Intent Solutions · how we build
How we build
A small team plus AI agents, running 40-plus software projects. This page shows the actual loop that makes that possible: tasks that live with the code, one cockpit that watches everything, and a governed second brain that answers with receipts. No slideware — every piece below is running right now, and the numbers on this page come out of it. We're an AI consultancy; this is the system we run on ourselves, and the playbook we bring to client work.
The loop
work → visibility → knowledge → workFive stages, one cycle. Work is tracked next to the code it changes, a cockpit keeps the whole estate honest, and only durable conclusions — never volatile snapshots — get compiled into memory agents can cite.
STEP 1
REPOS
43 task databases, one per project. The backlog ships inside the project itself.
STEP 2
MISSION CONTROL
A read-only scanner sweeps every database and rebuilds the cockpit on command.
STEP 3
COCKPIT + DRIFT
One page: what's open, what's active, what's stale — and where the docs disagree with reality.
STEP 4
GOVERNED BRAIN
Durable conclusions get compiled, then governed — facts with sources, never volatile snapshots.
STEP 5
AGENTS AT WORK
Agents answer with citations, not hallucinated memory — then pick up the next task.
-
REPOS
43 task databases, one per project. The backlog ships inside the project itself.
-
MISSION CONTROL
A read-only scanner sweeps every database and rebuilds the cockpit on command.
-
COCKPIT + DRIFT
One page: what's open, what's active, what's stale — and where the docs disagree with reality.
-
GOVERNED BRAIN
Durable conclusions get compiled, then governed — facts with sources, never volatile snapshots.
-
AGENTS AT WORK
Agents answer with citations, not hallucinated memory — then pick up the next task.
↺ and the loop returns to the repos — work continues.
Live snapshot
These totals come straight out of the loop above — a sanitizer script reads the estate-wide scan and publishes totals only. No project names, no task titles, just the shape of the work.
Snapshot as of 2026-07-02 (partial scan) · refreshed manually for now — weekly auto-refresh is on the roadmap.
The chapters
Three short reads, each one stage of the loop — written for humans first, engineers second.
Chapter 1
Tasks that live with the code
Every project carries its own task database in the repo itself. Agents find work, claim it, and close it with evidence — no external tracker to drift.
Read chapter 1 →Chapter 2
Mission control
One scanner sweeps every task database on the machine and regenerates a single cockpit page — plus drift findings where docs and reality disagree.
Read chapter 2 →Chapter 3
The governed second brain
Compile, then govern. Knowledge gets derived from real docs, gated by deterministic code, and every answer ships a citation and a tamper-evident receipt.
Read chapter 3 →It's open source
The pieces of this loop are public. Star counts checked 2026-07-01.
gastownhall/beads
Distributed graph issue tracker for AI agents. The upstream project our task layer runs on.
tobi/qmd
On-device hybrid search for markdown, by @tobi. Every hit is a citable qmd:// URI.
governed-second-brain
The map of the compile-then-govern stack: what it is, how it fits, why it beats recall-only memory.
governed-second-brain-plugin
The installable product — cited search and governed capture as a local-first Claude Code plugin.
intentional-cognition-os
Derives knowledge from raw corpus — summaries, concepts, contradictions — and emits a governance spool.
qmd-team-intent-kb
The deterministic control plane: dedupe → policy → promotion, with an append-only audit log.
Also public: claude-code-plugins-plus-skills (★ 2,400+ — 425 plugins, 2,810 skills, 200 agents) · tonsofskills.com · learn.intentsolutions.io · labs.intentsolutions.io · intentsolutions.io
Roadmap
This page is an MVP on purpose. Here's what's live and what's coming — no dates, just order.
- ShippedThis page — the loop, three chapters, and a real estate snapshot with manual refresh.
- PlannedWeekly auto-refresh — the snapshot regenerates itself every Sunday instead of by hand.
- PlannedVideo embeds — a YouTube walkthrough per chapter, showing the real thing running.
- PlannedMore chapters — agents & sessions, the drift deep-dive, and a compile-to-citation end-to-end.
- PlannedLive public stats endpoint — the totals as JSON, not just baked into this page.
- PlannedInteractive cockpit sample — a read-only cockpit you can click through, on synthetic data.
- ConsideringThis site as its own repo — with the disclosure gate running in CI on every change.
Stay in the loop with Intent Solutions
AI systems that actually ship — new builds, deep-dives, and releases. No spam.