andeye · agentic coding on your own Mac

vibe coding
verified

Adversarial agentic models plan, build, test and review each other's work in a secure sandbox – so you ship code you can trust.

This very page was planned, built, tested and reviewed by a vibe session.

In development – built on the same open-core values as everything andeye makes.

▾ watch a session run – click the commands yourself

The demo

Drive a session yourself

Click a command on the left and watch the terminal play out exactly what vibe does. The checklist under each command ticks off as it happens – and clicking again replays it.

click a command to run it

Launch

one command, from any project folder

  1. finds your project and its GitHub remote
  2. checks the repo's own fine-grained PAT – it reaches this repo, and only repos this session declares
  3. builds the sandboxed container (once – reused in seconds after that)
  4. verifies the firewall – outbound on a short allowlist
  5. signs Claude Code in with your Claude subscription – no API key

Build

nobody marks their own homework

  1. a planner turns your ask into a spec with real acceptance criteria
  2. a spec critic attacks the spec before any code
  3. a builder writes the code
  4. an independent tester runs every check
  5. fail means fix and re-test – claims don't count, passes do
  6. a reviewer hunts for reasons to say no

no arguments – it finds its own work

  1. reads your TODO.md backlog
  2. picks the highest-leverage open item itself
  3. runs the whole adversarial loop unattended
  4. commits with a full audit trail for your review

burn the rest of your session productively

  1. loops the solo run, item after item
  2. questions go to your notes – it never stops to wait
  3. commits keep landing while you sleep
  4. rolls into a fresh credit window when one runs out
  5. stops itself at the perfection gate

Red team

traffic off the allowlist

  1. every outbound connection is checked against a short allowlist
  2. GitHub, npm, Anthropic and a few named extras are allowed – the rest is refused

the content guard reads every diff first

  1. staged diffs and commit messages scanned before every commit
  2. secrets and personal data both stop the commit – secrets as BLOCK, personal data as WARN
  3. pushes re-scan the outgoing range for secrets before anything leaves

Ship

out through the one-repo PAT

  1. pushes with the repo's own fine-grained PAT
  2. lands on GitHub – Radicle support is on the way
  3. docs and CHANGELOG travel in the same commit

Also in the box

what this month cost

  1. month-to-date model spend, across every vibe on this Mac

lessons that travel

  1. captures a cross-project lesson to your learning library (with your say-so)

container → Mac clipboard

  1. copies Claude's last code block straight to your Mac clipboard

token thrift on demand

  1. lean mode – fewer subagents, terser output; /feast turns it back
yourproject — vibe — 96×28
click a command to run it
$

That transcript is the real shape of a vibe session – and this very page shipped from one.

Vibe andeye

Letting an AI write your code shouldn’t mean accepting whatever it produces. Vibe andeye puts adversarial agentic models between your idea and your codebase: one plans, one builds, and independent critics, testers and reviewers attack the result until it actually holds up.

Who it's for

For developers first

Vibe is for developers first – but we'd also welcome feedback from non-coding experts who know exactly what they want.