Browse / LLM wiki / nemo-knows
nemo-knows
Sizolity
nemo-knows turns raw source files into a maintained knowledge layer.
Open Sizolity/nemo-knows → Checked against the four rules.
Works with any agent
ships AGENTS.md, the cross-tool instruction file.
In its own words
A persistent, LLM-curated Markdown wiki with a local ingest pipeline, review gates, deterministic evaluation, and an optional browser console.
nemo-knows turns raw source files into a maintained knowledge layer. Instead of re-running retrieval and synthesis on every question, it pays the synthesis cost at ingest time, stores the result as interlinked Markdown, and lets future queries read from a wiki that already contains summaries, concepts, topic syntheses, citations, and known disagreements.
The pattern follows Andrej Karpathy's llm-wiki.md, with an opinionated file schema and a small Go toolchain around it.
What This Project Is
nemo-knows is both a knowledge-base convention and a local tool:
- A plain-file wiki under
wiki/, maintained by LLM agents according to
AGENTS.md.
- An immutable source area under
raw/, where original documents live. - A review buffer under
drafts/, where model output lands before it can be
accepted into the wiki.
- A Go CLI,
nemo, for draft generation, review, evaluation, linting, and
explicitly approved wiki writes.
- A local web console,
nemo-web, for browsing the wiki graph, importing
source-like pages, and starting reviewable ingest jobs.
The project deliberately uses Markdown, Git, and local files first. It does not require a database, vector store, search service, or always-on backend.
Quoting 19% of the README. Read the rest on GitHub ↗
Read from the repository's git tree on 2026-08-17.