Home
AI-native development
Context Engine
Every work session is distilled into a searchable knowledge base. Ask a question in plain English and get back the exact decisions, trade-offs, and system notes — ranked, cited, in under a millisecond.
Ask in plain English
❯cq"what happens when a planet absorbs a moon?"
- absorption-spec▸Mass + skill transfer on contact0.95
- moon-lifecycle-spec▸Orbit → fire → absorb states0.91
- stardust-mass-model▸Mass is HP; nothing is lost0.88
1,711 sections · 219 docs · <1ms query · local-first
What it does
Reasoning, not just code
- Every work session is distilled into a structured document — the questions that were asked, the decisions that were made, and the alternatives that were rejected, with the reasoning behind each.
- It captures intent, not just implementation: the trade-offs, constraints, and systems behind every choice — the context a new teammate or agent would otherwise have to reconstruct from scratch.
- Documents are distilled from real work as it happens, not a wiki of hand-maintained pages that drift out of date.
Ask in plain English
- Query the knowledge base the way you'd ask a colleague — natural-language questions, no exact keywords required.
- It finds the right context even when your words don't match the document's, and returns ranked, cited sections instead of a wall of text.
- Answers bring their own neighbours: related documents come along automatically, so one question surfaces the whole connected thread.
Knowledge that compounds
- Finished work writes itself back into the base — every completed task becomes context for the next one.
- Topics stay organised into clusters and a cross-linked graph, so the base grows richer and easier to navigate rather than noisier.
- Runs entirely on the machine — no external services, no network round-trips, nothing leaves your environment.
Deep dives
How we work
Memory is one habit of five.
See the rest of the loop that turns a one-line pitch into a game we can play.