Jan vs LM Studio: what each local AI app sends, and the hardware it needs
The Morse teamPublished Facts checked 11 min read
As of November 2026, Jan vs LM Studio is a choice between an app you can audit and an app that says less leaves it. Jan is Apache 2.0 open source, and its code shows it checking apps.jan.ai for updates when it starts and every hour after, with a random ID it keeps, and its built-in web search is on by default. LM Studio is closed source, says its app "does not include telemetry," and sends an update check when you open it. Both keep a local model's chats on your own disk.
We read Jan's privacy policy (last updated January 16, 2025), privacy page, desktop docs, changelog, and source code at release v0.8.8, and LM Studio's app privacy policy (last updated October 19, 2026), app terms (version of August 23, 2026), pricing page, docs, and blog, all on November 17, 2026. Jan earns credit for analytics that stay off in its code until you say yes, and LM Studio for a policy that names every occasion it hears from your copy of the app.
Jan vs LM Studio vs Morse at a glance
Read on November 17, 2026. Every Jan and LM Studio fact links to its own page or code.
| Jan | LM Studio | Morse | |
|---|---|---|---|
| Where your prompt is decrypted | Local model: your computer. Cloud model: the provider whose key you added (Jan) | Local model: your computer. Bionic cloud: "US-hosted" (LM Studio) | Anonymous: at the provider. Private: at a host that agreed to keep nothing. E2EE: only inside an enclave your browser has checked |
| Who verifies it | You can read the Apache 2.0 code for the app (GitHub); a cloud provider is its own word | LM Studio's word; the app's source is a trade secret (terms) | Your browser checks the enclave's attestation before sending (E2EE) |
| Where chat history lives | JSON files in Jan's data folder (Jan) | JSON files in .lmstudio/conversations (LM Studio) | Stored in your browser, encrypted, not on Morse's servers |
| What the app sends on its own | Update check at launch and hourly; web search when a model calls it (code) | Update check when you open it (docs) | Not compared here |
| Analytics | Off until you opt in; PostHog EU (Jan) | "does not include telemetry or user-specific tracking" (LM Studio) | Not compared here |
| Training | Jan doesn't receive your chats; cloud providers' own policies apply (Jan) | Cloud requests "not used for training or any other purpose" (LM Studio) | Morse doesn't train AI on your chats |
| Models | Whatever fits your machine, plus OpenAI, Anthropic, Gemini, and others on your own key (Jan) | Whatever fits your machine, plus Bionic's cloud models (LM Studio) | 80+, open and closed, including GPT, Claude, and Gemini |
| Price | Free (Jan); cloud providers bill you | Free locally; Bionic+ $20 and Pro $100 a month (LM Studio) | Free, 15 messages a day; Pro $19 and Max $49 a month |
What's the difference between Jan and LM Studio?
Jan is an open-source desktop app from Menlo Research Pte Ltd (Jan) that runs models with llama.cpp everywhere and with an experimental MLX engine on Apple Silicon (Jan), and plugs your own API keys into cloud providers. LM Studio is a closed-source app from Element Labs, Inc., a Delaware corporation (LM Studio), that runs llama.cpp and MLX models (LM Studio) and sells its own cloud through a separate app, Bionic.
Both serve an OpenAI-compatible API on your computer, Jan at http://127.0.0.1:1337/v1 (Jan) and LM Studio at http://localhost:1234/v1 (LM Studio), so a coding tool built for one can usually point at the other. LM Studio's comparison with the other big local app is in Ollama vs LM Studio.
What does each app send when you aren't chatting?
With a local model loaded, this is what crosses the network, from each app's docs and, for Jan, its code at v0.8.8 (October 28, 2026).
| Event | Jan | LM Studio |
|---|---|---|
| Opening the app | Update check to apps.jan.ai, after first paint (code) | Update check on macOS and Windows; the Linux updater is "in the works" (docs) |
| While it stays open | Another update check every hour by default (code) | Nothing listed in the policy (policy) |
| What the update check carries | App version, OS and chip type, and a random ID saved on your disk (code) | "app version and build, operating system," and your IP address via its CDN (policy) |
| Usage analytics | Only if you opt in: active users and retention (policy) | None (policy) |
| Searching for models | Hugging Face, from the Hub (Jan) | Requests to huggingface.co, plus "Anonymized search queries" to LM Studio (docs, policy) |
| A model that can use tools | Web search queries to Exa by default (Jan) | Web search is sold with Bionic: "Limited web search" on its free plan (pricing) |
Jan's privacy policy doesn't mention updates (Jan), so the code is where to look. On its first check, Jan makes a random ID with crypto.randomUUID() and saves it in a file called updater.json in its app data folder (code). Every later check sends that same ID in an X-Client-Session header, along with the app version, and a User-Agent of the form Jan/{version} ({os}; {arch}) (code). The code uses the ID to sign the request, and Jan doesn't say whether apps.jan.ai stores it, but an ID that stays the same would let that server tell your copy of Jan apart from others, hour after hour. It runs whatever you answered on analytics, and the setting that stops it is Automatic Update Check under Settings, General, which is on by default (code).
