Savvy

Product Introduction

1. Definition: Savvy is a local-first, source-grounded macOS meeting assistant application. Technically, it is a native desktop application built with Tauri (Rust backend, React frontend) designed for Apple Silicon Macs running macOS 13+.
2. Core Value Proposition: Savvy exists to provide AI-powered, real-time meeting guidance that is directly sourced from a user's private documents, ensuring relevance and confidentiality. Its primary value is delivering context-aware assistance without relying on the open web, prioritizing user privacy and data sovereignty.

Main Features

1. Local-First Document Indexing: Savvy operates by pointing at a local folder containing your documents (e.g., client briefs, project notes, contracts). It performs read-only scanning, extracts text, chunks the content, and builds a private, versioned SQLite index on your Mac. This process ensures your source documents are never copied or uploaded wholesale.
2. Source-Grounded Live Guidance: During a live meeting, Savvy transcribes audio via a third-party provider (Deepgram or AssemblyAI) but only activates its AI recommendation engine for three specific triggers: when a question is asked that your brief can answer, when a conversation touches a predefined "red line," or when you manually request "Advice." Every suggestion it provides includes citations back to the source document.
3. Privacy-Conscious Data Flow: Savvy enforces a strict data-handling policy. Documents, indexes, and meeting transcripts are stored locally with owner-only file permissions (0700, 0600). Only minimal data—selected excerpts from your brief and recent transcript turns—are sent to an AI model via a local CLI tool (codex or claude) that you control. Meeting audio is streamed for transcription but with data-sharing opt-outs enabled.

Problems Solved

1. Pain Point: Professionals waste time manually searching through documents during meetings, leading to delayed responses and potential oversight of critical details or contractual boundaries.
2. Target Audience: The primary user personas are consultants, client-facing account managers, legal advisors, and freelancers who conduct recurring meetings with clients and need instant, accurate access to a curated knowledge base derived from proposals, contracts, and past communications.
3. Use Cases: Essential for quarterly business reviews where specific metrics from a proposal must be referenced; contract negotiation meetings where agreed-upon terms and "red lines" must be enforced; and onboarding calls where consistent information from training materials must be delivered accurately.

Unique Advantages

1. Differentiation: Unlike cloud-based meeting assistants that index broad web data or upload entire document histories to remote servers, Savvy is architected for local-first operation. It contrasts with tools like Otter.ai or Fireflies.ai by keeping the core knowledge base and history on-device, giving users full control over what data leaves their machine and when.
2. Key Innovation: The product's key innovation is its "source-grounded" recommendation system. It doesn't just listen to the conversation; it continuously cross-references the dialogue against a pre-processed, private brief. This ensures every piece of advice is traceable to a specific source file, dramatically increasing trust and accuracy while mitigating AI hallucination.

Frequently Asked Questions (FAQ)

1. Is Savvy a cloud-based AI meeting assistant? No, Savvy is a local-first macOS application. Your source documents, derived indexes, and meeting transcripts are stored exclusively on your Mac. Only the live audio stream for transcription and minimal excerpts for AI recommendations are sent to external services under your own accounts.
2. What AI models does Savvy use for meeting recommendations? Savvy does not directly integrate with AI APIs. Instead, it shells out to command-line interface (CLI) tools you must install and authenticate separately: OpenAI's codex CLI or Anthropic's claude CLI. This design gives you direct control over the model provider, billing, and terms of service.
3. Can I use Savvy for offline meeting transcription? No, the current build of Savvy does not support offline speech-to-text. Real-time transcription requires an active internet connection and an account with either Deepgram or AssemblyAI, to which meeting audio is streamed.
4. How does Savvy protect my confidential client data during meetings? Savvy employs a privacy-by-design architecture. It streams only audio for transcription (with model-training opt-outs). Your briefs and documents are never sent in full; only relevant excerpts are passed to your local AI CLI at the moment a recommendation is triggered. All local data is protected by macOS FileVault and strict file permissions.
5. What are the system requirements to run the Savvy meeting assistant? Savvy requires a Mac with Apple Silicon (M-series chip) running macOS 13 (Ventura) or later. Intel Macs are buildable from source but are not officially supported. You also need to grant microphone and screen recording permissions for full functionality.