ChatWise vs Chatbox: client fees, subscriptions and BYOK

ChatWise and Chatbox charge for opposite things. ChatWise sells the desktop client once — Pro is a one-time $29 against a $49 crossed-out list price — and sells no model access at all. Chatbox gives the client away on every platform and sells model access as a subscription from $3.50 a month on annual billing ($3.99 month-to-month). Both accept your own API key as a documented path, so if you are bringing a key the client fee is the fee that differs most — though a few Chatbox features stay behind its own subscription even when the tokens are billed to your key, as the plan section below sets out. Prices read from ChatWise's pricing page and Chatbox's pricing page on September 19, 2026.

Identity first, because both names collide. ChatWise here means the desktop AI chat client at chatwise.app, closed source, made by a single developer — its pricing FAQ credits "Kevin", and chatwise.app/github redirects to the egoist/chatwise-releases repository, which publishes builds rather than source. It is not the App Store app "ChatWise: Chat with Ask AI" from a different seller, and not the UK social network at chatwise.co.uk. Chatbox here means Chatbox AI at chatboxai.app, whose Community Edition source lives at chatboxai/chatbox — GPL-3.0, more than 41,000 stars, not archived, latest release v1.23.3 published September 16, 2026 per the GitHub API. The older Bin-Huang/chatbox URL now redirects there. It is not the generic chat-widget sense of the word, not ChatGPTBox, and not ChatHub.

Chatwise vs Chatbox: which one should you pick

Read the same question from either direction — ChatWise vs Chatbox, or Chatbox vs ChatWise — and it resolves on the same five axes: platform, execution model, protocol, plan gates and what it costs to change your mind. Pick Chatbox if the same conversations have to be on your phone. That is the one difference neither side can configure away. Chatbox ships desktop, iOS, Android and web; ChatWise's own pricing FAQ says there is no plan to release a mobile app at the moment, and that it stores your data locally only with cloud sync merely planned. Pick ChatWise if you want the client fee to end. One $29 payment buys the current Pro features permanently; there is no month in which the client vendor bills you again unless you choose to renew for another year of feature updates.

Execution model. If you will let the model run commands, read the two agent modes before the price. ChatWise's Agent mode gives the model five built-in tools — Bash, Write, Read, Edit and AskUserQuestion — loads Claude Code-style skills from ~/.claude/skills and ~/.agents/skills, and states plainly that there is no sandboxing at the moment. It also disables MCP and Artifacts while in the preview, so on ChatWise today agent mode and MCP tools are mutually exclusive. Chatbox's Work Mode runs Node and Bash in a local sandbox, supports MCP servers and agentskills.io skills, automatically pauses after 25 consecutive tool calls, and locks the conversation into Work Mode once the model has started. It is desktop-only; the docs say the web and mobile apps offer regular chat only. A sandbox and a tool-call pause are a materially different risk posture from an unsandboxed Bash tool.

Protocol. ChatWise's custom-provider page says it can integrate any OpenAI-compatible or Anthropic-compatible service. Chatbox's documented custom types are OpenAI API Compatible and Google Gemini Compatible; no Anthropic-native custom type is documented, and its Claude setup guide tells you to keep the API Host at its default https://api.anthropic.com rather than pointing it somewhere else. So if you specifically want to speak the Anthropic Messages protocol to a third-party host, ChatWise documents the capability and Chatbox does not document it — an undocumented path, not a demonstrated refusal by the app. See the Anthropic base URL reference for what that surface involves.

Migration cost. ChatWise Pro is explicitly non-refundable; its pricing page says the 7-day card-free trial is the reason refunds are not offered, so use the trial as the decision window, not the purchase. Chatbox is a subscription you can stop, and its billing and subscription page documents both halves: auto-renewal is cancelled through the Stripe customer portal, with the service and quota retained to the end of the paid cycle, and a refund is requested through its support platform, which the page says shows eligibility and the amount there rather than stating them as a published policy. What is genuinely absent is an annual total — neither the pricing page nor the plans document publishes one, only a per-month figure under annual billing. Confirm the billed-upfront amount in checkout rather than multiplying the monthly figure yourself.

What each vendor actually charges

Sources: chatwise.app/pricing, chatboxai.app/en/pricing and the Chatbox licence page, all read September 19, 2026. The ChatWise page carries a "Discounts ending soon!" badge with no end date, so treat $29 as a promotional price on that date and $49 as the stated list. It shows a bare dollar sign with no currency code and no tax statement.

