Governed AI agents for engineering teams.

Cody runs on your infrastructure, for the tasks you define, under the rules your team sets.

One day of Cody workflows for a wealth management team. They returned a dependency bump PR for the portfolio service, a review that requested changes because a transfer retry breaks rule PAY-014, a statement export incident escalated with its cause, an end-to-end transfer test run that passed and a PR that removes an unused metric tag.

Works with the tools you already use

Codexagent
Claude Codeagent
Cursoragent
GitHubsource
Microsoft Teamschat
Jirawork
Datadogsignals
Snyksecurity

Use cases

Work you can hand to Cody.

From pull request review to incident triage, each workflow starts where the work arrives and reports back in the same place.

Request access →

PR review

Cody reviews each pull request against your team's rules and comments on the lines that break them.

Idea to PR

Cody turns a request into a PRD, a spec and a pull request your team approves.

Maintenance

On a schedule, Cody opens small pull requests for the chores nobody plans.

Support and incidents

Cody investigates failures with read-only access and hands on-call the cause.

Cost

Cody finds costly metric tags that nothing reads and opens a pull request to remove them.

End-to-end testing

Cody drives your app in a real browser and reports the result on each pull request and in a nightly run.

Your use case

Bring the work your team repeats

The use cases above are starting points. Any recurring engineering job, such as release notes, schema migrations or on-call handover, can become a Cody workflow, kept in your repo and reviewed like code.

Request access →

Multiplayer

One session your team can share.

Start a Cody session in your chat, your source control or the terminal, and let teammates with access join from theirs. Every question and answer stays in the same session.

Request access →
  • Start where the work is

    A pull request marked ready for review, a chat mention or a terminal command starts the session, with the code and access attached.

  • Join with the full history

    Teammates join from Slack or Microsoft Teams, from GitHub, GitLab or other source control, or from the terminal or web app, and see every question and finding so far. Cody replies where each question came from.

  • Access follows your permissions

    Only people with access to the repository or project can join. Sessions follow your repository and workspace permissions.

Enterprise

Agentic engineering built for the enterprise.

Each workflow names what it can read, what it may change, where it runs and when a person reviews its work.

Request access →

Access set per workflow

Each workflow lists the repositories, tools and actions it may use, and nothing else.

One sandbox per job

Each job runs in its own isolated session, apart from every other job.

Changes wait for review

Investigations start read-only, and riskier actions wait for a person to approve them.

Audit log for every run

Each run keeps its trigger, tool and model calls and checks for review.

Deployment options

Private access stays private.

Private gateways stay in your network and connect to Cody outbound over mTLS, with no inbound path. Agent sessions run in your cloud or with Cody.

Managed BYOC

Agent sessions run in your cloud.

Sandboxed agent sessions, private gateways and private systems stay in your cloud. Cody operates the platform, and your cloud connects to it outbound only.

Fully managed

Agent sessions run with Cody.

Cody runs the sandboxed agent sessions. Private gateways and private systems stay in your cloud and connect to Cody outbound only.

How it works

Every job runs the same way.

Each workflow is a set of files in your Git repository. Cody reads the files, starts a sandbox in your cloud and runs the agent and model they name.

Request access →
BUG TICKET TO FIX PR WORKFLOW
HARNESSCodex0.150.1
MODELGLM5.3OpenRouter
A Cody run assembled from a chosen harness and model, a sandbox in your infrastructure, workflow files in Git, and past runs replayed as private evals before a model change.

Frequently asked questions

Private beta

Bring one job to the private beta.

Cody is available to a limited number of engineering teams. We start with one recurring job and run it on your repositories, tools and rules.

  • Choose the first job

    A recurring task with a clear trigger and an owner who reviews the work.

  • Set access and checks

    Decide which repositories and tools Cody can use and which checks must pass.

  • Review the first runs

    Your team reviews the first runs and reports, then decides what to add next.

See Cody on your first job

We walk through how Cody would run it, what it would access and what it would report.

Request access