LM Studio's policy says of its own check, "We do not collect personal identifiers, nor track your individual behavior" (LM Studio). We can't check that against code, since the app is closed.
Does Jan have telemetry?
Only if you opt in, and then very little. At first launch Jan asks for your tracking preference, and its policy says "The Company does not collect any data until you explicitly allow tracking" (Jan). If you allow it, Jan counts daily active users and retention through PostHog EU, tied to a random user ID, and keeps that data "for up to 12 months." The code matches the policy: the analytics switch starts off, and when it's on, PostHog runs with autocapture off and session recording disabled (code, code). You can change the answer under Settings, Privacy.
Jan's docs also have a page called "MCP routing telemetry." Despite the name, it describes a hook for developers, and says "Shipped Jan does not register onRoutingTelemetry" (Jan).
Is Jan's web search on by default?
Yes, since v0.8.4 on July 21, 2026 (Jan). Jan's docs say "Web Search is on by default," and that the default provider, Exa, "uses a free, keyless endpoint" (Jan). Jan offers two tools, web_search and web_fetch, to any model that can call tools, and the model decides when to use them. The query it writes, which can carry words from your prompt, goes to Exa's hosted endpoint at mcp.exa.ai (code). Tavily, You.com, and a SearXNG server you host yourself are the other choices.
Jan's privacy page still says Jan "is used 100% offline on your own computer" (Jan), and its desktop policy doesn't mention search. For a chat that shouldn't touch the internet, turn off Enable Web Search under Settings, Web Search, or click the globe icon in the input toolbar, or pick a model without tool use.
LM Studio's offline guide lists chatting, chatting with documents, and the local server as working with no connection (LM Studio). Its web search lives in Bionic, where the pricing FAQ says it "uses Zero Data Retention (ZDR)" and warns that "querying and processing data from the web always carries its own privacy and security risks" (LM Studio).
Is Jan open source, and is LM Studio?
Jan is, under Apache 2.0, and LM Studio's app isn't. Jan's license file reads "Copyright 2025 Menlo Research" and "Licensed under the Apache License, Version 2.0" (GitHub). BetterClaw's comparison, updated October 27, 2026, still says Jan is "released under AGPLv3," which its license file no longer says.
LM Studio's app is closed, and only its MIT command line, MLX engine, and SDKs are public (GitHub); its terms and reverse-engineering clause are covered in Ollama vs LM Studio.
Can you use Jan and LM Studio at work?
Yes, both for free. Jan's Apache 2.0 license asks only that "Attribution is requested in user-facing documentation and materials, where appropriate" (GitHub), and LM Studio dropped its separate commercial license on July 8, 2025 (LM Studio). On a work network, Jan's local server listens only on 127.0.0.1 by default, takes an optional API key, and lets you list the trusted hosts allowed to reach it if you open it to other devices (Jan).
How do cloud models work in Jan and LM Studio?
Jan uses your own API keys, and LM Studio sells its own cloud. Jan's install guide goes from download to launch with no sign-in step, so local models need no account (Jan). It connects to other companies' APIs once you paste a key under Settings, Model Providers: OpenAI, Anthropic, Google, Mistral, Groq, OpenRouter, Azure, Hugging Face, and custom endpoints (Jan). Its policy says "The API provider processes your messages directly; the Jan app does not access or store these messages" (Jan), so each chat falls under that provider's account terms, with your name on the key.
LM Studio sells cloud models through Bionic, a separate app: Bionic+ at $20 a month runs Kimi K3, GLM 5.3, and DeepSeek V4 Flash, and Pro at $100 a month adds "Included usage for GPT Sol 6.1 and Claude Opus 5.5 (ZDR)" (LM Studio). What its zero-retention promise covers is in is LM Studio safe.
What hardware do Jan and LM Studio need?
Similar machines. On a Mac, LM Studio recommends more memory; on Windows and Linux, Jan asks for more graphics memory. From each app's own pages on November 17, 2026:
| Jan | LM Studio | |
|---|---|---|
| Mac | Apple Silicon, macOS 13.6 or newer; Intel Macs unsupported since v0.8.0, with local models gone in v0.8.5, so Jan tells Intel owners to stay on v0.8.4 (Jan) | Apple Silicon, macOS 14.0 or newer; Intel Macs "currently not supported" (LM Studio) |
| Mac memory | 8 GB "typically up to 3B models comfortably," 16 GB up to 7B, 32 GB up to 13B (Jan) | "16GB+ RAM recommended"; 8 GB Macs should "stick to smaller models" (LM Studio) |
| Windows | Windows 10 or newer, x64 with AVX2 or ARM64; 8 GB minimum, 16 GB recommended (Jan) | x64 with AVX2 or ARM; "At least 16GB of RAM is recommended" (LM Studio) |
| Graphics | 6 GB of VRAM minimum on NVIDIA, AMD, or Intel Arc (Jan) | "at least 4GB of dedicated VRAM is recommended" (LM Studio) |
| Linux | Flatpak, .deb, or AppImage; AVX2 required (Jan) | AppImage on x64 or ARM64, Ubuntu 20.04 or newer (LM Studio) |
| Disk | "10GB+ free space" (Jan) | Not stated |
Jan's guide tops out at 13B models on a 32 GB Mac. The local AI vs cloud AI guide covers where that line falls for other tools.
