A native, on-device AI assistant platform with embedded memory and zero cloud dependency. Your data never leaves your machine.
AI assistants today require you to send every conversation, every document, and every thought to someone else's servers.
For businesses handling sensitive data — legal, financial, medical, defence, IP-heavy R&D — that's a non-starter. You shouldn't have to choose between powerful AI and total data privacy.
Create named AI agents specialised for different domains — coding, research, compliance, writing, operations. Each agent maintains its own context, rules, and conversation history.
Connects to any OpenAI-compatible local model endpoint. Streaming real-time responses. Choose your model — from lightweight 4B to full 70B+ depending on your hardware.
Three native memory modules — Context Store, Fact Graph, and Learning Engine — built directly into the app. All stored locally in embedded SQLite with vector search. No external databases.
An included pipeline extracts knowledge from your conversations, distills it into training data, and fine-tunes LoRA adapters on-device. The AI improves over time based on how you work.
Inspired by OpenViking
Hierarchical knowledge organisation with layered summaries — deep context and resource memory.
Inspired by Graphiti
Temporal entity-relationship tracking with validity windows — knows when facts changed.
Inspired by PAL
Continuously improves retrieval quality and promotes raw conversation data into structured knowledge.
The safest data is data that never leaves.
All data stays on your Mac. No cloud sync. No remote servers. No third-party API calls.
No analytics. No crash reporting. No usage tracking. We literally cannot see what you do.
Secrets stored in macOS Keychain — Apple's hardware-backed security framework. Not in config files.
Fully functional with zero internet. The only network call is to your own local LLM endpoint.
LoRA fine-tuning runs on-device using Apple Silicon. Training data never leaves your machine.
No personally identifiable information leaves the device. By design, not by policy.
SwiftMaestro runs models natively on Apple Silicon via MLX. No external server required.
Apple M1 Ultra Mac Studio
64GB unified memory
macOS 14.0 Sonoma or later
Runs up to 35B parameter models at full speed
M2 Ultra / M4 Max Mac Studio
128GB+ unified memory
Runs 70B–122B parameter models
Concurrent model loading + fine-tuning
Windows & Linux ports planned
NVIDIA GPU acceleration under evaluation
Same privacy model — all inference stays local
Cross-platform timeline: 5–8 months
┌───────────────────────────────────────────────────────┐
│ SwiftMaestro.app │
│ │
│ ┌──────────┐ ┌────────────────────────────────────┐ │
│ │ SwiftUI │ │ Memory Orchestrator │ │
│ │ Chat │◀▶│ ┌────────┐ ┌───────┐ ┌──────────┐ │ │
│ │ UI │ │ │Context │ │ Fact │ │ Learning │ │ │
│ └──────────┘ │ │ Store │ │ Graph │ │ Engine │ │ │
│ │ │ └────┬───┘ └───┬───┘ └────┬─────┘ │ │
│ │ │ └─────────┴──────────┘ │ │
│ ▼ │ SQLite · FTS5 · vec │ │
│ ┌──────────┐ └────────────────────────────────────┘ │
│ │ MLX │ │
│ │ Inference│ HuggingFace Hub ←→ Local Model Store │
│ │ Engine │ macOS Keychain (secrets) │
│ └──────────┘ │
│ │ Native Apple Silicon GPU │
└───────┼───────────────────────────────────────────────┘
│ (optional fallback)
▼ HTTP (localhost / LAN)
┌─────────────────┐
│ External LLM │
│ LM Studio / │
│ Ollama / vLLM │
└─────────────────┘
Single installer per Mac. No IT infrastructure beyond Macs and a local model server (shared or per-seat).
Each installation is fully self-contained. No shared databases, no cloud state. Meets strict data sovereignty requirements.
Choose your LLM — Qwen, Llama, Gemma, Mistral — hosted on-premises. Full control over capabilities and training data.
Pre-configure agents with business-specific rules, knowledge, and workflows, then distribute to your team.
macOS-native today. Designed for portability tomorrow.
Native macOS app — multi-agent chat, streaming, secure storage
Embedded memory architecture — fact graph, context store, learning engine
Chat pipeline integration with memory-augmented responses
Learning engine, knowledge promotion, maintenance workers
Signed/notarised distribution, App Store evaluation
Interested in SwiftMaestro for your business?
Whether you're exploring private AI for your organisation, interested in integration, or want to discuss cross-platform deployment — we'd love to hear from you.
hello@swiftmaestro.com