Agent operations

Know what is running. Know who approved it.

BeansHost connects every agent deployment to its commit, its cloud, its runs and its approvals, so your team can answer what changed, when and why.

deployment states, every one recorded
17deployment states, every one recorded
prompts or responses stored
0prompts or responses stored
built-in roles with fixed permissions
7built-in roles with fixed permissions
days of audit history on Enterprise
365days of audit history on Enterprise
Operations without fragmentation

Deployment is one step, not the whole job.

The same record follows an agent from the first validation to the last run, so nobody has to stitch together provider dashboards, chat threads and spreadsheets.

Before release

Prove what you are shipping.

  • Validate the specification
  • Pin the commit and checksum
  • Pass the AWS or Azure build readiness checklist
  • Sync environment variables to the provider
During release

Watch it happen.

  • Stream build and deployment logs
  • Hold production for approval where required
  • Activate only after the health check
  • Record every state change
After release

Run it with context.

  • Call the agent through its API
  • Follow runs, failures and usage
  • Review the audit log
  • Roll back when needed
The deployment lifecycle

A traceable path from commit to active agent.

Deployments move through explicit states. Each transition is stored, so the history of a release reads the same in the console, the API and the audit log.

  1. 01

    Validate

    The commit and checksum are compared with the registered version and the plan is prepared.

    validatingvalidation_failedplanned
  2. 02

    Approve

    On Enterprise, production releases wait here for an owner or admin.

    awaiting_approvalapproveddenied
  3. 03

    Build

    Workers are packaged directly. Containers are built in your AWS or Azure account, or by the provider.

    buildingartifact_ready
  4. 04

    Release

    The release only becomes active after the health path answers.

    deployinghealth_checkingactive
  5. 05

    Recover

    Failures are recorded with the provider's logs. Team plans can roll back through the API.

    failedrolling_backrolled_backcancelleddisabled

Deployments start from draft and can be rolled back on Team and Enterprise.

Call deployed agents
Observe

Operational context for every release and run.

Plans and limits are enforced by BeansHost services, not only in the interface. Over a limit, calls return a clear error instead of a surprise charge.

Deployment history

Builder

Live status of every deployment across Vercel, Render, AWS and Azure in one organization view, with logs and Redeploy for each release.

Executions

Builder

Every run with status, timing, size and provider. Summary of runs, success rate, p95 duration and the most common failure over 24 hours, 7 days or 30 days.

Usage

Builder

Runs, failed runs, run time and data transferred by day, agent and project, for this month or last, measured against your plan.

Test run

Builder

Call an agent through BeansHost from the console and see the answer and execution ID without writing any code.

Audit log

Team

An append-only record of privileged actions with result and status. Filter by category, search and export as CSV.

Governance

Enterprise

Waiting releases, past decisions and policy in one place. Choose which risk levels need a person.

Privacy by default. Only run metadata is recorded: status, duration, provider, error category, byte counts and correlation IDs. What was sent to the agent and what it answered are never stored.
Access and governance

Clear roles today. Approvals when production matters.

Every request is scoped to your organization, and every route in the API maps to a permission. Members get one of seven built-in roles.

RoleWhat it can do
Organization OwnerEverything, including governance policy, the audit log and owner changes.
Organization AdminManages members, projects, connections, secrets and billing, and approves production releases.
DeveloperCreates projects, registers agents, manages secrets and API keys, and deploys.
OperatorManages deployments and calls agents. Reads projects, agents and runs.
AuditorReads and exports the audit log, governance decisions and access.
Billing AdminManages the plan and billing through Stripe and reads usage.
ViewerRead-only access to projects, agents, deployments and runs.
Audit

An append-only record.

A database rule stops audit events being changed, or deleted inside the 365-day retention window. Events are recorded on every plan and readable from Team.

Approvals

Production waits for a person.

On Enterprise, production releases that match your policy wait for an owner or admin to approve or deny them with a reason.

Fail closed

No answer means no release.

If the governance engine is unavailable, production releases that need it do not proceed. BeansHost rules work on their own, and ZonalGuard360 can be connected.