Cherry Studio and Chatbox are both free, open-source desktop AI clients that run on your own API key, and the one difference that decides between them is protocol: Chatbox documents two custom-provider modes, OpenAI-compatible and Gemini-compatible, and lets its agent run on that same key, while Cherry Studio speaks more protocols yet gates its agent behind an Anthropic-compatible /v1/messages endpoint. Pick Chatbox for reach and one simple path; pick Cherry Studio for protocol breadth and its coding-CLI manager. Neither vendor bills you for the app itself.
Both projects are active and independently maintained. Cherry Studio's latest release is v2.1.0, published September 18, 2026, under AGPL-3.0. Chatbox's latest is v1.23.3, published September 16, 2026, under GPL-3.0. Both checked against the GitHub API on September 19, 2026; neither repository is archived.
First, make sure you have the right two products
This comparison has an unusually bad name-collision problem, and one half of it involves real prices that belong to someone else.
| What you might land on | What it actually is |
|---|---|
| "ChatBox - AI Chat with Chatbot" on the iOS App Store | A listing from FLORATE LIMITED (florate.io) with $6.99–$8.99 weekly and $19.99–$59.99 yearly in-app purchases. A different company; those prices are not Chatbox AI's. |
| "chatbox pricing" results about live-chat widgets | "Chatbox" is also a generic UX noun for the support widget on a website. Customer-support vendor roundups are a different market; their numbers do not transfer. |
| Other businesses trading as "Cherry" | The name is shared with design and creative studios and with the peripherals maker Cherry. None of them is the AI client; check that the repository owner is CherryHQ. |
| cherry-ai.com, docs.cherry-ai.com and enterprise.cherry-ai.com | All three redirect to the cherryai.com.cn host: the site to www.cherryai.com.cn, the docs to docs.cherryai.com.cn, the enterprise site to enterprise.cherryai.com.cn. Older guides and link lists still point at the previous hosts, so a redirect is not evidence the project moved on. |
| CherryIN | A third-party gateway that Cherry's own provider table names as an example of the Anthropic-compatible type. Whether it is operated by the Cherry Studio team is not established, so treat it as a third party and do not read its pricing as Cherry Studio's. |
The correct products are CherryHQ/cherry-studio with documentation at docs.cherryai.com.cn, and chatboxai/chatbox with chatboxai.app.
Who should pick which
| If this is you | Pick | The deciding reason |
|---|---|---|
| You chat on a laptop, a phone and a borrowed browser | Chatbox | Its site lists Windows, macOS, Linux, iOS, Android and web. Cherry Studio's mobile repository still carries an in-development status badge and its only current releases are v0.1.0 beta prereleases, so the stable Cherry product is the desktop app. |
| You want the assistant to run commands and edit files | Either, but the permission models differ | Chatbox's Work Mode is desktop-only, needs a tool-use model, and defaults to "Approve" — it asks before running commands or changing files, with a Full Access switch its own docs call "fast but risky". Cherry's agent is subject to the protocol gate below. |
| Your endpoint is an OpenAI-compatible gateway and you want agents | Chatbox | Cherry's provider table says Cherry Agent requires the Anthropic-compatible type. Chatbox documents an OpenAI-compatible custom mode and Work Mode runs on it. |
| You want one app to manage Claude Code, Codex, Gemini CLI and friends | Cherry Studio | Its Code CLI manager installs and configures external coding CLIs from inside the app and will not overwrite system-installed copies. Chatbox documents no equivalent. |
| You juggle several gateway keys | Cherry Studio | One provider entry accepts a comma-separated key list and cycles through it. Chatbox's documented provider config carries a single apiKey. |
| You are moving between the two | Migration is cheap either way | The transferable asset is the key and the model ids, and both clients take those by hand in a minute. What does not transfer is chat history: Cherry's cross-device story is full backup and restore, and Chatbox stores conversations on-device. |
What each one costs
Neither app charges for itself. Cherry Studio has no consumer subscription of any kind; its README states the Community Edition is AGPL-3.0, that commercial use is permitted under that licence, and that an exemption requires a commercial licence from the vendor. Its Enterprise Edition is sold separately and publishes no price this check could read: the README lists its cost only as a buyout or subscription fee and routes buyers to a quote request, and the enterprise site's pricing URL serves a client-rendered page with no price in it. No seat price, currency or tier name is quoted here. Treat it as quote-only and confirm with the vendor.
