Controlled GitHub issue agents

Create a controlled
AI work agent.

Create a controlled AI work agent, give it credits, connect GitHub, set spend and approval rules, and inspect every action.

Join the waitlist →

Task, run, rerun GitHub issue flow Spend + approvals

Atlas agt_123

Avery · Auza Labs

Verified

PERMISSIONS

GitHub Issues Spending Approvals

MONTHLY SPEND LIMIT

$49 / $200

RUNS AUDITED

18 logs

APPROVAL REQUIRED

github.issue.create model.run spend threshold

LAST ACTION Created GitHub issue #42 now

Approval requested

github.issue.create

BUILT AROUND THE FIRST SHIPPED FLOW

Agent recordsPromotional creditsTask runsGitHub AppGitHub issuesSpend limitsApproval queueAudit logReruns

Current controlled action

github.issue.createserver-side admissionserver-only tokensstored run steps

The problem

Agents need more than prompts.

Even the first useful agent action needs guardrails. Before an agent turns a task into a GitHub issue, Auza checks its org, status, credits, permissions, spend limits, approval rules, and linked repo.

Identity

Every agent is tied to an authenticated user and organization, with status, purpose, model, credits, and control settings visible in one place.

Permissions

Grant only the actions the agent can take. The shipped flow requires an explicit server-side rule before github.issue.create can run.

Spending

Set monthly and per-action limits. Require approval above a threshold, and block work before any model or GitHub side effect when a rule trips.

Activity logs

Every task, run, debit, approval decision, tool step, and GitHub issue result is written to the run detail and audit history.

The Agent Record

One control record for every agent.

Each agent has a live control record showing who owns it, which GitHub repo it can touch, which actions are allowed, and what happened in each run.

Agent	Owner / org	Purpose	Allowed tools	Spend	Approvals	Linked repo	Runs	Status

Controlled GitHub issue flow View run history →

For developers

Start with one controlled action.

We're shaping the developer workflow with early teams before publishing a public code surface. The first release focuses on task runs, credits, GitHub issue creation, permission checks, spend controls, approvals, and audit trails that are easy to reason about.

✓ Agent and org-scoped control records
✓ Permission & spend-limit enforcement
✓ Approval queue with stored payload resume
✓ Every run and tool step written to audit

Join developer waitlist → Talk to the team

Create your controlled agent.

Get early access to the GitHub issue flow with credits, approvals, spend limits, audit, and reruns.