Add ask command for natural-language search across meeting notes

AGENTS.md

@@ -0,0 +1,50 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Commands
+
+```bash
+uv run pytest                        # run all tests
+uv run pytest -v                     # verbose
+uv run pytest -v -k test_search      # run a specific test module
+uv run pytest -v -k "TestRankMeetings::test_speaker_boost"  # single test
+uv run ruff check src/ tests/        # lint
+uv run ruff format src/ tests/       # auto-format
+```
+
+## Architecture
+
+**ownscribe** is a CLI tool for local meeting recording, transcription, and summarization. The main pipeline is: Record → Transcribe → Summarize → Output.
+
+### Plugin systems with abstract base classes
+
+Each stage has a base class in its subpackage and one or more implementations:
+
+- **Audio** (`audio/base.py`): `CoreAudioRecorder` (macOS, wraps a Swift binary in `swift/`) and `SoundDeviceRecorder` (cross-platform fallback). Selected in `pipeline.py:_create_recorder()`.
+- **Transcription** (`transcription/base.py`): `WhisperXTranscriber` (single impl). Data models (`Segment`, `Word`, `TranscriptResult`) live in `transcription/models.py`.
+- **Summarization** (`summarization/base.py`): `OllamaSummarizer` and `OpenAISummarizer`. Factory is `summarization/__init__.py:create_summarizer()` — used by both `pipeline.py` and `search.py`.
+- **Output** (`output/`): `markdown.py` and `json_output.py`, selected by `config.output.format`.
+
+### Key modules
+
+- **`cli.py`** — Click command group. Entry point: `ownscribe.cli:cli`. All subcommands (`ask`, `transcribe`, `summarize`, `devices`, `apps`, `config`, `cleanup`).
+- **`pipeline.py`** — Orchestrates the record → transcribe → summarize flow. Creates timestamped output dirs (`~/ownscribe/YYYY-MM-DD_HHMM_slug/`).
+- **`search.py`** — Two-stage LLM search over meeting notes. Stage 1 scores summaries for relevance, stage 2 synthesizes answers from full transcripts. Has keyword fallback and quote verification. Helper functions return data; only `ask()` calls `click.echo`.
+- **`config.py`** — Dataclass hierarchy (`Config` → `AudioConfig`, `TranscriptionConfig`, `SummarizationConfig`, etc.). Loaded from `~/.config/ownscribe/config.toml` with env var overrides (`HF_TOKEN`, `OLLAMA_HOST`).
+- **`summarization/prompts.py`** — Built-in prompt templates (meeting, lecture, brief) plus search prompts. Users can define custom templates in config TOML.
+
+### Testing conventions
+
+- Uses `pytest` with `pytest-httpserver` for mocking HTTP APIs (Ollama, OpenAI).
+- Shared fixtures in `conftest.py`: `sample_transcript`, `diarized_transcript`, `synthetic_wav`.
+- Tests use `FakeSummarizer` (in `test_search.py`) or `unittest.mock` for pipeline tests.
+- Markers: `@pytest.mark.hardware` (auto-skipped in CI), `@pytest.mark.macos` (auto-skipped on non-macOS).
+- When mocking the shared summarizer factory in pipeline tests, patch `ownscribe.pipeline.create_summarizer` (it's imported at module level).
+
+## Style
+
+- Python 3.12+. Ruff with line-length 120.
+- `from __future__ import annotations` in all modules.
+- Lazy imports for heavy dependencies (whisperx, ollama, openai) — imported inside functions, not at module level.
+- Helper functions return data; orchestrator functions (`ask()`, `run_pipeline()`) handle all `click.echo` output.

README.md

@@ -32,6 +32,7 @@ All audio, transcripts, and summaries remain local.
 - **Pipeline progress** — live checklist showing transcription, diarization sub-steps, and summarization progress
 - **Local LLM summarization** — structured meeting notes via Ollama, LM Studio, or any OpenAI-compatible server
 - **Summarization templates** — built-in presets for meetings, lectures, and quick briefs; define your own in config
+- **Ask your meetings** — ask natural-language questions across all your meeting notes; uses a two-stage LLM pipeline with keyword fallback
 - **One command** — just run `ownscribe`, press Ctrl+C when done, get transcript + summary
 
