Comparison · Updated October 2026
Msty vs Jan vs Chatbox — and the one where the models debate
All four let you bring your own API keys to GPT, Claude, Gemini and more. Msty, Jan and Chatbox are apps you install, and all three can also run a model on your own computer. PolyCog is a website you open, where several models argue their way to one answer.
The short version
Msty, Jan and Chatbox are desktop AI chat apps, and good ones: install one, plug in your own API keys or run a model on your own computer, and your chats stay on that machine. Jan is free and open source; Msty adds Split Chats, Knowledge Stacks and personas, with its advanced features on a $149-a-year plan; Chatbox also runs on your phone and sells an optional hosted plan. None of them has the models answer each other. PolyCog does: a difference scan, debate rounds where each model reads the others, a verify pass that shows each claim’s source, then one synthesis by the flagship you pick, on a website that keeps your sessions on every device. It runs no local models. If you want a private chat app, or a model that works offline, pick one of them. If you want a question argued out, that’s us.
Msty, Jan or Chatbox?
Three good desktop apps with different strengths. If you only want one of them, here is the short answer.
Jan
Free and open source (Apache 2.0), with local models built in: pick one and it runs on your machine. Cloud models on your own key too. Windows, macOS and Linux; no phone app.
Msty
The most features: Split Chats to compare models side by side, Knowledge Stacks for your documents, personas and an MCP toolbox. The desktop app is free; Aurum ($149 a year or $349 once) adds Studio Web and the advanced set.
Chatbox
The one on your phone: Windows, macOS, Linux, iOS, Android and the web. Free with your own keys; its Chatbox AI plans ($3.99 to $39.99 a month) sell hosted models if you’d rather not hold keys.
Feature comparison
Checked October 2026 against each product’s own site, documentation and repository. All three ship often.
| PolyCog | Msty | Jan | Chatbox | |
|---|---|---|---|---|
| What it is | A website for one question, argued outSeveral frontier models debate to one answer | A desktop AI workspaceStudio Desktop; Studio Web on Aurum | A free, open-source desktop appBy Menlo Research; Apache 2.0 | A chat app for desktop, phone and webPlus an optional hosted AI plan |
| Runs on | Any browserLaptop, phone or tablet; nothing to install | Windows, macOS, LinuxStudio Web in a browser on Aurum | Windows, macOS, LinuxNo phone app | Desktop, iOS, Android, webNative phone apps |
| The app | Free tier · $10/mo · $250 onceFree: 4 models, 3 debate rounds, 20 saved sessions | Free · $149/yr · $349 onceAurum adds Studio Web and the advanced features | FreeNo paid plan | Free on your own keysChatbox AI plans $3.99–$39.99 a month |
| Local models | NoCloud models through their APIs | Ollama, llama.cpp, MLXOnline models too | Built in (llama.cpp)Models from Hugging Face | Through Ollama or LM StudioA local server you run |
| Cloud models, your key | 6 providers · 13 AI labsOpenAI, Anthropic, Google, DeepSeek, xAI on their own keys, plus Qwen, Kimi, GLM, Mistral, Meta’s Llama and Muse, MiniMax, Venice and Nous through one OpenRouter key — cloud models only | Online providersAzure and Bedrock on Aurum | OpenAI, Anthropic, Mistral, Groq…Plus OpenAI- or Anthropic-compatible endpoints | OpenAI, Claude, Gemini, DeepSeek…OpenRouter and a dozen more |
| Several models at once | Every seat, every timeAll answer in parallel | Split ChatsOne message synced to every split, side by side | One model per chatIts docs describe no side-by-side mode | One model per chatIts site and guide describe no side-by-side mode |
| Models read each other | Debate rounds3 on Free, unlimited on Pro | No“Each split chat is independent” | No | No |
| Claims checked | Verify passEach factual claim against a web source, with its exact words | Real-time dataWeb results for the model; limited on Free | Web search toolSearch and fetch for the model | Web searchListed with its AI plans |
| Documents & tools | AttachmentsPer chat; no document library or MCP | Knowledge Stacks + MCPBoth limited on Free | MCP serversPlus a local API server | Knowledge base + MCPKnowledge base on desktop only |
| Where chats live | Your account20 saved on Free, 500 on Pro; every device | This computerStudio Web: that browser, no sync (its docs) | This computerIts data folder | Each deviceStored locally (its FAQ) |
| Free to try | Sample mode3 prompts a day, no account, no key | Free planA download | FreeA download | Free daily quotaChatbox AI’s trial, or your own key |
Split Chats or a debate
Msty sends one message to every split and lays the answers out side by side. In PolyCog the models answer each other first.
One prompt, two ways to ask several models: Msty’s Split Chats send one message to every split and show the answers side by side; PolyCog runs the models in parallel, scans for disagreements, has them debate, and synthesizes one answer.
