Work / Lab
Quorum
Local-first private-AI diligence: open-weight models via Ollama, five workstream leads, a risk committee, an IC memo.
- Tauri 2
- React 19
- TypeScript
- Tailwind 4
- Ollama
- SQLite
- Vitest
TL;DR
- Quorum was the local-first first cut of Concord: private AI for M&A due diligence that never has to leave the user’s machine.
- I forked my Redl app and, directing Claude Code, repositioned it around open-weight models on an Ollama sidecar, with optional per-role cloud routing.
- Its gated end-to-end test ran on a local 7B model and produced an IC memo that flagged a change-of-control clause.
What I built
Five workstream leads (financial, legal, commercial, operational and tax) read the data room and feed a risk committee. The committee includes a devil’s advocate, and its output is an investment-committee recommendation. By default every model runs locally through Ollama; any role can be routed to a cloud model instead.
Why it’s archived
Its one recorded end-to-end run showed the workflow holding together on a small local model. The next question was what frontier models from several labs would do on the same desk, checking each other. So I forked it again, and that fork became Concord.
Verified numbers
| Metric | Value | Source |
|---|---|---|
| Model in the gated end-to-end test | Qwen2.5-7B (local) | local: CONCORD-TRANSFER/repos/quorum.bundle |
| End-to-end memo verdict | PROCEED WITH CONDITIONS | local: CONCORD-TRANSFER/repos/quorum.bundle |