Runs on a MacBook Air.
4 billion parameters. No internet. No cloud. No waiting.
Three speeds. One machine.
COTID VX measured on a MacBook Air with 8GB of RAM — the minimum spec, not the demo spec. Text, browsing, and vision, each at conversational pace.
Text — 19–22 tok/s
Generation speed for chat, drafting, and tool calls. First token in under a second; sustained pace holds across full-length responses.
Browser — 1–2s per search
Live web search from your device through a private connection. Results land in seconds, then the model reads and synthesizes them at full text speed.
Vision — full text speed
Once an image is read, generation runs at 20–22 tok/s. First analysis of a new image takes 4–14s (resolution-dependent); repeat views are instant.
Designed to get faster without more compute — every update makes your existing hardware do more.
8GB minimum
Runs on the cheapest Mac Apple sells. 16GB recommended for peak performance.
0ms network latency
No round-trip to a server for inference. Responses begin instantly.
4B parameters
Large enough for deep reasoning. Small enough for consumer hardware.
2.8GB on disk
The entire model fits in less space than most mobile games.
8,192 context
Long documents, multi-turn conversations, complex reasoning chains.
Faster vision coming
The next-generation vision tower targets several-times-faster image encoding on the same hardware.
What the model can do.
A custom-built multi-modal model designed for reasoning, vision, and context-aware generation — running entirely on your Mac.
See
Understands images, charts, screenshots, and documents. Paste an image and reason about what it sees.
Reason
Multi-step thinking, chain-of-thought analysis, and complex problem decomposition — locally on your device.
Transcribe
Converts video and audio to text. Understands spoken content in context with your project.
Search
Finds sources, reads pages, and synthesizes research — all through a private connection on your device.
Summarize
Condenses long documents, articles, and research into clear, structured summaries you can use immediately.
Generate
Writes, rewrites, expands. Produces structured output adapted to your field and style.
See what’s on your machine — and what never leaves it.
DiffusionScope™ is COTID™'s transparency layer. It doesn’t just claim privacy — it shows you the evidence: what's stored, what was processed, and what stays on your device.
Storage Transparency
Open Settings and see exactly what lives on disk — your models, documents, and session data. Nothing hidden, nothing off-device.
Session Log
Every conversation with your AI is recorded locally — what you asked, what it did, tagged and searchable. Your history stays on your device.
The Local Boundary
Your documents, notes, and research are processed on your machine. You control what gets shared — and what never crosses the boundary.
Patent Pending — DiffusionScope™ interpretability method
Performance by device.
Speed scales with your machine's memory bandwidth. The 8GB numbers are the floor — here's what to expect across Apple Silicon.
| Machine | Text speed | With vision |
|---|---|---|
| MacBook Air 8GB (M1/M2) | ~19–22 tok/s | ~19–22 tok/s |
| MacBook Pro 16GB (M2/M3 Pro) | ~45 tok/s | ~30 tok/s |
| MacBook Pro 32GB (M3 Pro/Max) | ~60 tok/s | ~45 tok/s |
| Mac Studio 64GB+ (M3 Max/Ultra) | ~100+ tok/s | ~80+ tok/s |
Why local is faster than you think.
Cloud APIs are fast at raw token generation — but you pay for network latency, queue times, rate limits, and the cognitive overhead of knowing your data is leaving your machine.
| COTID (local, 8GB Mac) | Typical Cloud API | |
|---|---|---|
| First token latency | 1–2 seconds | 2–5 seconds |
| Generation speed | 19–22 tokens/sec | 30–60 tokens/sec |
| Works offline | Yes | No |
| Data leaves your machine | Never | Always |
| Rate limits | None | Throttled at peak |
| Monthly cost | $0 per token | $20–200+ |
Built for observability at scale.
DiffusionScope™ isn't just for individual users. The same technology that shows you your personal AI activity becomes a full observability platform for organizations — usage analytics, compliance monitoring, knowledge mapping, and deployment management across teams.
Usage Analytics
Who's using what. How much. Which projects. Which features drive value.
Compliance
Every interaction stays on-device. Audit trails without data exposure. Zero trust by design.
Deployment
Push model updates, manage licenses, configure TID templates across your organization.
Enterprise features coming soon. Contact us for early access.
It may not be the biggest model on the planet.
But it's yours. And so is everything you create with it.
Download COTID. Run it on your Mac. Own what you make.
Download Free Trial