WIKI / ENGINE View repository ↗
PUBLIC ALPHA / LOCAL-FIRSTRESEARCH SIGNAL DESK № 01

A research intake engine for Claude Code, Codex, and Obsidian

From saved sources to traceable signal.

Preserve the source. Read what matters. Connect recurring ideas only when the evidence earns it.

Plain Markdown + JSON · no hosted account · no runtime telemetry

Illustrative structureSource-bound
Capture
🎬 Video transcriptuntrusted intake
📄 Research articleuntrusted intake
⌘ Repository notesuntrusted intake
Signal

What matters—and why

  • Bottom line
  • Useful evidence
  • Action or disposition
  • Uncertainty
Exact Source ID + digest
Source preservedInterpretation labelledSynthesis provisional
01 Capture02 Signal03 Connect

Saved is not understood.

A folder full of links is still a backlog.

Capture tools preserve material. Generic summaries compress it. Automatic graphs can generate more structure than you have time to verify. Wiki Engine inserts a human reading layer before long-term knowledge.

Less backlog.
More usable research.

Four value lanes, one narrow promise: return attention without losing where the interpretation came from.

Capture

Keep exact source identity.

Admit one local file into content-addressed storage. Visible notes remain readable while exact payload bytes stay hidden from rendered Markdown.

Signal

Read the useful layer first.

Publish one concise, source-bound reading note instead of exploding every capture into a dozen speculative concept pages.

Connect

Make synthesis earn its life.

A Wiki page requires multiple distinct sources and stays provisional. Connection is an outcome—not a default animation.

Trace

Know what produced the claim.

Source IDs, hashes, records, receipts, reading state, and read-only Review make correction possible. Valid structure supports review; it does not certify truth.

Not another
generic summary.

A Signal is the one-source reading surface. Choose a lens to see the questions it is designed to answer.

SIGNAL / ONE SOURCEINSIGHT
source_idsrc-…source_sha256bound

BOTTOM LINE

What changed my understanding?

State the useful claim in plain language, then explain why it matters to the reader’s research intent.

Bound to exact Source identity—not memory or a loose URL.

🎬 Transcript first

Inspect the screen only when meaning depends on it.

Alpha.1 can produce a bounded route plan. It defaults to transcript evidence, asks for exact cue frames when the speaker points at the screen, and reserves broad visual review for visual-dominant material.

Acquisition, rights-safe downloading, and canonical frame Evidence remain under development. A link in a description stays uninspected until separately opened and admitted.

  1. 01
    Transcript-only

    Default for speech-led material.

  2. 02
    Targeted frames

    Exact cues for demos and deictic language.

  3. 03
    Full visual review

    Only where visuals carry the argument.

Every layer knows
what it can prove.

Wiki Engine separates deterministic custody from fallible interpretation.

  1. 📦
    Exact

    Source

    What bytes were admitted.

    Kernel-owned
  2. Fallible

    Signal

    What one source may mean.

    Human/model draft
  3. 🔗
    Provisional

    Wiki

    What distinct sources suggest together.

    Cross-source synthesis
  4. 🧭
    Diagnostic

    Review

    What needs attention next.

    Read-only Alpha.1 queue
🔒

Local by default

No hosted account, daemon, runtime telemetry, embedding database, or core network request.

🧱

Bounded mechanics

No-follow inputs, scan and file caps, path validation, a writer lock, and recoverable multi-file transactions.

No automatic truth

Captured instructions cannot authorize commands, browsing, credential use, publication, or authority changes.

Demonstrated core.
Visible limits.

19 / 19offline Alpha.1 tests
6passing public CI matrix jobs
0core network calls
MITlocal files you own

Repository-authored evidence, including one retained public CI run—not independent field validation. External users, semantic-quality evaluation, Linux accessibility, stable migration, backup/restore, and a long-duration soak remain future gates.

MASTER RESEARCH REPORT / 01

Karpathy showed how knowledge can compound.

Wiki Engine asks how that compounding can remain inspectable, reversible, and honest.

Read the attributed lineage from Vannevar Bush’s Memex through Andrej Karpathy’s influential LLM Wiki formulation, today’s community implementations, and Wiki Engine’s evidence-aware extension.

Read the master report

🧠 Second brain bridge

Feed the brain.
Do not become all of it.

Wiki Engine should own research custody and provisional interpretation. A second brain should own personal meaning, projects, decisions, commitments, and durable promotion.

This boundary is a thesis, not an implemented integration. No keyword, proximity, or model confidence should silently promote research into personal authority.

One source.
One Signal.
Then decide.

Requires Python 3.10+ on macOS or Linux.

TERMINAL / QUICK START
git clone https://github.com/youkistudios/wiki-engine
cd wiki-engine
python3 -m venv .venv
. .venv/bin/activate
pip install -e .

wiki-engine --vault ~/Desktop/My-Wiki init
wiki-engine --vault ~/Desktop/My-Wiki scan

SAVE A SOURCE. KNOW WHY IT MATTERS.

Keep the signal.
Keep the source.

Explore the repository ↗ Read the product thesis