2dph is a single embedded knowledge graph (LadybugDB = Kuzu successor) with native HNSW vector + BM25 full-text indexes, built from markdown, compose files, ssh config, docker state, and git history. Search is deduction: confirmed facts first, supporting info second, web-search as the independent second source when the local graph cannot confirm.

Architecture

graph TB
    subgraph corpus["Corpus"]
        OPS["ops stack<br/>chat · onlyoffice · npm · observability · ai-bot"]
        SSH["~/.ssh/config"]
        CV["portfolio yaml"]
        GH["git history · authors"]
    end

    subgraph dph["2dph tools"]
        EX["bin/facts/extract<br/>2-source pairing"]
        AU["bin/facts/audit<br/>confidence + staleness"]
        IDX["bin/kb/index<br/>chunk + embed"]
        MD["bin/md/import<br/>mistune leaves"]
        SR["bin/kb/search<br/>deduction + --hop"]
    end

    subgraph store["Ladybug var/kb.lbug"]
        FACTS["facts root<br/>confirmed"]
        INFO["info root<br/>narrative"]
        VEC["HNSW cosine"]
        FTS["BM25 FTS"]
        GR["File→HAS_VERSION→Commit→AUTHORED→Person"]
    end

    subgraph ai["AI"]
        M2V["model2vec<br/>potion-multilingual-128M"]
    end

    subgraph ext["External"]
        WS["web-search skill<br/>(2nd independent source)"]
    end

    OPS --> EX
    SSH --> EX
    CV --> EX
    GH --> EX
    EX --> FACTS
    EX --> INFO
    MD --> GR
    MD --> INFO
    IDX --> M2V
    IDX --> VEC
    IDX --> FTS
    IDX --> FACTS
    SR --> FACTS
    SR --> INFO
    SR --> VEC
    SR --> FTS
    SR --> WS
    AU --> FACTS

The method

Every assertion is Who / What / How / Where / When + evidence + confidence, mirroring the detective detective skill: ≥2 independent sources confirm a fact; conflicting sources or a single source → hypothesis → (not confirmed).

root	meaning	used for answers
`facts`	assertions backed by ≥2 sources (`confirmed`)	yes, with evidence links
`info`	descriptive/narrative leafs (how-tos, notes)	context only, marked `(not confirmed)`

Deduction search

bin/kb/search "Matrix federation over HTTPS"      # facts → info → web-search
bin/kb/search "what runs on arc-2" --hop 1        # walk graph edges
bin/kb/search "where is cs-lexicon" --json | yq '.'  # YAML by default
bin/kb/get <id> --body                            # full chunk on demand
bin/kb/stats                                      # index health
bin/kb/eval                                       # recall@5 gate

Mail is a first-class corpus (retrievable through the same search):

bin/mail/sync.go --source onlyoffice,gmail --workers 8 --out var/mail  # raw sync (Go)
bin/mail/import --from-raw var/mail                                     # JSON → markdown
bin/mail/index_mail                                                     # rebuild brain incl. mail
bin/kb/search "Mietwagen Nürnberg invoice"                              # now answers from mail

Storage

LadybugDB — single var/kb.lbug, Cypher property graph, HNSW + BM25 in one engine, embedded (no server), ACID, read-only-safe for concurrent readers. Never DROP INDEX FTS/VECTOR on Ladybug 0.19: DROP leaves ghost catalog tables (_0_Leaf_vec_UPPER) so recreate fails while SHOW_INDEXES omits HNSW. Fresh indexes = delete var/kb.lbug + bin/kb/index --rebuild. Use ensure_indexes() after upserts.
model2vec — potion-multilingual-128M static embeddings (256-dim), CPU-fast, deterministic, no Ollama runtime dependency.
facts and info split semantically by root column but written inside the same transaction.

Tooling conventions

bin/{subject}/{method} — self-describing: shebang on line 1, usage comment from line 2. bash + python primary; golang via the Go shebang when a compiled helper is right. YAML default output, --json for machines. Everything that touches network/db is read-only, throttled, cached. Tests gate every commit.

Development

uv venv .venv                                  # Python 3.12, uv-managed
uv pip install -r requirements.lock.txt        # pinned toolchain
bin/facts/audit self                           # lexicon consistency gate
go test ./... && python -m unittest discover -s bin/tools -t .

Docker (optional, cached model + var volumes):

docker compose run --rm brain index            # (re)index corpus
docker compose run --rm brain search "query"   # one-shot query
docker compose run --rm brain serve            # async Go HTTP server
docker compose up brain-watch                  # auto re-index on change

go-second-brain — the earlier Neo4j + Qdrant + Matrix RAG brain
agent-skills — upstream skills (web-search, db-yaml, …) that 2dph integrates
detective method — the two-source method

See PLAN.md for decisions, execution status, and v2 open questions.

README.md

2dph — deductionphile