sized by the environment each step names.
  • CloudflareSandboxes on Cloudflare’s network.
  • DaytonaManaged cloud sandboxes on Daytona.
Limits, access and where agents run →

05 Messengers

Decisions reach people where they already talk

When the work needs a person, the request says why it stopped, shows the evidence and recommends an answer. Your messenger tells the people who decide, and they answer there or in the web inbox. The work continues where it stopped.

  • SlackNotifications, and answers to decision requests.
  • Microsoft TeamsNotifications in your channels, and answers to decision requests.
How decisions work →

06 Coding agents

Use the coding agent you choose

Each step of your workflow names the agent that runs it: Claude Code, Codex, OpenCode or Pi. The process, with its steps, instructions, checks and limits, belongs to your team, not to the agent.

  • Claude CodeAnthropic’s coding agent.
  • CodexOpenAI’s coding agent.
  • OpenCodeThe open-source coding agent that works with many model providers.
  • PiA minimal coding agent for the terminal, built to be extended.
Agents & MCP →

07 Alerts

Any monitoring tool can start an investigation

Point any monitoring tool at a webhook, and an agent investigates the alert and writes a report. It changes no code. Any other system can start work through a webhook the same way.

  • WebhookSignatures are verified, duplicates dropped and every delivery logged. A dry run shows what a delivery would start before you switch it on.
  • SentrySentry alerts through a ready preset of the webhook.
  • DatadogDatadog monitor alerts through a ready preset.
  • PagerDutyPagerDuty incidents through a ready preset.
  • GrafanaGrafana alert rules through a ready preset.
Alerts, schedules and other triggers →

08 API

Start work with one API call, or from your own agent

One API call starts a task, and a retried call never starts it twice. MCP, the open protocol AI assistants use to call tools, lets Claude Code, Codex, Cursor or any other MCP client run the factory with the permissions of the person who connected it. Secrets never pass through it.

  • REST APICreate a task with one call and an idempotency key. Personal API keys are scoped and expire.
  • MCP serverA full set of tools: change workflows and rules, find out why work stopped and recover it, watch the queue. Clients connect over OAuth, with consent and revocation.
Run the factory from chat →

Keep your process.
Add the agents.