Use cases
The same git-native substrate serves a solo developer, a research fleet, and a whole network of people and agents. A pool of what people build with it.
For developers & their coding agents
Knowledge that lives where the work happens.
Invariants, gotchas, and design rationale that never make it into comments. Your coding agent — and your teammates — read them before touching an area, and write new ones when they discover something non-obvious.
invariants · gotchas · decisions — in the repo
When an agent makes a call — picks an approach, rejects an alternative, resolves a tradeoff — it records it as a typed, signed fact. Not memory: actionable knowledge. The next agent reads the rationale instead of re-deriving it, and you can trace exactly what was known when any choice was made.
self-documenting · signed · traceable
Run review or hypothesize at higher effort to discover emergent keystones — the load-bearing premises that underwrite facts across your codebase but were never written down. knomit ranks them by blast radius: how much breaks if the premise is false.
discovery · keystones · blast radius
For researchers & analysts
A corpus that sharpens itself.
Point agents at sources; they learn typed facts continuously. Synthesis distills patterns and higher-order insight as the corpus grows — the knowledge base sharpens with every cycle.
continuous learning · synthesis · compounding insight
Observations distill into syntheses; syntheses extend into falsifiable hypotheses with settlement criteria. When evidence arrives, the hypothesis transitions to observation — and git records the exact moment the loop closed.
observe → synthesize → hypothesize → confirm
Synthesis & hypothesesFacts beget facts; confidence rises and falls; query the graph as-of any point in time — so the corpus gets sharper over time, not just bigger.
confidence moves · query as-of any commit
The temporal graphFor fleets & open networks
Many minds, one converging truth.
One source of truth that humans and agents build together. Because facts are plain markdown in git, you review knowledge in pull requests and trust provenance instead of folklore.
author together · review like code
Peers, branches & consensusSpin a fleet at one task; each agent self-documents its approach and findings on its own branch; all converge to a shared repo. Synthesis distills where approaches agreed, where confidence diverged, and what only one agent found — a knowledge graph of why, not just a leaderboard.
fleet · per-branch findings · distilled comparison
A public knomit repo anyone contributes to via pull request — a shared commons with full provenance, where deduplication means the network pays each learning cost once.
public repo · PR contributions · paid once
For decentralized & edge deployment
It's just git — so it goes anywhere, and composes any way.
Instances compose. One instance points at a repo hosted by another, so you can wire a DAG: a raw-ingestion node feeds a distillation node feeds a consensus node. An ETL pipeline for facts, where every stage is itself a queryable knomit.
compose instances · ingest → distill → consensus
Every peer holds the entire knowledge base locally as a git repo. An agent in the field, airgapped, or simply disconnected has full recall and keeps learning on its branch — then syncs when it reconnects. The network is for consensus, not for access.
local-first · works disconnected · syncs later
It's open source. Clone it, point your peers at it, and see.