 ## Requirements
@@ -118,10 +119,27 @@ ownscribe devices                  # list audio devices (uses native CoreAudio w
 ownscribe apps                     # list running apps with PIDs for use with --pid
 ownscribe transcribe recording.wav # transcribe an existing audio file
 ownscribe summarize transcript.md  # summarize an existing transcript
+ownscribe ask "question"           # search your meetings with a natural-language question
 ownscribe config                   # open config file in $EDITOR
 ownscribe cleanup                  # remove ownscribe data from disk
 ```
 
+### Searching Meeting Notes
+
+Use `ask` to search across all your meeting notes with natural-language questions:
+
+```bash
+ownscribe ask "What did Anna say about the deadline?"
+ownscribe ask "budget decisions" --since 2026-01-01
+ownscribe ask "action items from last week" --limit 5
+```
+
+This runs a two-stage pipeline:
+1. **Find** — sends meeting summaries to the LLM to identify which meetings are relevant
+2. **Answer** — sends the full transcripts of relevant meetings to the LLM to produce an answer with quotes
+
+If the LLM finds no relevant meetings, a keyword fallback searches summaries and transcripts directly.
+
 ## Configuration
 
 Config is stored at `~/.config/ownscribe/config.toml`. Run `ownscribe config` to create and edit it.
@@ -149,6 +167,7 @@ backend = "ollama"        # "ollama" or "openai"
 model = "mistral"
 host = "http://localhost:11434"
 # template = "meeting"    # "meeting", "lecture", "brief", or a custom name
+# context_size = 0        # 0 = auto-detect from model; set manually for OpenAI-compatible backends
 
 # Custom templates (optional):
 # [templates.my-standup]

src/ownscribe/cli.py

@@ -97,6 +97,19 @@ def cli(
         run_pipeline(config)
 
 
+@cli.command()
+@click.argument("question")
+@click.option("--since", default=None, help="Only search meetings after this date (YYYY-MM-DD).")
+@click.option("--limit", default=None, type=int, help="Max number of recent meetings to search.")
+@click.pass_context
+def ask(ctx: click.Context, question: str, since: str | None, limit: int | None) -> None:
+    """Ask a question across your meeting notes."""
+    config = ctx.obj["config"]
+    from ownscribe.search import ask as run_ask
+
+    run_ask(config, question, since=since, limit=limit)
+
+
 @cli.command()
 def devices() -> None:
     """List available audio input devices."""

src/ownscribe/config.py

@@ -35,6 +35,7 @@
 model = "mistral"         # model name
 host = "http://localhost:11434"  # ollama: :11434, LM Studio: :1234
 # template = "meeting"    # built-in: "meeting", "lecture", or "brief"
+# context_size = 0        # 0 = auto-detect from model; set manually for OpenAI-compatible backends
 
 # Custom templates (optional):
 # [templates.my-notes]
@@ -79,6 +80,7 @@ class SummarizationConfig:
     model: str = "mistral"
     host: str = "http://localhost:11434"
     template: str = ""
+    context_size: int = 0
 
 
 @dataclass

src/ownscribe/pipeline.py

@@ -17,6 +17,7 @@
 
 from ownscribe.config import Config
 from ownscribe.progress import PipelineProgress, Spinner
+from ownscribe.summarization import create_summarizer
 
 # A standard WAV file header (RIFF + fmt + data chunk header) is 44 bytes.
 # Files at or below this size contain no audio frames.
@@ -87,15 +88,6 @@ def _create_transcriber(config: Config, progress=None):
     return WhisperXTranscriber(config.transcription, diar_config, progress=progress)
 
 
-def _create_summarizer(config: Config):
-    """Create the appropriate summarizer based on config."""
-    if config.summarization.backend == "openai":
-        from ownscribe.summarization.openai_summarizer import OpenAISummarizer
-        return OpenAISummarizer(config.summarization, config.templates)
-    else:
-        from ownscribe.summarization.ollama_summarizer import OllamaSummarizer
-        return OllamaSummarizer(config.summarization, config.templates)
-
 
 def _format_output(config: Config, transcript_result, summary_text: str | None = None) -> tuple[str, str | None]:
     """Format transcript and optional summary. Returns (transcript_str, summary_str)."""
@@ -233,7 +225,7 @@ def run_summarize(config: Config, transcript_file: str) -> None:
     """Summarize a transcript file."""
     transcript_text = Path(transcript_file).read_text()
 
-    summarizer = _create_summarizer(config)
+    summarizer = create_summarizer(config)
     if not summarizer.is_available():
         click.echo(
             f"Error: {config.summarization.backend} is not reachable at {config.summarization.host}. "
@@ -295,7 +287,7 @@ def _do_transcribe_and_summarize(
 
         # 3. Summarize
         if sum_enabled:
-            summarizer = _create_summarizer(config)
+            summarizer = create_summarizer(config)
             if not summarizer.is_available():
                 click.echo(
                     f"\nWarning: {config.summarization.backend} is not reachable "