Lacan Knowledge OS

A corpus-grounded research environment for Lacanian psychoanalysis. Evidence first. Interpretation second. Fabrication never.

English · 中文 · 日本語 · Français · Deutsch · Italiano

The one rule this system is built around

Every claim must land on a passage number.

If the corpus cannot support a question, the system abstains — it does not answer from model knowledge. An abstention is a result, not an error.

⚠️ Rights notice — read before installing the corpus

The engine is open source. The corpus is not. Two different things, two different rules:

	License / status
Engine source code (this repository)	Apache-2.0 — use, modify, redistribute, build on it
Reference corpus (companion repo)	third-party copyrighted texts. Publicly readable for research and study only; public visibility is not a licence — not licensed for commercial use, redistribution or re-serving.
Public-domain demo corpus (demo-corpus/)	public domain (Falret 1890 · Binet 1892 · Janet 1909) — freely redistributable

In plain terms. The corpus repository contains French working transcriptions of Lacan's seminars, a text extraction from the Seuil print edition, and a community Chinese translation project. None of these rights belong to this project. It is published openly so that researchers can obtain it, and public availability is not a licence: it does not permit commercial use, redistribution, mirroring, repackaging, or serving the text to others. If you are unsure whether your intended use is lawful, that question is yours to answer — this project cannot answer it for you. The full analysis is in RIGHTS.md.

If you need something you can publish, share or ship commercially, use the engine with a corpus you have the right to use — the public-domain demo, your own texts, or a licensed edition. That path is fully supported: CORPUS.md · docs/DEMO_CORPUS.md.

Related MCP server: docs-index

What it is

Lacan Knowledge OS turns a corpus of seminar and écrits text into a citable evidence store, and puts a frozen scholarly core between your question and any answer.

The chain you can always audit

What it refuses to do

These refusals are the design, not limitations:

Get started

Are you an AI agent? Run this first and follow what it prints — it decides whether research can run at all:

python3 tools/ensure_corpus.py --status      # exit 0 = corpus ready · exit 3 = tell the user
python3 tools/ensure_corpus.py --json        # machine-readable status

1 · Clone the engine

git clone https://github.com/YanKaFei/Lacan-Knowledge-OS.git
cd Lacan-Knowledge-OS
python3 -m workspace_ui.server.cli --port 3090
# → http://127.0.0.1:3090/help    (13-page Help Centre — fully usable with no corpus)

2 · Give it a corpus — one of three paths

Path A in full. The companion repository is public, so the pack downloads without a token — but it is third-party copyrighted text: research and study only (RIGHTS.md). It holds a corpus pack: one hash-manifested archive that tools/fetch-corpus.py verifies file by file before installing — a partial or altered download cannot silently corrupt the corpus.

curl -sLO https://github.com/YanKaFei/Lacan-Knowledge-OS-corpus/releases/download/corpus-v1/corpus-pack-v1.tar.gz
curl -sLO https://raw.githubusercontent.com/YanKaFei/Lacan-Knowledge-OS-corpus/main/corpus-pack-v1.manifest.json
python3 tools/fetch-corpus.py --pack corpus-pack-v1.tar.gz \
    --manifest corpus-pack-v1.manifest.json --into .
python3 _scripts/_tools/core_freeze.py --verify      # → SCHOLARLY_CORE_READY
python3 -m workspace_ui.server.cli --port 3090       # research now answers

Path A is a plain public download and is for research use only — no commercial use, no redistribution, no re-serving (RIGHTS.md). Paths B and C need no access at all. Pack mechanics (handing a corpus to a colleague without publishing it): docs/CORPUS_PACK.md.

3 · Verify the contract (this is the point of the project)

python3 _scripts/_tools/core_freeze.py --verify      # 39 scholarly components
python3 _scripts/_tools/freeze_lineage.py --verify   # zero semantic drift across 7 segments
python3 _scripts/_tools/build_i18n.py --check        # UI copy ↔ dictionary ↔ call sites
python3 _scripts/_tools/build_help.py --check        # Help links, anchors, 0 fiction
bash _scripts/run_all_tests.sh                       # 146 suites · 78 validators

Requirements: Python 3.9+ (standard library covers the core path) and a modern browser. No bundler, no build step — index.html loads ES modules directly. Vector retrieval and real LLM synthesis are optional extras (docs/EMBEDDING_PROVIDER.md).