Chatbox's paid plans buy a hosted model service, not the client.
| Chatbox AI plan | Monthly | Billed annually | Compute points | AI images |
|---|---|---|---|---|
| Free | $0 | $0 | 10,000, daily reset | 5 |
| Lite | $3.99 | $3.50 / month | 2,000,000 / month | 25 |
| Pro | $19.99 | $16.70 / month | 12,000,000 / month | 100 |
| Pro+ | $39.99 | $33.33 / month | 30,000,000 / month | 200 |
Read from chatboxai.app/en/pricing on September 19, 2026. Advanced models require Pro or above — the Lite row is marked standard models only — and every tier row lists a five-device limit. The page does not state a currency, and whether unused points roll over is not documented anywhere found. Note that the plans guide omits the Free column that the pricing page shows, so read the tiers from the pricing page.
One trap in that table: at least some of its rows describe what the hosted Chatbox AI service covers rather than what the app unlocks. Chatbox's knowledge-base documentation says paid users can pick Chatbox AI as the knowledge base's model provider and that free users can also use their own API — so that row is about whose models you use, not about access to the feature. Do not generalise the pattern, though: the Free column marks Work Mode "Trial only", and nothing in the Work Mode documentation reconciles the two, so that row stays unresolved.
Chatbox's own plans guide converts points to tokens — Pro's 12M points as roughly 6.7M tokens of one Claude model — which, against Pro's $19.99 monthly price, works out to about $2.98 per million blended tokens. Treat that as derived arithmetic rather than a rate card: the input/output mix behind it is not disclosed, and the conversion table names an older model set than the pricing page's own current roster, so the two pages are not a coherent snapshot.
The bring-your-own-key boundary, field by field
Both clients accept a third-party OpenAI-compatible endpoint. The URL fields are shaped differently, and each client invites its own mistake.
| Question | Cherry Studio | Chatbox |
|---|---|---|
| Where you add it | 设置 → 模型服务 → 添加服务商 | Settings → Model Providers → Add |
| URL field shape | One API address; the app appends the route | An API Host plus a separate API Path |
Does /v1 go in the host? | Enter the base; Cherry appends /v1/chat/completions. Ending the address with # makes it verbatim instead. | No. The path already defaults to /v1/chat/completions, so the host is the origin. |
| The failure this causes | A # on an ordinary endpoint freezes the path and breaks the route | A host ending in /v1 yields /v1/v1/chat/completions — a 404 that reads like an auth error |
| Documented protocol modes | OpenAI-compatible, Anthropic-compatible, Gemini, Bedrock, Azure OpenAI, local inference, and special gateways | OpenAI-compatible and Gemini-compatible in the provider docs. The import schema is narrower still: its type takes "openai" only, and says more API types are planned. |
| Model ids | Added from the 获取模型列表 button, or typed by hand | Typed per model, with capability checkboxes — unconfigured models are treated as plain text, so vision and tool use must be ticked manually |
| Several keys on one provider | Yes, comma-separated and rotated front to back | One apiKey per provider in the documented config |
| One-click provider import | None documented | chatbox://provider/import?config= with a base64 config, from version 1.15.1 |
Cherry field rules from the provider settings page and the custom-provider page; Chatbox field semantics from its model-configuration guide, which names the OpenAI-compatible and Gemini-compatible options and says the API Path "generally doesn't need to be filled in, defaulting to /v1/chat/completions", and from its import schema, which carries apiHost, an optional apiPath with that same default, and a single optional apiKey. Both checked September 19, 2026. One caveat worth stating rather than papering over: a widely repeated community rule says a Cherry address ending in / behaves as a third case. That branch could not be pinned to official text in this check, so use no trailing slash and no trailing #, which is safe under the two branches the docs do state.
The one structural difference: Cherry's agent is protocol-gated
Cherry Studio classifies providers by wire protocol, and its provider table states in the Anthropic-compatible row, served at /v1/messages, that Cherry Agent requires this type. Two repository issues describe the same boundary from the user side: #15549, reporting that agent mode fails against an OpenAI-compatible gateway because it routes through an Anthropic-style /v1/messages path, and #15168, asking for OpenAI-compatible models in Cherry's agent. Read them as corroborating reports, not as rulings: both were closed as not planned by the repository's inactivity bot, with no maintainer reply on either. Cherry's agent-setup documentation does not restate the requirement, so this claim rests on the provider table, with those two reports as supporting evidence.