Where do Jan and LM Studio keep your chats?
On your disk, as JSON, and neither company says it sees them. Jan's policy calls your conversations "private and inaccessible to the Company" (Jan), and LM Studio's says that with local models "none of your messages, chat histories, and documents are ever transmitted from your system" (LM Studio). Jan saves each thread as messages.jsonl in its data folder, at ~/Library/Application Support/Jan/data on a Mac, %APPDATA%/Jan/data on Windows, and ~/.local/share/Jan/data on Linux (Jan). LM Studio says "Conversations are stored in JSON format" in ~/.lmstudio/conversations/ (LM Studio). Neither app's docs say the files are encrypted, so anyone who can open your account on that computer can read them. Jan's factory reset erases "all user data, including downloaded models and chat history" (Jan).
How Morse gives you frontier models without the hardware
Jan and LM Studio keep a chat private by keeping the model on your computer, so your memory decides which models you get. Past that line, Jan hands the prompt to a provider under your own API key, and LM Studio's cloud asks you to trust processors its policy doesn't name (LM Studio). Morse is a web app, so its models also run in the cloud. Morse changes what reaches the provider and, in E2EE, who can read it at all.
- E2EE, for the big open models your laptop can't load, sealed before they leave. Open-weight models like GLM, Kimi, and DeepSeek run in secure enclaves. Your browser checks the enclave's hardware attestation against versions pinned in the app, then encrypts the message so only that enclave can read it, and Morse's servers relay text they can't read. Third-party providers run the enclaves, not Morse, and the protection rests on the enclave hardware and attestation working as designed. E2EE has no web search or image generation, and comes with Pro at $19 a month.
- Anonymous, for GPT, Claude, and Gemini without your own API key. Jan sends these chats under a key tied to your account with the provider. On Morse, our proxy removes your account, email, and IP address and sends the message under Morse's own account, so the provider sees the message without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy.
- Private, the default, for open-weight models and xAI's Grok when you don't need an enclave. Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
Morse runs no advertising trackers or third-party analytics, shows no ads, and doesn't train AI on your chats, as our privacy page says. Where Jan and LM Studio leave JSON files on your disk, Morse history is stored in your browser, encrypted, not on Morse's servers, and locked when you sign out. Clearing the browser deletes it and we can't recover it, so keep a copy of anything you need. How your browser checks an enclave is walked through in how to verify an AI enclave's attestation.
Choose Jan, LM Studio, or Morse
- Choose Jan if you want code you can read, you'll bring your own API keys for cloud models, and you'll turn off the hourly update check and web search when you need a chat to stay offline.
- Choose LM Studio if you want an app whose policy lists the fewest things it receives, you're happy to take that on its word, and you'd buy cloud models from the same company.
- Choose Morse if your computer can't hold the model you want, and you want E2EE that your own browser checks, GPT, Claude, and Gemini without the provider knowing it's you, and history encrypted in your browser, in one account.
We read every page and file linked above on November 17, 2026, and took each company at its written word or its published code; we didn't install either app or watch its traffic for this comparison.
Send a first message from the Morse home page with no account, or start free with no card.
Morse publishes this comparison. We've tried to keep it fair; if something's wrong or out of date, email hello@trymorse.ai.
Jan is a trademark of Menlo Research Pte Ltd. Morse is not affiliated with or endorsed by Menlo Research Pte Ltd.
LM Studio is a trademark of Element Labs, Inc. Morse is not affiliated with or endorsed by Element Labs, Inc.
Common questions
Does Jan run on an Intel Mac?
Not for local models. Jan's Mac guide says Intel Macs haven't been supported since v0.8.0 and that local models stopped running on them in v0.8.5, so it tells Intel owners who need local models to stay on v0.8.4. LM Studio's requirements say Intel-based Macs are currently not supported.
How do I stop Jan's update check and web search?
Turn off 'Automatic Update Check' under Settings, General, and turn off 'Enable Web Search' under Settings, Web Search. You can also switch web search off for one chat with the globe icon in the input toolbar.
Do I need an account to use Jan?
Not for local models. A cloud model in Jan needs an API key from that provider, which you paste in under Settings, Model Providers, and the provider bills you for it.
Does Jan or LM Studio see my chats?
Neither says it does for local models. Jan's privacy policy says your conversations are 'private and inaccessible to the Company,' and LM Studio's says that with local models none of your messages, chat histories, and documents are ever transmitted from your system.