How to use it

1. Ask — one question per run.

2. Read the state — VALIDATED, VALIDATED_WITH_QUALIFICATIONS, PARTIALLY_SUPPORTED, VALIDATION_FAILED, INSUFFICIENT_EVIDENCE, ABSTAINED. Read it literally.

3. Verify — click a citation chip; the Evidence Inspector shows the quoted span, its context window and the source chain. The answer is what the core concluded; the inspector is what the corpus actually says.

4. Go deeper — open the same passage in Explore and read around it.

5. Keep it — Add to Project, or Save to Obsidian.

The interface

Every control name inside Help is rendered from the real interface, and every functional claim is machine-checked: 42/42 claims verified · 0 documentation fiction · 0 broken links.

Use it from DSH / any MCP client

This repository is a DeepSeek Harness plugin target: it ships an MCP server, a portable composition row, an installable bundle and an idempotent installer.

dsh plugin --profile web add github:YanKaFei/Lacan-Knowledge-OS
python3 tools/install-dsh-row.py --profile web     # bind the server path for your clone
python3 _scripts/_tools/lacan-kb-mcp               # or run the MCP server standalone

Tools appear as mcp__lacan-kb__search_passages, …get_passage, …get_context, …resolve_entity, …list_concepts, …search_terminology, …compare_concepts, …find_relation, …list_seminars, …get_sources — 10 tools, protocol 2025-11-25. Proven with DSH's own MCP SDK (_data/mcp/DSH_CLIENT_PROOF.json, 14/14 checks). See docs/DSH_PLUGIN.md.

Advantages and disadvantages

Verifiability over fluency. Every claim ties to a passage id; you can audit the chain by hand.

Honest failure. Abstention, INSUFFICIENT_EVIDENCE and explicit unavailability are first-class outputs. No hallucinated filler.

An executable contract. The 39-component freeze and the seven-segment lineage are checked by code, not by a promise in a document.

Offline-first. Offline / Mock produces deterministic answers with no credentials, no network.

One contract, many clients. The same MCP surface serves the web UI, DSH, and any editor with an MCP client.

Auditable history. Every run is an immutable snapshot; re-asking creates a new item instead of overwriting the old one.

Corpus is separable. Engine public (Apache-2.0), corpus third-party with its own boundary or bring your own — the two never mix in one repository.

It needs a corpus you have the right to use. Out of the box it shows its interface, Help Centre and contracts; it answers only once a corpus is installed. The reference pack is public to download but research-use-only material.

The reference corpus is research material, not a product asset. No commercial use, no redistribution — see RIGHTS.md.

It is opinionated about scope. It answers about a corpus. Questions about publication history, attendees or dates abstain, because the corpus does not carry that metadata.

No quality score. There is no “confidence %” anywhere; interpretation quality remains your judgement.

Heavy at laptop scale. The reference corpus is ~2.5 GB on disk including derived indexes.

Vector retrieval is optional. Without the embedding extras, retrieval falls back to lexical matching (the UI says which is active).

Real-LLM synthesis is a single blocking call. Long runs take 45–60 s; the UI shows real elapsed time, and “stop waiting” does not cancel the backend job.

Fast product layer, frozen scholarly layer. Convenience features change often; changing the semantics requires a formal Core Change Request.

Presupposes familiarity with the domain. It will not teach you Lacan.

Documentation map

Status

• Scholarly core: frozen v1 — SCHOLARLY_CORE_READY, 39 hash-pinned components, semantic drift 0.

• Product layer: complete for daily use; last acceptance run 20/20 blocking gate items, 146 regression suites / 78 validators / 0 failed / 0 skipped.

• Distribution: engine public (Apache-2.0) · reference corpus public, research use only · public-domain demo corpus in progress.

• Roadmap: docs/ROADMAP.md.

Code: Apache-2.0 · Source texts: not distributed, not licensed for commercial use (RIGHTS.md)

If this system saves you a week of citation checking, it did its job.