Coordination infrastructure
Lacunari
Coordination infrastructure for distributed work — a living roadmap that reads an entire project and tracks what exists, what is missing, and what to do next.
The bottleneck in a scaled team was never how fast the model writes code. It is that the plan lives in a handful of people's heads.
Lacunari is what happens when you make the plan machine-readable. It reads an entire project — code, documents, scans, research — and turns it into a living roadmap: what exists, what is missing, and what to do next. Then it lets dozens of workers, human and otherwise, claim tasks from one shared queue without stepping on each other.
What it actually does
Content-addressed work. Two people writing the same document produce the same document, and the system knows it. Identical contributions are recognised as identical regardless of who made them, which is what stops a large team from paying twice for one piece of work.
Gap detection. Undocumented code, untested functions, orphaned files, missing data. These are found by set-based SQL over data you already hold — no external APIs, no network calls for core indexing. The gaps are the roadmap.
Atomic claiming. Task assignment runs through Postgres transactions, so two workers cannot pick up the same item. This sounds obvious and is the single hardest thing to get right once there are more than a handful of contributors.
File locking that declares conflicts before work begins. Not after. Merge conflicts are cheap to prevent and expensive to resolve.
The design position
No vendor lock-in, deliberately. It runs entirely on Postgres 12+, and the workers can come from anywhere — Claude, Llama, GPT, or a shell script somebody wrote on a Tuesday. Nothing in the core depends on a provider staying in business or keeping its prices.
It is licensed AGPL-3.0, with commercial licences available for organisations that need them.
Why I built it
Every project I run has the same failure: work gets duplicated, gaps go unnoticed until they are expensive, and the map of what needs doing exists only in my head. That is survivable with one person. It is not survivable with thirty, and it is definitely not survivable when some of them are agents working in parallel at three in the morning.