- verification runs, which IronBee executes on its own infrastructure for your deployments and pull requests
- sessions, which your AI coding agent records with the IronBee CLI
Account
Your account is the workspace your team shares. It holds your projects, integrations and settings. Everyone in it has a role: Owner, Admin, Member or Billing Admin. See Roles and permissions.Project
A project is what IronBee verifies:- a GitHub repository
- a Vercel project
- a Netlify site
- a CLI project: a name your IronBee CLI reports under
- Web: open the deployment in a browser and exercise it as a user would.
- API: call the deployment over HTTP and check the responses against its contract.
Verification run
A verification run is one verification of a change: a preview deployment, a push, a pull request, or a run you start yourself. An agent reads what changed, exercises it against your running app, and records a result with the evidence behind it. Every run has:- A trigger, what started it: Vercel, Netlify, GitHub Actions, API, CLI or Console.
- A status, where it is: Created, Queued, Starting, Running, Completed, Failed, Cancelled.
- A result, what it concluded: Pass, Fail, Not applicable (the change left nothing to verify), Soft fail or Unknown.
- The source it verified, for runs tied to a GitHub repository: repository, branch, commit and pull request.
Environment, variable and secret
An environment is one deployment of a project that runs are verified against, such aspreview or staging. Integrations create environments automatically. Its URL, OpenAPI document and Domain properties tell the run about the deployment. See Environments.
Variables are plain values a run is told outright. See Variables.
Secrets are credentials a run can use without seeing them: sign-in credentials, tokens, headers, a deployment protection bypass. They’re written once and never shown again. See Secrets.
Variables and secrets apply to all environments, or to one.
Evidence
The evidence of a run is everything it recorded:- the recording and screenshots
- the actions it took
- the network requests
- the traces
- the logs
Session
A session is one working session of your AI coding agent with the IronBee CLI, from start to finish. It records what the agent did: its activities, its verification cycles, and their verdicts. An activity is one agent turn: from your prompt to the moment the agent stops.Verification cycle
A verification cycle is one pass of verification inside a session. The CLI verifies through six platforms:
Every enabled platform that matches the changed files takes part in the same cycle. Turn platforms on or off with
ironbee <platform> enable|disable. See Verification.
In the console, CLI cycles appear among the project’s verification runs with the CLI trigger.
Verdict
A verdict is the agent’s assessment of whether its change works:status:pass,fail, ornot_applicablewhen the change has nothing to verifychecks: what the agent confirmed, such as “form submits successfully”issues: what went wrong. On a pass they’re advisory; on a fail they must be fixedfixes: what was changed to resolve earlier issues
Fix
A fix is a code change the agent makes after a failed verification. The agent then verifies again, up to a retry limit. The GitHub Action can fix findings too, and commit the fix for you.Verification modes
The CLI runs in one of three modes:
See Verification modes.
Analysis, findings and recommendations
Once a day, IronBee analyzes each project’s recent sessions:- The Quality analysis measures how well the agent verifies.
- Findings are observations grounded in that data, such as a file with a high fix rate or a pattern of retries. They aren’t necessarily bugs.
- Recommendations are directives generated from findings.
Config files
The CLI reads its config from three layers, deep-merged (project-local wins over project, which wins over global):
The
IRONBEE_SERVICE_API_KEY environment variable overrides service.apiKey from any config file. See Configuration.