Chat models are separate
The desktop app, native runtime, Kokoro voices, and English speech recognition come in the installer. Chat models are your choice: download one in Settings when you are ready.
Why LCOA
LCOA and ChatGPT
ChatGPT is a hosted service; LCOA runs on your computer. They solve different problems, so this lists differences, not a winner. Where we have not measured something, it says so.
| LCOA | ChatGPT | |
|---|---|---|
| Where the AI runs | On your own PC | On OpenAI’s servers |
| Works offline | Yes, after you download a chat model | No, it needs the internet |
| Where your chats live | On your PC | In your OpenAI account, with settings for history and training |
| Price to start | $0 (Free) | Free, with limits |
| Paid plans | Plus $10 a month. Pro, Family and Developer are planned. | Go $8, Plus $20, Pro from $100, Business from $20 per user a month |
| Message limits | No per-message limit on local chat; your PC sets the pace. Free limits voice minutes and Google requests. | Limits by plan, and they change |
| Charges per token | None; models run on your PC | None in the apps; the API is billed separately |
| Models | The Trixe family, sized to your PC’s memory | OpenAI’s hosted models |
| How capable the models are | Not benchmarked against ChatGPT. Hosted models can be far larger than a PC runs. | OpenAI’s largest models |
| Voice | Voices and listening run on your PC. 15 minutes a day on Free, unlimited within LCOA on Plus. | Voice in the app, processed in the cloud |
| Image generation | Plus, with Trixe Imagine on your PC | Included, with limits by plan |
| Files and folders | Works on folders you allow, shows the change first, and can undo most of them | Works on files you upload or connect |
| Mail and calendar | Google, optional. Reads mail and saves drafts; it never sends. | Connectors for Gmail and Google Calendar |
| Ads | None | Go may show ads |
| Platforms | Windows x64 only, beta | Web, Windows, Mac, iPhone and Android |
| Maturity | A beta, unsigned, from a small team | A long-established product |
Checked November 3, 2026. ChatGPT details from OpenAI’s pricing page, the ChatGPT Go announcement and its Business help page. Prices and limits change; check the source before you decide.
How LCOA differs from other local AI tools.
For a PC-first workflow, LCOA brings model choice, realistic speech, listening, planning, and computer tools into one place.
Chat, code and reminders together
Move between chat, Code, timed reminders, web research, and authorized file tools. The advantage is how those pieces come together in your day.
Offline and in your control
Run downloaded chat models and included voice tools offline, choose your model, and decide which folders and actions to allow. Local inference needs no hosted-model API key and incurs no hosted token charges.
| What matters to you |  DESKTOP COMPANION · BETA | LM StudioLOCAL MODEL WORKSPACE | JanLOCAL AI & AGENTS |
|---|---|---|---|
| Local, offline model chat | Yes, after downloading a chat model | Yes, after getting model files | Yes, with downloaded local models |
| Getting your first local model | Choose and download a Trixe model in Settings | Choose and download model files | Choose and download a local model |
| Tools beyond a conversation | Code, reminders, web, file and command tools | APIs and tool integrations | Agents, projects and MCP connectors |
What LCOA bundles: voice and listening in the installer, with chat models chosen separately. Local processing and tool use are shared strengths across this category. Choose LCOA for its combination of features, with realistic speech and listening in the box, and chat models chosen separately.
LCOA is an unsigned beta. Model quality and speed depend on your hardware; no comparative performance advantage has been established. Other tools may be a better fit for a mature model workspace or a different agent workflow. Web services need internet, and local inference uses your PC’s memory, power, and storage.
Checked October 27, 2026 · Sources: LM Studio offline, model setup, integrations, Jan quickstart. Selected features, not performance benchmarks. Products evolve.
Who it is for, and who it is not
A good fit
- You use Windows and want AI that works without sending chats to a server.
- You want to decide which folders and actions an assistant may touch, and see changes before they happen.
- You are happy to pick and download your own chat model.
- You want voice and listening that run on your own machine.
Not a good fit yet
- You need a Mac, Linux, phone or tablet app. LCOA is Windows x64 only for now.
- You need the largest hosted models. A PC runs smaller ones.
- You want chats synced across devices. They stay on one PC.
- You need a signed, mature product today. This is an unsigned beta.