SDTK-BRAIN

A standalone second-brain vault — your agent is the knowledge engine

SDTK-BRAIN is deliberately NOT a pipeline stage and NOT part of the sdtk-kit umbrella. It is a separate local-first tool: one vault folder holds immutable raw/ sources and a compiled markdown wiki/ knowledge layer, maintained by deterministic rails (ingest, compile, lint, search, graph viewer) while your AI agent — reading the vault's own CLAUDE.md contract — does the actual thinking. Everything is local: no LLM in the CLI, no network, no telemetry, no entitlement.

Standalone package (not in sdtk-kit)Local-first vaultNo LLM / no network / no telemetryObsidian-compatible markdown
sdtk-wiki — atlas + kanban
Real capture: local Atlas graph, node detail, shared Kanban board

When a solo founder reaches for this

You keep finding useful open-source repos, papers, and notes while building your product — and losing them. Drop each one into raw/inbox, run sdtk-brain ingest + compile, and they become linked knowledge pages (sources, comparisons, syntheses) your agent can search and cite next time you ask 'which orchestration library did we shortlist, and why?'. Open the same folder in Obsidian or your agent — it's all plain markdown.

Product truth

What this toolkit does today

These cards summarize the current shipped or internal surface exactly as it exists in the local source tree.

Two layers that never blur

raw/ holds immutable source documents (articles, papers, repos, notes, meetings) — the agent reads but never edits meaning there. wiki/ is the compiled knowledge layer (sources, concepts, entities, comparisons, syntheses) the agent maintains freely. .brain/ is CLI-owned machine state.

The agent is the answer engine

There is deliberately no `ask` command. The vault ships a CLAUDE.md operating contract; open the vault in your agent and ask there — grounded synthesis with page citations is the agent's job. The CLI provides only the deterministic rails.

Shares SDTK's viewer — and nothing else

sdtk-brain open builds a self-contained docs + graph viewer (the same guarded viewer SDTK-WIKI uses). If you also use SDTK on a project, sdtk-wiki's multi-workspace switcher can show your project wiki and your brain vault side by side. That is the only connection — the tool installs no SDTK skills and needs no other kit.

Current command and workflow surface

Core workflow

npm install -g sdtk-brain-kit
sdtk-brain init
sdtk-brain ingest raw/inbox
sdtk-brain compile --mode safe --apply
sdtk-brain search "<topic>"
sdtk-brain open

Maintenance

sdtk-brain lint
sdtk-brain maintain --mode safe
sdtk-brain discover --plan
sdtk-brain enrich --source github --mode review
sdtk-brain status

Runtime support

RuntimeProjectUserNotes
Clauden/an/aInstalls no skills for either runtime — the vault's own CLAUDE.md is the agent operating contract. Open the vault folder in your agent and it knows how to maintain it.
Codexn/an/aIdentical: no skills, no runtime coupling. Any agent (or Obsidian) that reads markdown can operate the vault.

Validated truth

  • Published as sdtk-brain-kit@0.1.0 and verified with a real npm registry install.
  • Dogfooded against the maintainers' real second-brain vault, unmodified: 111 raw files / 325 wiki pages read, lint produced 858 report-only findings, graph built 426 nodes / 1,581 edges.
  • Full core workflow (init -> ingest -> compile -> search -> open) verified end-to-end from the published package on a fresh vault.
  • Shared viewer/builder byte-identity with SDTK-WIKI is CI-guarded (sha drift test).

Boundaries and honest limits

  • Standalone by design: not in the sdtk-kit umbrella, installs no SDTK skills, links to no other kit.
  • No `ask` command, no LLM call, no network, no telemetry, no entitlement — the CLI is deterministic rails only.
  • raw/ content is never mutated; .brain/ machine state is CLI-owned; init refuses non-empty non-vault folders.

Source-of-truth docs

Package README

products/sdtk-brain/distribution/sdtk-brain-kit/README.md

Install, quick start, full command list, vault layout, boundaries.

Family links

Move across the SDTK suite without losing context

These product pages map the current suite boundaries so users can move from specification to coding to operations deliberately.