Let agents work.
Don't let them run your stack.
Run Claude, GPT-4o and Gemini in deliberation.
Approve every action. Audit everything across your projects.
Founders' pricing locked for life. Refundable until launch (Sun 11 May, 9pm AEST).
Your agents are shipping code while you sleep.
You've already used Cursor. Claude Code. Maybe Devin. They're powerful. They're also opaque, autonomous, and one bad prompt away from rewriting your production database at 2:47am.
You don't need another agent. You need a layer that sits above them.
Unbounces is the cockpit. Three models deliberate before any action. You approve before anything ships. Every keystroke logged, signed, and replayable. Kill switch under your thumb.
See the cockpit.
The control surface for everything your agents do across every project.
Built for solo builders running production.
The pieces that turn raw model access into operator-grade tooling.
Multi-LLM deliberation
Three providers weigh in before any action. Disagree? You see the dissent. Cost capped per call.
Cross-project memory
Context follows you across every repo. The cockpit remembers what you decided in one project yesterday when you're working in another today.
Approval gates
Nothing executes without you. Every action surfaces in the queue. Approve, reject, or send back with notes.
Audit trail you can prove
Every event HMAC-signed. Replay any action. Export to your own storage. Nothing happens off-the-record.
Three steps to a controlled fleet.
Connect your projects. Watch deliberation happen. Approve before anything runs.
Connect
Bring your own keys. Plug in Claude, GPT-4o, and Gemini. Add your repos. Cockpit boots in under a minute.
Deliberate
Every prompt routes through three models. They debate. They vote. The dissent is visible to you.
Approve
Nothing ships without your tap. Audit trail records the verdict, your call, and the result. Replayable forever.
Founders' deal — first 50 only. $20 refundable deposit. Lock pricing for life.
Built for the operator, priced for the indie.
No seat tax. No usage tax. Bring your own keys. Cap your own spend.
Solo
For one operator running a small fleet.
- Up to 3 connected projects
- Multi-LLM deliberation (BYO keys)
- Approval gates + audit trail
- Cross-project memory
- 24/7 SSE stability monitor
- Multi-user RBAC
- Webhook event stream
Studio
For the operator running a serious team.
- Unlimited connected projects
- Multi-LLM deliberation (BYO keys)
- Approval gates + audit trail
- Cross-project memory
- 24/7 SSE stability monitor
- Multi-user RBAC
- Webhook event stream
No credit card today. Deposit fully refundable until launch (Sun 11 May, 9pm AEST).
About
Built solo by Andrew Pogany.
Solo developer in Melbourne. Twenty-plus years shipping Australian fintech, mortgage origination, and infrastructure software. Eight active codebases right now, all running through the cockpit you're looking at.
Unbounces is the tool I built because I needed it. Three models deliberating, every action audited, kill switch under my thumb — that's the workflow I trust with production. I figure if it's good enough for my own systems, it might be good enough for yours.
Direct line. Email goes straight to me. Bug reports get triaged in hours, not weeks. Feature requests get shipped or rejected with a real reason. No ticket queues. No support tiers.
Questions you probably have.
General
What does Unbounces actually do?
Unbounces is a cockpit that sits above your AI agents (Claude Code, Cursor, Devin, etc.) and gates every action through multi-model deliberation, your approval, and a signed audit trail. You bring your own keys; we provide the control surface.
Who's this for?
Solo developers and small teams running real production with AI agents. If you've already had a 2am moment where an agent did something you didn't authorise, you're the audience. If you're still curious about whether to use AI in production, this might be early for you.
Is this a wrapper around the model APIs?
No. It's an orchestration and governance layer. The models don't talk to each other natively — Unbounces sequences them, formats their outputs into structured deliberation, captures dissent, and routes the final decision to your approval queue.
Security & control
Where do my API keys live?
Your keys live in your own vault, encrypted at rest with AES-256-GCM. They never appear in logs, never leave your environment unencrypted, and never touch our servers in plaintext. We use them just-in-time to make calls on your behalf.
Can I kill everything in flight?
Yes. Single keystroke (Ctrl+K from anywhere in the cockpit) cancels every in-flight model call, halts every approval queue, and freezes every webhook. Hard kill, audited, reversible.
What about data residency?
Audit logs are stored in your own Supabase instance (your project, your region). We hold zero customer data on our infrastructure. Your model calls go directly from your environment to the providers.
Pricing
Why a deposit?
Founders' pricing for the first 50 customers locks in for life — that's a meaningful commitment, so the deposit confirms intent. It's $20, fully refundable until launch (Sun 11 May, 9pm AEST). If you change your mind, you get every dollar back.
Can I change tiers later?
Yes. Up or down, any time, prorated. Solo and Studio share the same core engine — the difference is project count and team features.
Technical
Which models are supported?
At launch: Claude (Anthropic), GPT-4o (OpenAI), and Gemini (Google). Provider list will grow — xAI's Grok and Mistral are queued for evaluation.
Does it work with my existing agent setup?
Yes if you're using Claude Code, Cursor, or any agent that exposes a webhook or polls a queue. We're working on direct integrations for the major agent platforms — current shipping integrations: GitHub webhooks, Slack, custom HTTP endpoints.
Ship in May.
Lock founders' pricing for life.
$20 deposit. Refundable until Sun 11 May, 9pm AEST. First 50 spots only.
Reserve your spot — $20 deposit