Add CLAUDE.md with architecture and development guidance

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

CLAUDE.md

@@ -0,0 +1,74 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Running the App
+
+```bash
+# Development
+./venv/bin/uvicorn app:app --host 0.0.0.0 --port 8080 --reload
+
+# Production (via systemd)
+sudo systemctl restart jarvischat
+
+# Direct run
+./venv/bin/python app.py
+```
+
+## Dependencies
+
+```bash
+./venv/bin/pip install -r requirements.txt
+# Also requires: psutil jinja2 python-multipart (not in requirements.txt)
+```
+
+## Architecture
+
+Single-file FastAPI backend (`app.py`) + single-template frontend (`templates/index.html`). No build step. SQLite database auto-created at `jarvischat.db` on first run.
+
+### Request Flow: `/api/chat`
+
+1. User message saved to DB → conversation created if new
+2. `build_system_prompt()` assembles: profile + FTS5 memory search results + preset prompt
+3. Streamed to Ollama (`/api/chat`, `stream: true`, `logprobs: true`) via SSE
+4. **Auto web search trigger**: if perplexity > 15.0 OR response matches `REFUSAL_PATTERNS`, re-queries Ollama with SearXNG results prepended to system prompt
+5. Final response saved to DB; SSE `done` event sent with perplexity + tokens/sec
+
+### Request Flow: `/api/search` (explicit search)
+
+Bypasses perplexity/refusal detection entirely — queries SearXNG directly then asks Ollama to summarize with results as system context.
+
+### Memory System
+
+FTS5 virtual table (`memories`) in SQLite. `search_memories()` uses BM25 ranking. `process_remember_command()` intercepts "remember that..." / "forget about..." before the message reaches Ollama and returns a confirmation string. Topic auto-detection via keyword matching in `detect_topic()`.
+
+### Key Constants (top of `app.py`)
+
+- `OLLAMA_BASE` — `http://localhost:11434`
+- `SEARXNG_BASE` — `http://localhost:8888`
+- `PERPLEXITY_THRESHOLD` — `15.0` (controls auto-search sensitivity)
+- `DEFAULT_MODEL` — `llama3.1:latest`
+
+### External Services
+
+- **Ollama** — required, must be running on port 11434
+- **SearXNG** — optional, port 8888; `GET /api/search/status` probes availability
+- **wttr.in** — weather shortcut in `query_searxng()`, bypasses SearXNG for weather queries
+- **rocm-smi** — AMD GPU stats via subprocess; gracefully degrades if not available
+
+### Database
+
+`get_db()` opens a new connection per request (no connection pool). `init_db()` runs at startup via the FastAPI `lifespan` handler. The `profile` table uses a singleton row (`id = 1`). Default settings are seeded but never overwritten by `init_db()`.
+
+### SSE Protocol
+
+All streaming endpoints yield `data: {json}\n\n`. Key event shapes:
+- `{token, conversation_id}` — streaming token
+- `{searching: true}` — web search triggered
+- `{search_results: N}` — N results retrieved
+- `{done: true, perplexity, tokens_per_sec, searched?}` — terminal event
+- `{error: "..."}` — error event
+
+### Deployment
+
+Runs as systemd service under user `jarvischat`, working directory `/opt/jarvischat`. Logs via syslog (`journalctl -u jarvischat`).