The practical consequence is small but easy to get wrong: a Cherry custom provider holds separate OpenAI and Anthropic base URLs, with OpenAI-Responses, Gemini and image base URLs behind the dialog's extra-endpoint toggle, which the documentation page writes as More Settings and the v2.1.0 client labels 更多端点 (More Endpoints). Chat working over the OpenAI slot tells you nothing about agent mode. If you want Cherry's agent, fill the Anthropic slot too, with an endpoint that genuinely serves /v1/messages. Kunavo implements that route — the Claude Code setup depends on it — but Cherry Agent has not been runtime-tested against it here, so treat this as what Cherry requires, not as a promise that the pairing works. The same protocol constraint propagates to the Code CLI manager, which filters its picker by the interface type each CLI needs; which CLI maps to which type is not enumerated in the docs.
Chatbox documents no equivalent protocol gate. Its Work Mode overview names one model requirement — a model that supports tool use — and no protocol requirement beyond the provider modes above; what it does not do is run on mobile or web, which that page states directly.
A worked monthly estimate on one key
This is illustrative token arithmetic, not a measured bill and not a ceiling. Assume the same daily chat habit in either client: twenty working days, and per day 250,000 uncached input tokens across accumulated conversation context plus 30,000 output tokens — so 5,000,000 input and 600,000 output in the month. Rates are live Kunavo catalog prices per million tokens; the client fee is $0 in both apps.
| Model | Input / output per 1M | Estimated month |
|---|---|---|
| Claude Haiku 4.5 | $0.40 / $2.00 | $3.20 |
| Gemini 3.8 Flash | $0.525 / $2.625 | $4.20 |
| GPT-5.6 Terra | $0.70 / $4.20 | $6.02 |
| Claude Sonnet 4.6 | $1.20 / $6.00 | $9.60 |
Under these assumptions the cheap lane lands near $3.20 on Claude Haiku 4.5 while the same month on Claude Sonnet 4.6 is $9.60. Scale by your own habit before treating either as a budget, and note that the cheapest listed rate and the lowest cost to finish your work are different questions — a model that needs a second and third attempt can cost more than a pricier one that answers once. AI cost optimization covers measuring that on your own traffic.
Two boundaries. Kunavo's catalog amount is a billing floor rather than a cap: when the upstream reports its charge, the bill is the greater of catalog cost and upstream cost times the applicable markup. And the minimum top-up is $10 in prepaid credit — a funding minimum, not a task fee or a subscription. See billing details. Cache charges and external tools sit outside this example.
One gap to plan around if the knowledge base is why you are choosing a client: a knowledge base needs an embedding model to index documents, and Kunavo does not serve embedding models. That step runs locally or against an outside provider, on its own key, whichever client you pick.
Where your data sits
Chatbox's data-storage FAQ says chat messages and settings are stored on your device and that it will not upload data to any server without permission; using the hosted Chatbox AI service does transit messages to Chatbox's servers, while a bring-your-own key sends them straight to your provider. Cherry Studio documents no first-party account-based cloud sync — the issue requesting an account system and cross-device sync is still open — so its documented cross-device path is backup and restore to a local directory, WebDAV or S3-compatible object storage. That is a real operational difference, though note what is and is not established here: Cherry's documented cross-device path is a file you move yourself, while Chatbox at least runs on the second device. Whether Chatbox replicates conversation history between devices was not verified in this check.
Set either one up
Kunavo publishes a configuration guide for both clients, and each one prints the field shape that client actually wants rather than a generic base URL. These are published setup references, not compatibility tests: neither app has been runtime-tested against Kunavo's endpoint here. Start at the Cherry Studio integration guide or the Chatbox integration guide, run one bounded conversation, then read the charge your account recorded for it. Create a Kunavo account when you are ready to fund a key.
Comparing self-hosted web UIs instead of desktop apps? LibreChat vs Open WebUI and AnythingLLM vs Open WebUI run the same comparison one layer over, and the OpenAI-compatible API guide covers the field shapes that trip up every client on this list.
FAQ
Is Cherry Studio better than Chatbox?
Only for a specific kind of work. Cherry Studio is the better choice when you need more than one wire protocol behind a single provider entry, when you want the app to install and configure external coding CLIs for you, or when you want its agent workspace. Chatbox is the better choice when you want the same client on desktop, phone and web, and when you want its agent to run on a bring-your-own-key setup with no protocol caveat. Both apps are free and both read your own API key, so the honest framing is workflow fit, not a ranking.
