Notely

AI Setup ​

Configure LLM provider models, API tokens, and local vector index settings inside AI → AI Settings.

1. Text Generation Providers ​

Notely connects to cloud and custom LLM providers using the Vercel AI SDK:

  • Google Gemini: Requires a Gemini API key. Default provider (gemini-2.0-flash), recommended for rich tool calling.
  • Groq: Requires a Groq API key (supports models like llama-3.3-70b-versatile, llama-3.1-8b-instant, deepseek-r1-distill-llama-70b).
  • OpenAI / OpenAI-Compatible: Connect to OpenAI (gpt-4o, gpt-4o-mini) or custom compatible endpoints by setting an API Key and custom Base URL.
  • Connection Diagnostics: Click the Test button next to any configured provider in AI Settings to verify connection status.

2. Embedding Index Setup ​

Vector embeddings enable Semantic Search and Context Retrieval:

  • Local BGE Model (Recommended): Runs entirely offline inside your app. Downloads a lightweight BGE-small-en-v1.5 ONNX model (~130MB) into %AppData%/notely/ai-model/ and runs vector calculations locally via onnxruntime-node.
  • HuggingFace API: Runs cloud-based embeddings using an API key token.

3. Knowledge Graph Engine ​

Relationship extraction and entity graph generation:

  • GLiNER2-Relex ONNX (Always Local): The Knowledge Graph uses a dedicated gliner2-multi-v1-onnx model running locally via ONNX Runtime. This is separate from your text generation provider — it runs entirely offline with no API key required and is not user-configurable.
  • Model Location: Downloaded automatically to %AppData%/notely/models/gliner2-relex/ on first graph build.
  • Confidence Threshold: Adjustable in AI Settings (graphConfidence, default 0.45–0.60). Higher values produce fewer but more precise relationships.

4. Speech-to-Text (STT) Engine Setup ​

Configure meeting and audio transcription inside AI → AI Settings → Speech-to-Text:

  • Local ONNX Whisper (Offline): Download whisper-tiny.en (~40MB), whisper-base.en (~140MB), or whisper-small (~460MB) for private, zero-latency on-device transcription via WebAssembly/ONNX Runtime.
  • Cloud Whisper (Groq / OpenAI): Sub-second cloud transcription using your saved Groq or OpenAI API key.
  • Audio Capture Preference: Set default recording mode to Meeting: Mic + System Audio, Microphone Only, or System Audio Only.

5. SQLite Database Locality ​

All AI databases are workspace-scoped and stored inside the hidden {workspace}/.notes-app/ folder to keep your data local and portable:

  1. ai-embeddings.db: Stores chunk text, line mappings, content hashes, and indexing queues.
  2. ai-graph.db: Stores extracted entity nodes and relationships.
  3. ai-memory.db: Stores conversation sessions, message logs, and persona configurations.
  4. ai-logs.db: Stores 5-stage execution traces, flow telemetry logs (FlowTracker), and prompt tracking payloads (LogDB).

PRAGMA journal_mode = WAL and synchronous = NORMAL are enabled across all databases for high performance without write blocks.