Vakr — Mission Control for Coding Agents
Vakr is a native SwiftUI Mac app with a small background daemon that supervises the coding agents working on your Projects. One window shows each Project's progress, its agents and its share of your weekly usage limit. Everything waiting on a human — questions, permission prompts, failures, reviews — lands in a single Needs you list you can clear a keystroke at a time. Its core idea is in the headline: don't read the diff, read the proof. Every run is checked before merge — the tests you approved, whether they passed, their exit codes — and Vakr blocks the merge when what the agent says it did doesn't match what it actually ran. Optionally, a second agent from a different vendor tries to break the change by writing a failing test. Each run gets its own git worktree and branch, so agents never touch main. Claude Code and Codex are supported natively and about 22 more agents over ACP, using the subscriptions you already have. Per-Project policies block commands like force-pushes, a single switch halts every run, other Macs or a Linux VPS can join as Machines, and approvals can come from your phone over Tailscale. Code, sessions and the database stay on your Mac — no account, no analytics.
Products I own and operate end to end — my roadmap, my users, my on-call.
- ▸Running several coding agents across several Projects turns into a dozen terminals, each silently waiting on a question nobody saw
- ▸Agents report success they didn't earn — a summary that says tests pass is not evidence that tests ran
- ▸Reviewing every line of an agent's diff doesn't scale, but merging without review isn't acceptable either
- ▸Weekly usage limits get burned by whichever Project happens to run first, not the one that matters
- ▸One window over every Project, plus a single Needs you list that collects questions, permission prompts, failures and reviews from every agent and Machine
- ▸Proof-based review: each run records the approved tests, their results and exit codes, and the merge is blocked when the agent's claims and its actual commands disagree
- ▸Optional cross-vendor adversarial check — a second agent from another vendor tries to write a failing test against the change
- ▸Isolation by default: every run gets its own git worktree and branch, and per-Project policies block destructive commands such as force-pushes, with a global halt switch
- ▸Per-Project budgets as a share of the weekly limit, with the queue paced against it so overnight runs land without starving the day's work
Live at vakr.app — free with all features during early access
Supports Claude Code and Codex natively and about 22 more agents over ACP, on the subscriptions you already pay for
Local-first: code, sessions and the database stay on your Mac — no account, no analytics, 0% CPU at rest
Multi-machine: pair another Mac or a Linux VPS and approve from your phone over Tailscale
JobJam.io — AI Job Search & Application Platform
AI-powered job search platform: discover roles, evaluate fit, tailor applications, and close skill gaps. One-time pricing, no subscriptions.
Fathohm — Comprehension-Debt System of Record
See how much of your codebase no human understands. A treemap of your repo colored by human comprehension score, with a tracked, assignable metric for 'who understands what' in an AI-native codebase.
Grenz — Firewall for AI Agents
A local firewall for AI agents: scoped, revocable, observable permissions instead of your full GitHub token. Deny-by-default, MIT licensed, live at grenz.dev.