The repository README is the user-facing starting point. The documents in this directory have narrower responsibilities so installation, extension contracts, and future plans do not become mixed together.
-
Automatic context and caching: local model limits, remote provider behavior and real OpenAI, Claude and Gemini cache checks.
-
Remote model presets: shared AI Engine and Live Assistant selectors, Custom endpoints and provider compatibility.
-
Bonsai 8B validation: memory measurements and quality limitations for smaller GPUs.
-
Voice sampling validation: interior samples, embedding reuse and the ITV recognition checks.
-
Performance demo: native Bonsai on an 8 GB GPU, measured cold-load and follow-up timings, and the recording methodology.
-
Automatic LLM setup: hardware profiles, native Bonsai, automatic context sizing and KV reuse between notes and assistant questions.
-
Automatic audio setup: transcription/diarization profiles, model memory management and long-job activity reporting.
-
Ollama: discovery, model selection, context and assistant streaming.
-
Linux CUDA: private-library detection, repair and CPU fallback.
-
Safe updates: stable releases, backups and preserved settings.
-
Meeting library: transcript search, tags, reusable assistant actions, and their privacy and scope boundaries.
-
Audio capture: microphone + system recording and setup on Windows, Linux, and macOS.
-
Architecture: process boundaries, workers, storage, and the provider registry.
-
Plugin API: how users discover, install, enable, and remove plugins, plus the stable hook reference.
-
Plugin and provider development: package authors, provider contracts, model registration, permissions, tests, and listing submissions.
-
Privacy: local data, network access, credentials, and threat model.
-
Webhooks: outbound event contract, delivery guarantees, signatures, Live agents, security, and rules for future changes.
-
Live AI Assistant: native real-time assistant, independent worker, settings, persistence, API, and resource limits.
-
Local MCP server: read-only desktop-client integration, lifecycle, tools, configuration, and security boundary.
- Roadmap: completed foundations and planned work.
- RAG and MCP analysis: historical design research behind the implemented retrieval and MCP boundaries.
- AI engine research: model/runtime evaluation notes; not a user guide or stable API.
The root product specification records the original product brief. When it differs from the application or current documentation, the README, source code, and the public contracts above are authoritative.