Your first repository and agent run.

Quickstart

You need a workspace admin or owner role for these setup steps and manual runs. New GitHub projects start with the runner on and PR reviewer, CI fixer, issue triage, issue implementer and mentions enabled. Review which events you want to trigger work before supplying an AI key. If your workspace already has a key, connecting a repository can start work immediately: arrange a quiet setup window or remove that key until you have disabled unwanted agent triggers.

1. Create an account and confirm your e-mail. Open Sign up. If sign-up is closed, an existing workspace admin must invite you.
2. Connect GitHub. On Get started or Integrations, choose Connect GitHub. Install the app for the account or organization you administer and select the repositories it should reach. Return to InfraStudio Agents and use Add from GitHub to create a project.
3. Choose the automatic triggers first. Open Agents, choose Settings on the project's card, then switch off unwanted agents. Automatic CI-failure runs comment their diagnosis by default; automatic pushing is a separate option. An explicit @app fix command through Mentions can still request a push. For a comment-only evaluation, disable Mentions as well as automatic CI fixer push. Auto-merge starts off and requires readiness checks before live mode.
4. Add your AI key. Open Integrations for the workspace or project, choose a provider below, and save its credentials. A project key is used for that project; a workspace key is shared with its projects unless a project has its own. Use Test where available.
5. Check the project settings. Confirm the repository and default branch. Under Models, choose model IDs your provider account can use; Azure uses the deployment name. Set the CI workflow and deployment job names to match your repository.
6. Run one review. Open Agents, choose PR reviewer, press Run now and enter an existing pull request number. A manual run can start even when automatic triggering is off. Follow it on Runs and read the resulting review in your code host.
7. Review the first result. Adjust the enabled triggers and limits before leaving the project to run automatically.

The pricing page explains project and run allowances. The AI provider bills token usage separately. A working provider key is required even on Free.

Bring your own AI key

Save credentials through Integrations. The platform seals them and supplies them to your provider through a gateway; the agent receives a run token, not your provider key. Replace a key in the same card when you rotate it.

Anthropic

Create a Claude API key in the Claude Console, then save it in the Anthropic card. Use an API account with access to the model you select. A personal Claude chat subscription is not the hosted service's API credential. See the provider's API-key guide.

OpenAI

Create a project API key, or a project service-account key for a shared workload, and save it in the OpenAI card. Select an OpenAI model available to that project. Your API account's permissions and usage limits still apply. See the official OpenAI quickstart and API-key documentation.

AWS Bedrock

In AWS, select a region and a Claude model or inference profile available to your account. The integration accepts JSON with region, access_key_id and secret_access_key; optional session_token supports temporary credentials, which must be replaced before they expire. Grant only the model-invocation access the integration needs, including streaming. Put the Bedrock model or inference-profile ID in the project's model settings. See Bedrock model IDs and regional availability.

{
  "region": "us-east-1",
  "access_key_id": "YOUR_ACCESS_KEY_ID",
  "secret_access_key": "YOUR_SECRET_ACCESS_KEY"
}

Azure OpenAI

Deploy a model that supports the Responses API in your Azure resource. Save JSON with endpoint and api_key in the Azure OpenAI card. The endpoint is the resource origin, such as https://your-resource.openai.azure.com; InfraStudio adds its API route. Use your deployment name in the project's model settings. This integration currently uses an API key, not Microsoft Entra authentication. See Azure endpoints and deployments.

{
  "endpoint": "https://your-resource.openai.azure.com",
  "api_key": "YOUR_AZURE_API_KEY"
}

OpenAI-compatible endpoint

Save JSON with an HTTPS base_url ending in /v1 and an api_key. Optional model selects the endpoint's default model. Compatibility means the Responses API, including the tool use the coding workflow needs; a Chat Completions-only endpoint is insufficient. Test your endpoint before enabling automatic work.

{
  "base_url": "https://your-provider.example/v1",
  "api_key": "YOUR_PROVIDER_KEY"
}

Connect product signals

• GA4: add a service-account JSON key with Viewer access to the property.
• PostHog: add a project-scoped personal API key using its MCP Server preset, then set the PostHog project ID in project settings.
• Sentry: add a read-only token for the organizations, projects and events you want reviewed.
• Clarity: add the project's data-export token. The scheduled collector performs the check; repeated manual probes would consume its daily request allowance.

The analytics review needs a configured analytics source; error triage needs Sentry. Enable the relevant schedule only after its integration is ready. Inspect the first finding and its evidence before asking the agents to implement it.

Using GitLab

For GitLab, manually create a project and set its repository, code host and GitLab address in project settings. Add a token with api and write_repository scopes and Developer or Maintainer access. Review the agent triggers before adding any AI key, generate a GitLab webhook from project settings, and install the supplied URL and secret in GitLab. Manually created projects start with the runner off: turn on the project runner in settings after setup, or webhook deliveries will be ignored. A token for a custom host belongs to that project, and changing the destination requires entering it again. GitLab uses merge requests and pipelines; GitHub App installation and Auto-merge are GitHub-specific.

Repository instructions

Defaults work without a custom folder. To tailor a workflow, commit a .agents/ folder to the default branch. Prompts use .agents/prompts/<name>.md, including review, plan, implement, revise, pr-review and ask. Collector queries use .agents/clarity-queries.txt. Review instruction changes like code. Platform restrictions on credentials, protected paths and allowed tools remain in force.

If the first run does not complete

• No provider key: check the selected provider and whether the key is stored for this project or its workspace.
• Authentication or quota failure: check the provider account's model access, key expiry and usage allowance. Changing the platform plan does not change the provider's quota.
• No repository: check the GitHub installation selection or GitLab token and host, then confirm the default branch.
• Queued work: inspect Runs and the plan's concurrency/monthly allowance. A project with its runner off cannot start work.
• Unexpected output: read the run summary and review comment, then ask for a change or reject the finding. Start with a smaller, explicit task.

For help, contact contact@infrastudio.io. Read how access is limited and the buyer questions.