b61b17ea7b
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.
|
||
|---|---|---|
| .. | ||
| src | ||
| tests | ||
| Cargo.toml | ||
| README.md | ||
kei-buddy
Maturity: concept / scaffold — no business logic yet.
Purpose
kei-buddy is the runtime crate that composes existing KeiSeiKit
primitives (kei-pet, kei-memory-sqlite, kei-cortex,
kei-notify-telegram) into a personal-assistant Telegram bot called
KeiBuddy.
On first contact the bot walks the user through an 11-state onboarding flow: name, tone, interests, hobbies, per-topic decomposition (specifics → now-or-later → research preference → source selection), and digest schedule. After onboarding the bot enters ongoing conversation mode, drawing on the stored persona and memory.
This crate provides the state-machine enum and skeleton driver. The
onboarding FSM is ported from
keisei-marketplace/src/lib/keibuddy/chat-onboard.ts.
Status
Scaffold only. The OnboardState enum and TransitionInput struct are
defined. All transition logic is stubbed (next() returns self.clone()).
The binary entry point prints a placeholder message and exits 0.
Running
Environment variables
| Variable | Required | Default | Description |
|---|---|---|---|
TELEGRAM_BOT_TOKEN |
yes (serve) | — | Bot token from @BotFather |
TELEGRAM_WEBHOOK_SECRET |
yes (serve) | — | Secret token for webhook verification |
KEI_BUDDY_PORT |
no | 8080 |
HTTP port to bind |
KEI_BUDDY_DB_PATH |
no | ./kei-buddy.db |
SQLite database path |
OPENAI_API_KEY |
no | — | Enables OpenAiExtractor when set (requires extractor-openai feature) |
Subcommands
# Apply schema (idempotent; run once before first serve)
kei-buddy migrate
# Register the webhook URL with Telegram
kei-buddy webhook-set https://your-domain.com/webhook
# Start the HTTP server
kei-buddy serve
# Remove the registered webhook (revert to polling)
kei-buddy webhook-delete
Example systemd unit
[Unit]
Description=KeiBuddy Telegram bot
After=network.target
[Service]
EnvironmentFile=/etc/kei-buddy/env
ExecStart=/usr/local/bin/kei-buddy serve
Restart=on-failure
User=keisei
[Install]
WantedBy=multi-user.target
Roadmap
- OpenAiExtractor wiring — pass real OPENAI_API_KEY to OpenAiExtractor in serve.rs when feature enabled.
- Persona binding — read persona manifest via
kei-pet; apply tone overlay to outgoing replies. - Digest scheduling — wire
kei-cron-schedulerfor morning/evening digest delivery.