Local AI Chat
MeetNote gives your team a smarter way to work with meeting history: on-device AI when you want local speed and privacy, plus custom provider mode when you want cloud-scale power.
One summary workflow for fast-moving teams
In high-frequency collaboration, the bottleneck is rarely access — it’s interpretation.
Teams repeatedly ask:
- What did we decide last week?
- What changed between meetings?
- Who owns this follow-up?
- What is still unresolved?
Without an intelligent layer, people scan long records by hand and decision latency grows.
Local AI Chat turns meeting context into a query-able knowledge layer, so teams move from “searching” to “acting” faster.
Model lineup (choose what fits your device)
MeetNote does not force a one-size-fits-all model. It gives you a practical range:
- SmolLM2 (135M) — ultra-light, fast, broad device compatibility
- Qwen 2.5 (0.5B) — balanced speed + quality
- Qwen 2.5 (1.5B) — stronger reasoning with larger footprint
- Phi-3.5-mini (3.8B) — deeper logic for advanced use cases
- Gemma-2-9B — richer synthesis and analysis
- Mistral NeMo (12.2B) — high-capacity multilingual / long-context tasks
So “Summarize this meeting” can run efficiently on lightweight phones, or scale up on high-memory devices.
Custom API Key support
MeetNote now supports custom AI providers in addition to local models.
In AI Settings, users can choose a meeting chat provider:
- Local (on-device)
- OpenAI-compatible
- Anthropic
- Gemini
- Ollama
For supported cloud providers, users can add a Base URL, choose a Model ID, and save their API key securely on-device.
Model lists can be refreshed directly from the provider, and responses can be streamed live as tokens arrive.
If a cloud/custom provider fails, MeetNote can automatically fall back to a local model (when enabled), so workflows keep moving.
How MeetNote adapts across device classes
MeetNote’s local AI path is designed for real-world hardware diversity:
- Downloads model files to device storage (GGUF flow)
- Uses native acceleration preferences where available (Vulkan on Android, Metal on iOS)
- Applies model-aware context and inference settings
- Falls back to compatible runtime/model behavior when needed
- Uses safety checks to avoid unstable heavy-model loads on constrained devices
In short: lighter models for reach, heavier models for depth — without making users guess what will work.
Built for summaries, recaps, and Minutes composition
Inside Meeting AI Chat, users can trigger quick workflows like:
- Summarize
- TO-DO List
- Budget Overview
Responses are grounded in:
- Current meeting transcript content
- Meeting metadata (title, date, status, owner, length, team, project)
- Focused context selection and transcript compression for long sessions
That grounding makes outputs practical for MoM workflows: cleaner drafts, clearer ownership, and faster preparation.
What teams get
- Faster context retrieval for decisions
Ask directly, get relevant answers quickly, and move from context lookup to execution. - Better continuity in high-frequency collaboration
Reconnect to prior discussions without repetitive re-alignment loops. - More useful outputs for busy stakeholders
Leaders and contributors can absorb key outcomes, risks, and next steps with less cognitive load. - Stronger follow-up quality
Clearer decision trails improve accountability and reduce ambiguity. - Lower friction between recorded and usable
Meeting data becomes an active planning asset, not a passive archive. - Better support for iterative planning
Track evolving decisions, compare reasoning, and carry forward unresolved issues with confidence.
Example use cases
- Product teams: Track scope shifts across planning meetings
- Operations teams: Extract recurring blockers and ownership patterns
- Client services: Summarize stakeholder concerns and confirm next steps
- Leadership: Get concise pre-read context before strategic reviews
