Skip to content →
  1. Agents and automation
  2. Linear cloud coding agent
Sign upOpen app

Linear cloud coding agent

Overview⁠

Linear’s cloud coding agent turns issues into draft pull requests using Claude Code or Codex. Delegate a well-scoped issue to Linear, and the agent writes and tests the change in a secure cloud environment, starting with the context already in your workspace. Your team can delegate work without setting up a local development environment.

Teammates with repository access can follow progress, add context, steer the approach, and review the diff in Linear. Existing GitHub permissions and review requirements still determine what ships.

Use the coding agent to:

  • Fix well-scoped bugs as they enter triage using Loops
  • Make contained changes with clear outcomes and constraints
  • Address review comments, rebase branches, fix lint issues, or add tests
  • Delegate changes without setting up a local environment

Linear cloud coding agent is available on Free, Basic, Business, and Enterprise plans. AI credits are required. Legacy plans are not supported. Usage draws from your workspace’s AI credits. See AI credits for more information.

To review spending and manage spend limits, open Settings → Usage & limits. Learn more about usage and spend limits.

Setup⁠

A Linear owner or admin with GitHub organization owner access can enable the coding agent for your workspace:

  1. Navigating to your Settings > Features section > AI & agents page
  2. Under “Linear Agent” select Cloud coding agent
  3. Click the drop-down and select Enabled. On Enterprise plans, select Enabled for teams… to restrict access to specific teams.

If your GitHub organization or services accessed by the agent use IP allow lists, allow the cloud coding agent IPs listed under GitHub Enterprise Cloud in the GitHub integration documentation.

Once enabled, any workspace member with a GitHub account linked in Connected Accounts can delegate coding work to Linear.

Choose a model⁠

Our current default model (tracked under Auto) is GPT-6 Sol. The model selected here is used for all subsequent coding work in the workspace.

Workspace owners or admins can update the model selection from Cloud coding agent settings. We currently support Claude Opus 5.5, Claude Fable 5, Claude Opus 5, Claude Opus 4.8, and Claude Sonnet 5; GPT-6 Sol, GPT-6 Luna, GPT-5.6 Sol, GPT-5.5, and GPT-5.4.

Bring your own key (BYOK)⁠

Coding sessions use Linear credits for model usage by default. To bill model usage to your own provider account instead, connect a provider API key, often called bring your own key (BYOK), or an eligible Anthropic organization.

Your team can continue delegating work, following progress, and reviewing code in Linear as usual.

An admin configures the connection once for everyone in the workspace. Members don’t need to enter their own API keys or sign in to the provider, and repository access and review requirements stay the same.

BYOK is available on Free, Basic, Business, and Enterprise plans. Legacy plans aren’t supported.

Your provider bills you directly for model usage. Linear still bills for the cloud environment where the agent runs your code, so your workspace needs AI credits to run coding sessions even when using your own API key.

See sandbox runtime pricing for rates by sandbox size.

Set up your workspace⁠

A workspace owner or admin with proper permissions can configure inference from Settings > Linear Code. Find the Inference section below Agent configuration.

The Anthropic and OpenAI rows control the connections for Codex and Claude Code, respectively. Configure either or both, or leave a row set to Linear credits to have Linear handle model billing. Note that each connection is shared across the workspace.

Use an Anthropic API key to bill Claude Code model usage to your own provider account.

In the Anthropic row under Inference, select API key. Choose Anthropic in the Provider dropdown, enter your API key, then select Save.

To use Amazon Bedrock, choose Amazon Bedrock and enter your AWS region and API key.

For another service, choose Custom endpoint and enter its Endpoint URL and API key. The endpoint must support the Anthropic Messages API.

Use an OpenAI API key to bill model usage to your OpenAI account.

In the OpenAI row, select OpenAI API key to open Add OpenAI API key. Choose OpenAI in the Provider dropdown, enter your API key, then select Save.

To use another service instead of OpenAI, choose Custom endpoint in the same dialog. Enter the service’s Endpoint URL and API key. The endpoint must support the OpenAI Responses API.

Select Save to validate and save your provider credentials--you can also select Test connection first if preferred.

Use and manage your connection⁠

Once setup is complete, your team can continue delegating work to Linear as usual. New coding sessions use the connection configured for their agent, including sessions started through a Loop. No additional setup is needed for individual members.

Admins can return to the same settings to replace a key or switch back to Linear-managed.

These changes apply to new sessions. Existing sessions continue using their original provider, so keep a valid key configured for that agent to resume work started with your own key.

Configure coding environments⁠

Coding environments let workspace owners and admins configure repository-specific setup for the coding agent. Use them to provide the runtimes, tools, configuration, and instructions the agent needs to work in a particular codebase.

To create an environment, go to Workspace settings → AI & Agents → Cloud coding agent → Environments.

Configure an environment⁠

Configure the tools, runtime settings, and resources your cloud coding agent needs to work in your repository.

Runtimes and tools

Node.js, Python, Ruby, Go, PostgreSQL, and Redis, including specific runtime versions.

Environment variables

Use these for required runtime configuration. Names and values are visible to the coding agent and should not be treated as agent-hidden secrets. To add values from a .env file, choose Import from .env, paste the file contents, and select Import variables. Review the imported variables, then save the environment. Environment variables are applied during setup.

