# Chatbox vs Jan for as local platform for Inference Cooperative members ## Bottom line Chatbox is the better default for most members because it runs on phones; Jan is the better choice for desktop members who also want to run open models on their own computer. Both keep chat history on the member's device and both reach the co-op through its OpenAI-compatible gateway. - **Reach:** Chatbox covers Mac, Windows, Linux, iOS, Android and the web. Jan is desktop-only. For members who chat mainly on a phone, only Chatbox works. - **Ease of connecting:** Jan fetches the co-op's model list automatically. In Chatbox, members type model names in by hand, which is the most error-prone step for non-developers. - **Openness:** Jan is Apache 2.0 throughout. Chatbox's public code is a GPL-3.0 "Community Edition" synced from a separate "pro" repository, and the app sells its own paid AI plans, which compete with the co-op's. - **Running models locally:** Jan runs open models itself. Chatbox needs a separate local server such as Ollama or LM Studio. - **Neither** solves cross-device history without a server, which members have raised in issue #57. Suggested path: document both, Chatbox first for phones and Jan for desktop. If the co-op later customises one app for members, both licenses allow it; GPL keeps any co-op version open, Apache allows more. ## Capabilities side by side | | Chatbox | Jan | | --- | --- | --- | | Platforms | Mac, Windows, Linux, iOS, Android, web | Mac (Apple silicon, 13.6+), Windows 10+, Linux; no mobile app | | Connecting to the co-op | Settings → Model Provider → Add, type "OpenAI API compatible", API host + key, add models by hand, then Check | Settings → Model Providers → Add Provider, OpenAI-compatible, base URL + key; model list fetched from the gateway | | Where history lives | On the device, unless the member opts into a Chatbox AI account | On the device | | Cross-device sync | Only through Chatbox's own servers (paid account) | None | | Runs open models on your own computer | No; connects to Ollama or LM Studio | Yes, built in (llama.cpp; MLX on Apple silicon) | | Images and files | Image input with vision models; document chat listed under paid Pro plans | Image input with vision models; file attachments | | Tools (MCP) | Listed under paid Pro plans | Free, with per-call approval | | Web search | Paid Chatbox AI plans | Built-in web search tool (v0.8.4+) | | Interface languages | 9 | Not checked | | Cost to members | Free with your own key; optional plans $3.99–$39.99/month | Free; no paid tier | Web search, document search and image generation are things the co-op's web chat already offers; members who switch to either app lose them unless the app supplies its own. ## Usability for non-developers Jan is easier to connect; Chatbox is easier to live with on a phone. The hardest step in both is the same: copying an API key from the Member Dashboard without exposing it. | Step | Chatbox | Jan | | --- | --- | --- | | Install | App store on phones; installer on desktop; or no install via the web version | Installer on desktop (.dmg, .exe, Flatpak) | | Account | None needed with your own key; the app also offers Chatbox AI sign-up | None | | Fields to fill | Name, API type, API host, key, then each model ID typed by hand | API format, name, base URL, key | | Confirm it works | A built-in Check button | Model list appears after Create | | Typical mistake | Typo in a model ID like `greenpt/deepseek-v4.1-flash` | Base URL missing `/v1` | | Tools and images | Set per model when adding it | Pencil icon → Capabilities per model | | Distractions | The app also offers its own paid Chatbox AI plans and models | None; no paid tier | For a guide, Chatbox needs one extra page: a short list of co-op model IDs to copy. Chatbox's API path default (`/v1/chat/completions`) means members enter the host without `/v1`; that detail should be checked and spelled out. ## Fit with the co-op's values and roadmap Both apps advance the charter's privacy commitment by keeping transcripts off the co-op's server. They differ most on transparency and on how well they serve less-technical members. | Charter value | Chatbox | Jan | | --- | --- | --- | | **Trustworthiness** ("protect each other's privacy") | History on device; Chatbox AI sync is opt-in and goes to the vendor's servers | History on device; usage analytics off unless the member opts in | | **Transparency** | Community Edition is GPL-3.0, but it is synced from a separate pro codebase | Whole app Apache 2.0 in one public repo | | **Participation** | GPL lets the co-op ship its own customised build, which must stay open | Apache lets the co-op ship its own build under any terms | | **Experimentation** | Many cloud providers; local models via Ollama or LM Studio | Local models built in, plus MCP tools, for