"All tests pass" - they don't
The agent reports success. You check. Half the suite is red. You only find out after it already committed.
A guard for your AI coding agent
Your AI can write code fast. Aufsicht puts a simple gate in front of every change: allowed, checked, or stopped. It runs on your machine and shows you what happened.
01 · The problem
The agent reports success. You check. Half the suite is red. You only find out after it already committed.
It fails, retries the same broken approach, fails again. No memory of why - just quietly burning your budget in a loop.
"Don't touch the auth module." Three prompts later it's rewriting your auth module. Rules in a prompt are suggestions, not guardrails.
When something breaks, you're left reconstructing what the agent did from git blame and vibes. There's no record you can trust.
02 · How it works
Nothing touches your files until it passes two checkpoints. Your agent can only ask - "here's what I want to do, and here's my result." Aufsicht answers yes or no. And the AI can't bluff: instead of trusting what it claims, Aufsicht re-reads the real result from your disk and confirms it. A step that breaks your rules, or a result that doesn't match reality, is refused - not just discouraged.
The first checkpoint asks one thing: is this step allowed right now? Everything it depends on must already be done, and it can't skip ahead. If the agent claims "done" without doing the work, it's refused instantly - your file isn't even created yet.
The change is written to a temporary file next to the real one - never the real file yet. Aufsicht re-reads it from your disk and checks it against your rules. If it fails, the temporary file is deleted and the rejection is logged - the file you care about was never touched.
When something fails, the agent gets a plain reason - like "expected 20 rows, got 14" - not a cryptic error. It gets a few tries to fix it, then it stops and flags you. No infinite retry loop quietly burning your budget.
Every yes and no is written to a tamper-proof log that's re-checked each time Aufsicht starts. "Did it really happen?" is answered with proof, not memory - and nothing can be edited or deleted quietly.
You give one prompt. Aufsicht turns it into a clear plan, the way a team would: it breaks the work into steps, gives each step an owner, and suggests using your best model where it matters and cheaper ones for the simple parts. You can read the whole plan in the dashboard before any work starts - and every step still has to pass the two checkpoints above, in order.
Set up a team like a real org - CTO, PM, architect, backend & frontend, QA, security, DevOps. Each role owns its own steps and they work in parallel, so nothing collides. Only have one AI? Point every role at it. Your call.
Handy built-in tools: calculator, JSON, text, regex, dates, a safe web fetcher, and a locked-down terminal with time and output limits (no free shell access). Any agent can use them, and every call is recorded.
When the agent needs an API key, it asks Aufsicht by name - it never sees the value, and neither do we. You paste it once; it's encrypted on your machine and never reaches git, your logs, or the AI. It's injected only into the exact command that needs it, then wiped from the output.
Ask it to ship to Vercel, Netlify, Fly, Railway, or Docker and Aufsicht runs the real deploy with your secrets injected. If it fails, you get a clean reason back - with no secrets exposed.
Want the technical internals? They're all on the technician page.
03 · Live demo
You play the agent submitting work. Aufsicht inspects it at two checkpoints - is this step allowed, and is the result real - with a holding stage in between where your change is parked but not applied yet. Pick a story and watch each stage give its verdict.
The stages run in order: Register (your order is logged) → Plan (split into steps, one owner each) → Post 1 · Allowed? → Held (change parked, not applied) → Post 2 · Real? → Commit (written to the file, decision logged). A step that fails never reaches the next stage, and the rejection stays on the record.
Type an order like you would to your agent, pick a scenario, hit Run. You'll see the whole path: order recorded, split into steps, two inspections, verdict.
04 · Pricing
Choose one path: Free to start without payment, Pro for a small team, or Max for a larger organisation. Click a plan and we will keep your choice through sign-in and payment.
Choose a plan once. Sign in once. You will return to the same checkout.
For solo builders & trying it out
Rp0 forever
No card required
For small teams
Rp149rb / month
or Rp1,49jt / year (2 months free)
For orgs & enterprise
Rp399rb / seat / month
or Rp3,99jt / seat / year (2 months free)
Let a key expire or deactivate it and you drop to Free automatically. Your ledger stays intact - only the paid flags switch off. Each term is bought separately, so nothing renews or charges you on its own.
05 · Proof
Straight talk: Aufsicht is not here to make your AI faster or smarter. It guards and records - sometimes that costs a little more, sometimes less. What never changed across our tests: no bad changes slipped through after the fix, a full record every run, and the retry limit held every time. Cost went both ways (+20% on one job, −58% on another), because how much the model explores matters more than the guard does.
06 · FAQ
Any modern AI coding agent that supports the MCP standard (most do). One setup command connects whatever it finds on your machine. You connect once, then keep prompting like normal.
No. It guards and records. Sometimes that costs a little extra, sometimes the tidier flow costs less. What it never does is fake the result.
Every rejection comes with a plain reason you can read, fixes are capped so it never loops forever, and a single switch turns the checks off while you look into it. A guard you can't switch off is a risk, not a feature.
The guard itself, yes - everything runs locally on your machine. The rest depends on your model: cloud needs internet, a local model doesn't.
Nowhere. Reports carry only statuses and proof codes - never your files or secrets. The logbook, the lock, and the saved results all live in your own folders. There's no cloud to send them to.
The agent asks Aufsicht for a secret by name, never from you and never the value. You paste it once; it's encrypted on your machine and never reaches git, the log, or the AI. When a deploy needs it, Aufsicht injects it into that command only and scrubs it from the output. The AI sees the name and whether it worked - never the value.
A prompt requests. Aufsicht enforces: rule-breaking moves are refused automatically, and every approval leaves recorded proof. Try the demo above.
No. Pick "Run in Background" from the launcher (or run one command) and the gateway detaches from your shell and keeps running after you close the terminal, like a small always-on service. Stop or check it any time. Works the same on macOS, Linux, and Windows.
Bonus · Play
Break the pixel AUFSICHT above, piece by piece, until 100% consensus. Move the paddle with mouse, finger, or arrow keys.