Skip to content

Airscker/ASREngine

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

564 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Aide

Turn meetings into execution — inside your workflow, on-prem.

Aide is a comprehensive AI agent designed to improve work efficiency, reduce time and resource costs, and make daily work clear and organized. Not another note-taker — Aide is a workflow-native execution layer with governed enterprise memory that answers the four questions every team asks later: Why? Where? Who? What next?

Flutter macOS Windows Privacy On-Prem


🎯 The Problem Aide Solves

Teams waste hours every week reconstructing "why / who / what next" — and policy often forbids cloud recording. This creates:

  • Decision drift — Decisions are made, but their context is lost quickly
  • Ownership gaps — Unclear who is responsible for what after meetings
  • Follow-up loss — Action items fail to be reliably captured or transferred into tools like Jira, CRM, or email
  • Cross-team amnesia — New team members struggle to understand past decisions and their rationale

The cost: Repeated meetings, manual notes + status chasing + documentation overhead, slow onboarding, painful handoffs across teams, and higher audit and compliance friction.

Why existing tools fail:

  • Cloud-first tools are blocked by policy due to sensitive conversations
  • Many "meeting bots" introduce security/compliance friction, leading to slow adoption
  • Tools produce notes, not workflow execution + governed memory

✨ What Makes Aide Different

🎯 Decision Trail by Default

Decisions + context + owners + approvals are always traceable. Every meeting decision is automatically captured with full context, making it easy to answer "Where was this decided — and who approved it?"

Execution, Not Notes

Action items with deadlines/status are not merely recorded — they're pushed into Jira/Slack/Email/CRM (coming soon). Aide transforms meeting discussions into actionable workflow items.

🧠 Shared Enterprise Memory

Cross-meeting search answers "why/who/what next" questions in seconds. Aide maintains a comprehensive knowledge base across all your meetings, making historical context instantly accessible.

🏢 On-Premise & Privacy-First

Run Aide with local embedded engines or a self-hosted unified HTTP backend. No third-party cloud dependency is required.


🚀 Core Capabilities

🎙️ Real-time Transcription

  • Live audio transcription with low latency (<200ms)
  • Frame-based streaming with automatic commit detection
  • Voice Activity Detection with intelligent buffering
  • Support for system audio and microphone
  • File upload transcription for recorded meetings

📊 Intelligent Summarization (Pro Feature)

  • AI-powered meeting summaries with structured output
  • Executive summary, key topics, decisions, and action items
  • Streaming token generation for real-time feedback
  • Metal GPU acceleration on Apple Silicon
  • Powered by local Llama.cpp inference

🤖 AI Agent Assistant (Pro Feature)

  • Role-based meeting agents for Sales, Engineering, Legal, and Management
  • Chat UI with tool-grounded answers
  • Natural language query interface
  • Shared Enterprise Memory — agents leverage cross-meeting knowledge
  • Cross-agent search for comprehensive insights

🔍 Advanced Search & History (Pro Feature)

  • Full-text search across all meeting transcripts and summaries
  • Smart filtering by date, duration, status, tags, and metadata
  • BM25 ranking algorithm for relevant results
  • Highlighted match snippets with context
  • Cross-meeting search to answer "why/who/what next" questions

📅 Calendar Integration

  • Automatic calendar access with permission management
  • View upcoming events and meetings
  • Create calendar events directly from the agent
  • Seamless workflow integration

🔒 Enterprise-Grade Security

  • Local or self-hosted processing — no third-party cloud dependency required
  • Policy, Permissions & Audit — built-in governance and traceability
  • On-premise deployment ready
  • No account or internet required (after model download)
  • Full user control over data

🏗️ Core Engines

Aide is built on three foundational technologies:

🎤 ASR (Automatic Speech Recognition)

  • Local embedded mode: whisper-based native runtime
  • Remote mode: unified Aide AI server (/v1/asr/transcriptions + streaming APIs)
  • Real-time transcription with low latency
  • Support for multiple audio sources

🧠 LLM (Large Language Model)

  • Local embedded mode: llama-based native runtime
  • Remote mode: unified Aide AI server (/v1/summaries, /v1/chat/completions)
  • Summarization, action item extraction, and agent reasoning
  • Context + output guardrails for stable memory usage

🛡️ Policy, Permissions & Audit

  • Built-in security and governance features
  • Permission management for calendar and microphone
  • Audit trail for all meeting data
  • Compliance-ready architecture