Prepare script

Run setup commands during environment preparation.

Files

Add text files at relative paths in the environment home directory, such as .npmrc.

Custom guidance

Provide repository-specific instructions for how the agent should work.

Agent and model

Choose an agent and model for the environment. Available options depend on the selected model and workspace configuration.

Sandbox size

Choose the CPU and memory available to cloud coding agents. Resource changes apply to new sessions, not sessions already running. For runtime cost, see AI Credits below.

Resource changes apply to new sessions. They do not affect sessions already running. Environment variables are applied during setup.

Environment secrets⁠

Coding environments can securely store credentials for private registries and authenticated services outside your repository. This gives agents access to the private dependencies and services they need to build and test your code.

When creating a new environment variable for your environment, check the Secret checkbox to denote this state:

Environment secrets are not visible to the coding agent by default, but are visible to anyone who can run a session in that environment. Treat them as development-time secrets, like a local .env file, not as values hidden from your team.

Variables marked Secret are available only during sandbox preparation, not while the coding agent is running. If the agent needs a variable to run a command, save it without marking it secret. Note that this exposes its value to the agent, so only provide credentials you are comfortable sharing with it.

Limitations⁠

  • An environment contains one repository, and a repository can belong to only one active environment.
  • Repositories must be available through your workspace’s GitHub integration.
  • Archived environments cannot be used for new sessions.

Use the coding agent⁠

When you delegate work to Linear, Linear Agent creates an secure coding session tied to that issue and begins working from the context already there. Click on the session to steer or follow up with the agent, or let it run autonomously—you’ll be notified when your input is needed.

When you ask Linear to make a change, Linear Agent starts working right away. That includes direct requests like fixing a bug, implementing a feature, opening a PR, or starting a coding session. If Linear has already offered to handle something, a quick reply like “proceed” is typically enough to begin.

If you’re asking Linear to look into something first, it won’t start coding yet. Requests to investigate, debug, research, plan, or create the actual task are treated as scoping the problem, not implementing it. The same goes for messages that only say where the work should be tracked, like a team, project, cycle, milestone, or triage.

While working, Linear Agent can prepare projects that use supported runtimes, including Python, Ruby, Go, Rust, Java, and Node.js, and install dependencies from repository configuration and lockfiles. It can also start applications in its cloud environment, use browser automation to test changes, and capture screenshots or recordings for verification.

Review code changes⁠

When the coding agent produces a pull request, the workflow continues in your Reviews tab. There, you can inspect the resulting diffs and follow review discussions.

When a session includes verification artifacts, you can review the screenshots or recordings alongside the proposed changes.

You can also delegate code review actions to the Linear Agent directly to address any left review comments from other teammates, requirements to rebase open PRs onto master, or fixing tedious lint issues.

Guidance and skills⁠

The coding agent works from the issue you delegate, the repository access available to your workspace, and the guidance configured for Linear Agent.

The agent can use your team’s existing Claude Code or Codex setup, depending on your preferred model. If your repositories include guidance in a dedicated skills.md file or supported tool-version configuration, Linear Agent can use that information alongside the issue and repository context.

Writing effective issues⁠

The coding agent starts with the issue you delegate. The more clearly an issue defines the desired outcome, scope, and constraints, the less time Linear spends exploring the codebase or inferring requirements.

Too ambiguous

Search is broken.

Improve the search experience.

Users can’t find archived projects through search.
This should respect the Include archived toggle.

Well-scoped

Issue search excludes issues from archived projects, even when Include archived is enabled.

Update issueSearch.ts so issue search respects the existing includeArchivedProjects flag. When enabled, issue search should include issues from archived projects that match the search query. When disabled, existing behavior should remain unchanged.

Reuse the existing includeArchivedProjects filtering behavior used elsewhere in search rather than implementing archived project filtering independently.

Do not modify search ranking, pagination, or project search behavior.

The well-scoped example above points Linear toward the relevant code, existing patterns, and expected behavior while defining what should not change. Code Intelligence can also be a useful tool to help iterate on issue descriptions for optimization prior to delegation.

Information that reduces ambiguity or codebase exploration helps Linear reach an implementation faster and use AI credits more efficiently.

For more information about AI credits, see AI credits.

Agent automations⁠

Once you’ve enabled Linear cloud coding agent, you can use Loops to delegate work automatically.

Loops let you run agent behaviors, including coding work, each time an issue arrives in Triage. Optionally, filter to run only when an issue meets specific criteria, like a label, creator, or other filterable property.

Internally, our favorite way to use these automations has been to have Linear take the first pass at issues coming into Triage. Each time Triage Intelligence identifies a new issue in triage issue as a bug, this automation runs to investigate and draft a PR if called for. You can set up automations of your own in a team’s triage settings.

Pricing⁠

Cloud coding agent usage includes model tokens and sandbox runtime, which are charged separately. With BYOK, your provider bills model usage directly, while sandbox runtime still uses Linear AI credits.

Sandbox runtime is billed in 5-minute increments. See AI credits for sandbox runtime pricing and how to review your usage.