Replaces the rigid FSM after Intro/AskLanguage with a single LLM call per
turn that sees:
* persona (what's already known — slots not re-asked)
* recent 10 chat_log messages (history)
* top-5 kei-sage atoms relevant to user_text (graph-RAG, not embeddings)
* raw user_text
LLM returns JSON {slot_updates, response_text, done, focus} which drives
the next state + persona patch + reply. No embeddings, no vector store —
kei-sage's FTS5 + Obsidian-style atom graph is the retrieval layer.
New files:
* src/retrieval.rs (101 LOC) — retrieve_context(chat_log, topics,
chat_id, query, history_n, atoms_k) -> RetrievalContext
* src/conversational.rs (157 LOC) — conversational_step
(state, persona, context, text, extractor, lang) -> StepOutput
Modified:
* src/serve.rs::run_fsm — branch on state: Intro/AskLanguage still go
through legacy handle_step (jump-start); everything else routes to
conversational_step with retrieval context.
* src/lib.rs — module declarations.
Tests (5 new, 60 total passing):
* parses_well_formed_llm_response
* done_true_transitions_to_ready
* invalid_json_falls_back_gracefully
* retrieve_returns_empty_on_empty_stores
* retrieve_finds_seeded_data
Verify:
* cargo check -p kei-buddy: PASS
* cargo test -p kei-buddy --lib: 60/0 (was 55, +5)
Why graph-RAG instead of embeddings: kei-sage already in tree (atoms +
edges + BFS + PageRank + FTS5). Explicit edges (message → topic →
contact) beat opaque cosine similarity for personal-assistant memory
where relationships are typed. No sqlite-vec dep, no embedding cost.
NOT deployed yet — needs server rebuild.
|
||
|---|---|---|
| .. | ||
| _rust | ||
| templates | ||
| design-scrape.sh | ||
| figma-tokens.sh | ||
| frontend-inspect.sh | ||
| harden-base.sh | ||
| kei-ci-lint.sh | ||
| kei-docs-scaffold.sh | ||
| kei-doctor.sh | ||
| kei-sleep-queue.sh | ||
| kei-sleep-setup.sh | ||
| kei-sleep-sync.sh | ||
| live-preview.sh | ||
| log-ship.sh | ||
| MANIFEST.toml | ||
| metrics-scrape.sh | ||
| provision-hetzner.sh | ||
| provision-vultr.sh | ||
| README.md | ||
| screenshot-decode.sh | ||
| tomd.sh | ||
_primitives — first-class building blocks
_primitives/ holds standalone utilities that agents, hooks, and skills
(including /compose-solution) depend on. Unlike _blocks/ (behavioral
markdown) or _manifests/ (agent TOML), primitives are executable shell
programs installed at $HOME/.claude/agents/_primitives/ by install.sh.
Current primitives
| Primitive | Purpose | Invocation |
|---|---|---|
tomd.sh |
Universal non-native-format → markdown converter (PDF, DOCX, XLSX, PPTX, CSV, images, code). | ~/.claude/agents/_primitives/tomd.sh <file> |
tomd.sh is a first-class primitive. Universal non-native-format →
markdown converter with configurable cache directory
(KEISEI_TOMD_CACHE) and KeiSeiKit-style error tags ([tomd]).
Hook integration
hooks/tomd-preread.sh is a PreToolUse(Read) hook that auto-redirects
Claude to the converted markdown when a Read targets .docx / .doc / .xlsx / .pptx / .csv. Cached under $KEISEI_TOMD_CACHE (default
/tmp/keisei-tomd-cache).
/compose-solution discovery
Phase 3 prior-art sweep greps _primitives/ alongside _blocks/,
_manifests/, skills/, _bridges/, hooks/. If a user task involves
file-format parsing, the meta-composer surfaces tomd automatically —
reuse over rewrite (RULE "No Patching").