🆕 Recent Updates

  • 🔀 Unified backend mode switch — one setting controls ASR + LLM transport (Local Embedded vs Remote HTTP Server)
  • 🌐 Remote backend UI panel — base URL, API key, and health-check in Settings > AI Backend
  • 🟢 Global engine status indicator — app-wide readiness icon for local/remote engines
  • 🚫 No silent fallback in remote mode — tasks fail fast with explicit configuration errors
  • 🎙️ Remote realtime streaming — app now supports server-side streaming ASR endpoints
  • 🧾 Operational logging split — server routes concise logs to console, detailed logs to server.log/engine.log

🚀 Quick Start

For End Users

macOS

# Download the appropriate DMG for your Mac:
# Apple Silicon (M1/M2/M3/M4): aide_macos_arm64_1.6.3.dmg
# Intel:                       aide_macos_x86_64_1.6.3.dmg

# Open DMG and drag to Applications
open aide_macos_arm64_1.6.3.dmg

Windows

# Download aide_windows_x64_1.6.3.zip
# Extract and run asr_app.exe

First Launch:

  • Grant microphone and calendar permissions (auto-requested)
  • Models download automatically (~2.5 GB, 5-10 minutes)
  • Start recording or upload audio files!

For Developers

# Clone repository
git clone <repo-url>
cd ASREngine/asr_app

# Get dependencies
flutter pub get

# Run in debug mode
flutter run

Requirements:

  • Flutter SDK 3.24+
  • macOS: Xcode 15+
  • Windows: Visual Studio 2022 + C++ workload

📚 Documentation

Document Description
USAGE.md End-user manual & guide
aide-ai-platform/README.md Unified ASR+LLM server deployment
AI HTTP Server (Dev) Server setup, API usage, tuning, and troubleshooting
SERVER_TECHNICAL.md Server runtime architecture and APIs
GETTING_STARTED.md Complete setup guide for developers
BUILD.md Build instructions for macOS & Windows
MODELS.md Model management and configuration
ARCHITECTURE.md Technical deep dive: FFI, dylibs, Metal
API_STREAM.md Streaming engine C API reference
TROUBLESHOOTING.md Common issues and solutions

🏗️ Architecture

Aide Desktop App (Flutter)
        ↓ (Dart FFI)
Native Helper Libraries
        ↓ (C++ Inference)
whisper.cpp / llama.cpp
        ↓ (GPU Acceleration)
Metal / CoreML (Apple Silicon)

Key Components

  • libstream.dylib - C++ streaming engine with frame-based API
  • whisper.cpp - ASR inference engine with Metal acceleration
  • llama.cpp - LLM inference engine for summarization and agent reasoning
  • Silero VAD - Voice activity detection
  • ScreenCaptureKit - System audio capture (macOS)
  • Flutter - Cross-platform UI framework
  • Tool Orchestrator - Agent tool system for SEARCH, LIST_HISTORY, MEETING_DETAILS, CALENDAR_GET

Data Storage

~/.Aide/
├── models/                     # AI models (~2.5 GB)
│   ├── ggml-distill_lv3.bin   # ASR model
│   └── Qwen3-0.6B-Q8_0.gguf   # LLM model
├── history/                    # Session data
│   ├── aide.db                # SQLite database
│   └── sessions/              # Individual session data
│       └── <session_id>/
│           ├── audio.wav      # Original recording
│           ├── transcript.txt # Segmented text
│           ├── timestamp.bin  # Binary timestamps (Int64 pairs)
│           ├── metadata.bin   # Segment count (Int32)
│           └── summary.md     # AI Summary
└── logs/                       # Application logs (if enabled)

🔧 Building from Source

macOS Release Build

cd asr_app
bash build_release.sh

Output: Aide-x.x.x-macos.dmg (signed and notarized)

Features:

  • Code signing with Developer ID
  • Notarization for Gatekeeper
  • Professional DMG with drag-to-install
  • Automatic retries for reliability

Windows Release Build

cd asr_app
bash build_release_windows.sh

Output: Aide-x.x.x-windows-x64.zip

Optional installer:

bash build_release_windows.sh --installer

See BUILD.md for detailed instructions.