members who want to tinker | | **Sustainability** | Neutral: model energy depends on which co-op model is chosen | Running small models locally avoids data-centre use for some tasks, at the cost of the member's own power | **Where the roadmap points** - **#57 Improving the privacy of our chat app** (Nathan Schneider, 2026-11-18): lists a local desktop-and-mobile app as one option and names Chatbox for its GPL-3.0 license. It rejects LobeChat because its license blocks permissionless customisation, the same reason the co-op dropped Open WebUI. Both Chatbox and Jan pass that test. - **#57, comment from Joe (jbwexler):** wants cross-device history where possible. Neither app offers it without a third-party server; Chatbox's sync runs through the vendor. - **#56 Local chat app walkthrough** (solo.ist, 2026-11-18): proposes testing Jan and Chatbox, including streaming on Tinfoil and green models, where history is stored, and telemetry. - **Nathan's comment on #56:** suggests the co-op might support only a customised local app such as Chatbox. Chatbox's mobile reach makes it the stronger base for that; a co-op build would likely need its in-app paid plans removed. ## Resilience and transparency Jan is the more transparent project; Chatbox has the larger mobile footprint and a clearer business model. Neither has community governance; the co-op itself is the only member-governed part of this stack. | | Chatbox | Jan | | --- | --- | --- | | Maker | Chatbox AI, created by developer Benn (Bin) Huang; AlternativeTo lists China as origin | Menlo Research Pte. Ltd., Singapore (formerly Homebrew), founded 2023 by Daniel Ong and Nicole Zhu | | License | Community Edition GPL-3.0; regularly synced from a separate pro repository; AlternativeTo lists the app as "Proprietary (GPL-3.0)" | Apache 2.0 since May 2025 (previously AGPL), with an attribution request | | Public code | github.com/chatboxai/chatbox, about 42k stars | github.com/janhq/jan, about 45k stars, 9.3k commits | | Funding | Not found; revenue from Chatbox AI subscriptions | Described in one secondary source as bootstrapped and employee-owned; not confirmed | | Business model | Free client; sells Lite, Pro and Pro+ plans for its own hosted models | No paid tier and no markup on cloud use | | Governance | Company-run; GitHub issues | Company-run; GitHub issues, Discord, CONTRIBUTING, Code of Conduct | | Strategic risk | The free client's interests can pull towards selling Chatbox AI, which competes with the co-op | Menlo now calls itself an R&D lab "building the brain for robots"; Jan could become a side project | The co-op's exposure is small either way: both use a standard OpenAI-compatible connection, so members can switch apps in minutes, and both licenses let the co-op fork if a maker changes course. ## Open questions for the co-op to test - [ ] Are the Chatbox apps in the App Store and Google Play built from the GPL Community Edition or from the pro codebase? - [ ] Does Chatbox's free tier send any analytics or crash reports? Its data-storage FAQ doesn't say. - [ ] Can Chatbox fetch the co-op's model list from `/v1/models`, or must members always type IDs? - [ ] Do Tinfoil and green models stream correctly in both apps (issue #56's test list)? - [ ] Exact host format in Chatbox: `https://gateway.inference.coop` or with `/v1`? - [ ] Which features of a co-op-customised Chatbox would need changing: removing Chatbox AI plans, pre-filling the gateway, listing co-op models? - [ ] Could either app sync history through infrastructure the co-op controls, to answer the cross-device request in #57? ## Sources - Inference Co-op: [charter](https://git.inference.coop/co-op/docs/src/branch/main/governance/charter.md), [API access](https://git.inference.coop/co-op/docs/src/branch/main/guides/api-access.md), [models](https://git.inference.coop/co-op/docs/src/branch/main/guides/models.md), [issue #56](https://git.inference.coop/co-op/support/issues/56), [issue #57](https://git.inference.coop/co-op/support/issues/57) - Chatbox: [GitHub (Community Edition)](https://github.com/chatboxai/chatbox), [adding providers](https://docs.chatboxai.app/en/guides/providers), [data storage FAQ](https://docs.chatboxai.app/en/faq/faq/data-storage), [pricing](https://chatboxai.app/en/pricing), [PromptQuorum review, Sep 2026](https://www.promptquorum.com/local-llms/chatbox-review), [AlternativeTo listing](https://alternativeto.net/software/chatbox-ai/about/) - Jan: [custom endpoints](https://docs.jan.ai/desktop/remote-models/custom-endpoint), [changelog](https://www.jan.ai/changelog), [GitHub](https://github.com/janhq/jan), [Homebrew is now Menlo Research](https://menlo.ai/blog/homebrew-to-menlo), [RobotHub profile](https://www.robothub.app/en/companies/menloai), [VDF AI comparison, Oct 2026](https://vdf.ai/blog/jan-vs-lm-studio/)