Local models: the honest trade
All three can run a model on your own computer. PolyCog can’t.
A local model works offline, sends nothing anywhere and costs nothing per answer. Jan has one built in (llama.cpp, with models from Hugging Face), Msty runs Ollama, llama.cpp and MLX, and Chatbox talks to an Ollama or LM Studio server you run. What it takes is memory, and the models that fit on a laptop are smaller than the flagships.
- A 3B model: 8 GB
- A 7B model: 16 GB
- A 13B model: 32 GB
The flagship models PolyCog seats run on their providers’ servers; most of them can’t be downloaded at all.
PolyCog runs cloud models only, through each provider’s API on your own key. If you need a model offline, use one of these three; if you want the frontier models arguing a question out, that’s PolyCog.
Where they beat us
A comparison you can trust has to cut both ways, and these are three very good apps.
Free to use
Jan is free and open source; Msty’s desktop app has a free plan; Chatbox is free with your own keys. PolyCog’s free tier is free too, with limits (4 models, 3 debate rounds, 20 saved sessions); Pro is $10 a month.
Models on your own computer
All three can run a model on your machine, offline: Jan has llama.cpp built in, Msty runs Ollama, llama.cpp and MLX, and Chatbox talks to Ollama or LM Studio. PolyCog runs cloud models only.
Chats that stay on your machine
Each keeps your chats on your own device, and with a local model nothing goes anywhere at all. PolyCog stores sessions in your account, which is what lets them open on your phone.
Your documents and tools
Msty’s Knowledge Stacks, Jan’s and Msty’s MCP connectors, and Chatbox’s desktop knowledge base let a model work with your own files and apps. PolyCog takes attachments per chat, with no document library or MCP tools.
Where PolyCog wins
Models that answer each other
Msty’s Split Chats show answers side by side, each split on its own; Jan and Chatbox run one model per chat. In PolyCog each model reads the others and replies, round after round, before one synthesis names what they still disagree on.
Claims checked, with the exact words
A verify pass looks up each factual claim and shows you the source and the sentence it found, before the synthesis is written. The three apps can give a model web search; none of them describes a claim-by-claim check you can read.
Your chats on every device
Sessions save to your account (20 on Free, 500 on Pro), so a debate started on the laptop opens on the phone. In the three apps a chat stays on the device that made it.
Nothing to install or update
Open a web page. No installer for each computer, no model downloads, no updates to keep up with; the same account works on Windows, a Mac, Linux, a phone or a borrowed laptop.
Common questions
Msty, Jan or Chatbox: which should I pick?
Jan if you want free and open source with local models built in: pick a model and it runs on your machine. Msty if you want the most features: Split Chats to compare models side by side, Knowledge Stacks for your documents, personas and an MCP toolbox, with the advanced set on its $149-a-year Aurum plan. Chatbox if you want an app on your phone too: it has iOS and Android apps and an optional hosted plan. If what you want is several models arguing a question out, that is PolyCog.
Are Msty, Jan and Chatbox free?
Mostly. Jan is free and open source (Apache 2.0), with no paid plan. Msty Studio Desktop has a free plan; Aurum, at $149 a year or $349 once, adds Studio Web and the advanced features. Chatbox is free with your own API keys, and its Chatbox AI plans ($3.99 to $39.99 a month) sell hosted models instead. In all three your API keys are billed by the providers at list price, as they are in PolyCog, which is free, $10 a month or $250 once.
Can they run several models at once?
Msty can: its Split Chats send one message to every split and show the answers side by side, and its own post says each split is independent. Jan and Chatbox run one model per conversation; neither describes a side-by-side mode. In none of them does a model read another’s answer and reply, which is PolyCog’s core: a difference scan, debate rounds, a verify pass and one synthesis.
Can PolyCog run local models?
No. PolyCog runs cloud models through their providers’ APIs. For a model on your own computer, Jan has one built in, Msty runs Ollama, llama.cpp and MLX, and Chatbox connects to Ollama or LM Studio. Jan’s README puts a 7B model at about 16 GB of memory on a Mac.
Can I use the same API keys in PolyCog?
Yes. An OpenAI, Anthropic, Google, DeepSeek, xAI or OpenRouter key works in PolyCog and in all three apps, at the same list prices. Plenty of people keep a desktop app for everyday chat and open PolyCog when a question deserves a debate.
Where are my chats stored?
In Msty, Jan and Chatbox, on the device you use. Msty’s docs say Studio Web keeps its data in that browser and doesn’t sync between devices; Jan keeps a data folder on your computer; Chatbox’s FAQ says chat history is stored locally. PolyCog saves sessions to your account, 20 on Free and 500 on Pro, so they open anywhere you sign in.
See the difference in one prompt
Run the same question through several frontier models, watch them debate, and read the synthesis. Free sample mode, no signup, no keys needed.
Free tier forever · $10/mo · $250 lifetime · your API keys, your bill