🎯 Use Cases

For Individuals

  • 📝 Transcribe meetings, interviews, lectures with instant results
  • 🎤 Record voice memos with automatic transcription
  • 📊 Generate meeting summaries automatically (Pro)
  • 🔍 Search through conversation history to find past decisions (Pro)
  • 🤖 Get AI-powered insights and action items (Pro)

For Teams

  • 💼 Internal meetings and brainstorming sessions
  • 🎓 Training sessions and workshops
  • 📞 Client calls (with permission)
  • 🗣️ Standup meetings and retrospectives
  • 🔄 Decision trail tracking — Always know where decisions were made
  • 👥 Ownership clarity — Track who approved what and when
  • Action item execution — Transform discussions into workflow items

For Enterprises

  • 🏢 Regulated organizations — On-premise deployment with full audit trail
  • 🔒 Security-sensitive environments — No cloud, no data leaks
  • 📋 Compliance requirements — Built-in governance and traceability
  • 🔍 Cross-team knowledge — Shared enterprise memory across departments
  • 🎯 Workflow integration — Connect to Jira, Slack, CRM (coming soon)

For Creators

  • 🎙️ Podcast transcription
  • 📹 Video content planning
  • ✍️ Blog post drafts from voice notes
  • 🎬 Interview transcription

🔒 Privacy & Security

What We DON'T Do

  • ❌ No cloud processing
  • ❌ No data collection
  • ❌ No telemetry or analytics
  • ❌ No account required
  • ❌ No internet connection needed (after setup)
  • ❌ No third-party services

What We DO

  • ✅ 100% local AI inference
  • ✅ Data stored only on your device
  • ✅ Open-source components (whisper.cpp, llama.cpp)
  • ✅ Transparent architecture
  • ✅ Full user control over data
  • ✅ On-premise ready
  • ✅ Enterprise-grade security

📊 Performance

Metric Value
Transcription Latency <200ms
Real-time Factor (RTF) ~0.1x (10x faster than real-time)
Model Size ASR: 1.5 GB, LLM: 1 GB
Memory Usage ~2-3 GB during active use
GPU Acceleration 5-10x speedup on Apple Silicon
App Size ~50 MB (models downloaded separately)
Context Window 32k tokens (with intelligent truncation)

🛠️ Tech Stack

Component Technology
UI Framework Flutter 3.24+
ASR Engine whisper.cpp
LLM Engine llama.cpp
VAD Silero VAD
GPU Metal (macOS), DirectML (Windows planned)
Database SQLite
Search BM25 ranking algorithm
Audio dart_soundboard, ScreenCaptureKit
Agent Tools Custom tool orchestrator with EXECUTE protocol

🤝 Contributing

We welcome contributions! Here's how:

  1. Fork the repository
  2. Create a feature branch (git checkout -b feature/amazing-feature)
  3. Commit your changes (git commit -m 'Add amazing feature')
  4. Push to the branch (git push origin feature/amazing-feature)
  5. Open a Pull Request

Development Setup

See GETTING_STARTED.md for complete setup instructions.


📝 License

See LICENSE file for details.


🙏 Acknowledgments

Built with amazing open-source projects:

AI Models:


📞 Support


🗺️ Roadmap

Current (v1.6)

  • ✅ Real-time transcription
  • ✅ AI summarization (Pro)
  • ✅ Advanced search (Pro)
  • ✅ AI Agent Assistant (Pro)
  • ✅ Session history
  • ✅ Calendar integration
  • ✅ macOS & Windows support
  • ✅ Dark/Light themes

Planned (v1.7)

  • 🔄 Workflow connectors (Jira, Slack, Teams, Zoom, CRM)
  • 📤 Export to multiple formats
  • 🎨 Custom themes
  • 🌐 Multi-language UI
  • 🔄 Auto-update system

Future (v2.0)

  • 🗣️ Speaker diarization
  • 📊 Advanced analytics and insights
  • 🔌 Plugin system
  • 🐧 Linux support
  • 🌍 Multi-language transcription
  • 📈 Team collaboration features

💡 Key Differentiators

Not another note-taker — Aide is a workflow-native execution layer with governed enterprise memory. Aide answers the four questions every team asks later:

  1. Why? — Decision trail with full context
  2. Where? — Cross-meeting search finds where decisions were made
  3. Who? — Ownership tracking and approval records
  4. What next? — Action items pushed into workflow tools

On-premise & privacy-first — Perfect for enterprises that need security without sacrificing functionality.

Execution, not notes — Transform meeting discussions into actionable workflow items, not just passive documentation.


Made with ❤️ for teams that value privacy, efficiency, and execution

No cloud. No data leaks. Just you, your AI, and your workflow.

About

Aide is a comprehensive AI agent designed to improve work efficiency, reduce time and resource costs, and make daily work clear and organized. Not another note-taker — Aide is a workflow-native execution layer with governed enterprise memory that answers the four questions every team asks later: Why? Where? Who? What next?

Resources

License

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors