Blog
Concrete scenarios and workflows — what becomes possible when people and agents build on shared, proven knowledge instead of re-deriving it every time. Roughly weekly.
Latest
One command turns a knomit knowledge base into 566 markdown files. The docs stop at git push; this is what you have afterwards — the anatomy of a bundle, and four ways to spend it, including Gemini and Claude both answering a design question straight off the files, with no knomit, no server and no MCP.
The same task, with and without the knowledge base, eight runs per arm, judged blind. The assisted arm was always faster and cheaper — but one subtly flawed fact made it worse than no knowledge at all. Repairing that single fact flipped 30 of 32 blind comparisons and prevented a mistake models invent unprompted.
Store a model's speculation in the same shape as a verified observation and retrieval will grant them equal authority — that is how a guess gets laundered into a decision. knomit's answer is the move programming languages made decades ago: type every fact by its epistemic status and its origin, and enforce what may be derived from what. A guess can be stored, retrieved, even confirmed later — but it cannot pass as an observation.
Twenty-five years ago the industry tunneled RPC through HTTP POST until REST pointed out that the protocol already had the semantics. Agent memory is making the same mistake today — rebuilding provenance, history, identity, and sync inside vector stores, or auto-committing files into git while ignoring what its operations mean. knomit's bet: a commit is an assertion, a branch is an identity, a merge is consensus. Use git the way REST uses HTTP.
You shipped the feature and merged the PR — and somewhere in your knowledge base, a dozen facts just quietly went out of date. The fix isn't to write more carefully; it's to make the corpus introspect on itself. Here is a real, interactive pass where Claude Code swept knomit's own knowledge base fact-by-fact, checked every claim against the source instead of trusting the corpus, and corrected the drift — recording each correction instead of erasing it. This is the learning loop closing.
knomit's own repo runs on knomit. We started where everyone starts — a fat CLAUDE.md and a per-machine MEMORY.md — and watched them rot and crowd out context. Here is what replaced them: typed, scoped knowledge surfaced at the moment of the decision, and one real fact that climbed from hunch to proven.