Is Chatbox better than Cherry Studio?
Chatbox wins on reach and on simplicity. Its own site lists Windows, macOS, Linux, iOS, Android and web, while Cherry Studio's mobile client repository still carries an in-development status badge and its only current releases are v0.1.0 beta prereleases, so the stable Cherry product is the desktop app. Chatbox's custom-provider docs name two modes, OpenAI-compatible and Gemini-compatible, and its Work Mode agent runs on that same key. Cherry Studio wins when you need its wider protocol surface or its coding-CLI manager. Checked September 19, 2026.
Are Cherry Studio and Chatbox free?
Both clients are free to download and use with your own API key. Cherry Studio's Community Edition is AGPL-3.0 and has no consumer subscription at all; its only paid track is a separately sold Enterprise Edition whose price is not published. Chatbox's Community Edition is GPL-3.0, and its paid Chatbox AI plans -- Free at $0, Lite at $3.99 per month, Pro at $19.99 and Pro+ at $39.99, checked September 19, 2026 -- buy a hosted model service, not the app. If you bring your own key you pay neither vendor.
Can I use one API key in both Cherry Studio and Chatbox?
Yes, if the endpoint is OpenAI-compatible, because that is a mode both clients document for a third-party provider. The field shapes differ and that is where setups break. Chatbox splits the endpoint into an API Host and an API Path, and the path already defaults to /v1/chat/completions -- so the host is the bare origin, and a host ending in /v1 produces /v1/v1/chat/completions. Cherry Studio takes one API address and appends the route itself, so you enter the base and let it append. Copying the same string into both fields is the most common way to get a 404 that reads like an authentication failure.
Does Cherry Studio's agent work with an OpenAI-compatible gateway?
Cherry Studio's own provider-classification table states that the Anthropic-compatible type, served at /v1/messages, is the one Cherry Agent requires. Two reports in the repository describe the same boundary from the user side: one that agent mode fails against an OpenAI-compatible gateway because it routes through an Anthropic-style /v1/messages path, and one asking for OpenAI-compatible models in Cherry's agent. Both were closed as not planned by the repository's inactivity bot rather than answered by a maintainer, so they corroborate the reported behaviour without adding an official statement. So chat over an OpenAI base URL is not evidence that agent mode will run. A Cherry custom provider holds separate OpenAI and Anthropic base URLs, so the practical answer is to fill both slots with an endpoint that really serves /v1/messages. Checked September 19, 2026.
Do I need a Chatbox subscription to use its knowledge base or agent mode?
Not for the knowledge base. Chatbox's knowledge-base documentation says that paid users can choose Chatbox AI as the knowledge base's model provider and that free users can also use their own API. Work Mode is less clear-cut: its overview page states only a desktop and tool-use-model requirement and never names a plan, while the pricing table's Free column marks Work Mode 'Trial only'. Those two vendor pages are not reconciled anywhere we could find, so treat a free account's long-run Work Mode access as unsettled rather than granted. Note separately that a knowledge base needs an embedding model, and Kunavo does not serve embedding models.
Which Chatbox app on the App Store is the real one?
Check the developer and the website, not the name. A prominent App Store listing called 'ChatBox - AI Chat with Chatbot' is published by FLORATE LIMITED, whose site is florate.io and whose in-app purchases run $6.99 to $8.99 weekly and $19.99 to $59.99 yearly. That is a different company and those are not Chatbox AI's prices. The product this page compares is chatboxai.app, built from github.com/chatboxai/chatbox.
Checked September 19, 2026 and re-checked September 21, 2026: both repositories, licences and latest releases via the GitHub API, plus the mobile repository's releases and the two agent issues, whose closures were made by the repository's inactivity bot; Cherry Studio's provider-type table, provider settings page, custom-provider page, Code CLI page and data-settings page on docs.cherryai.com.cn; Chatbox's pricing page, plans guide, model-configuration guide, provider import schema, knowledge-base doc, Work Mode overview and data-storage FAQ. Cherry Studio Enterprise pricing, the trailing-slash address branch, whether a free Chatbox account keeps Work Mode beyond a trial, whether Chatbox replicates history between devices, and whether Cherry Agent works against Kunavo's endpoint were not established and are marked as such above. Kunavo token rates come from the live catalog; every dollar example here is illustrative token arithmetic rather than a measured task cost.