Chatbox AI plans, and why this page does not convert them to a token price

Read from chatboxai.app/en/pricing on September 19, 2026. Compute points are a Chatbox-AI-only billing unit — the tokens page says each model has its own compute-point-per-token ratio and that BYOK users read their usage in the provider's dashboard instead. This page does not convert compute points into a dollar-per-token figure, because the per-model ratios were not read, so any such conversion would be invented. The gates above are the rows of the pricing page's own comparison matrix; web search, document and image analysis and the 5-device allowance are ticked on all four plans there, including Free. Note that two official Chatbox pages diverge: the plans document lists different model families and uses different feature names — Document Chat and Document Preprocessing, both marked Pro and above — and has no Free row at all. Take the pricing page as current and confirm in the app.

One gate is documented in a form that does not depend on whose key pays for the tokens: the Work Mode configuration page gives its three web-search providers a Requires column, and the built-in Chatbox AI one reads "A paid Chatbox AI license", against "Free (limited, subject to Microsoft)" for Bing Search — which the same table marks as not reading page content — and "Your own API key" for Tavily. The Knowledge Base is the one worth testing yourself: the pricing matrix marks it Pro and above, while the older knowledge-base tutorial says paid users may pick Chatbox AI as its model provider and free users may use their own API. Those two are a real disagreement, and ten minutes on a free install settles it. The Work Mode docs say nothing either way about plan gating for the Knowledge Base or MCP, which is an unanswered question rather than a denial.

Kunavo serves no embedding model. A Chatbox Knowledge Base needs one to index documents, so that step is not something a Kunavo key covers — it runs on whatever embedding provider you configure in Chatbox. The feature itself is desktop-only and capped at 50 MB per file, per Chatbox's service FAQ.

Pointing either client at your own endpoint

Chatbox splits the endpoint into two fields, and that is the trap. Go to Settings → Model Provider → Add → Add Custom Provider, name it, and choose OpenAI API Compatible. Then fill API Key, API Host and API Path. Two live official conventions exist and both resolve to the same URL: the current ZenMux provider guide puts the /v1/ in the API Host and /chat/completions in the API Path, while the older generic providers page says the API Path usually needs no value because it defaults to /v1/chat/completions. The binding rule on both: the full request URL must contain /v1/chat/completions exactly once. Kunavo's own Chatbox setup page follows the older convention — host https://api.kunavo.com with the path left empty — which is why a host ending in /v1 produces /v1/v1/chat/completions and fails. Models are then added under Models → New with a Model ID, nickname, type, context window and max output tokens, and a Check button tests the connection.

ChatWise is Settings → Providers → "+", then a preset or a custom configuration, then key and endpoint settings. Models can be fetched automatically if the endpoint serves a /models route, or typed in by hand as Model ID, display name, capability flags for image and video or audio input, a reasoning flag, and a context length. Hand-declared metadata is a quiet hazard on this path: nothing is discovered, so a wrong capability flag or context length is simply wrong until you notice. This page deliberately publishes no base URL for ChatWise. Its custom-provider documentation never shows a complete example URL, so whether it appends /v1 to the host or expects the /v1 root typed in could not be established — read the field label in the app and confirm the first request lands before you rely on either shape. Both clients are configuration targets here, not tested integrations: neither was installed and pointed at Kunavo for this page. For the general shape of that endpoint see the OpenAI-compatible API guide.

A worked estimate for the shared part of the bill

Whichever client you pick, a bring-your-own-key setup bills the same tokens. These figures are illustrative token arithmetic, not measured task costs and not a bill ceiling. The light column assumes a month of chat totalling 3,000,000 input tokens and 250,000 output tokens; the heavy column assumes 12,000,000 input and 1,000,000 output. Input dominates in a chat client because every turn re-sends the transcript so far. Rates are live Kunavo catalog prices per million tokens.

Read these against the subscription column, not instead of it. Under the light assumption, Claude Haiku 4.5 at $1.70 sits below Chatbox AI Lite at $3.50 a month, while Claude Sonnet 5 at $8.50 sits between Lite and Pro. Under the heavy assumption Claude Sonnet 5 reaches $34.00, above Pro's $16.70. Those are dollar-to-dollar comparisons and nothing more. A Chatbox AI plan is not a flat all-you-can-use fee: it buys a monthly compute-point quota, and because this page does not convert points to tokens, nothing here establishes whether Pro's quota would actually cover the heavy month above. So the defensible statement is the narrow one: of the four models priced here, the three cheaper ones come in under Lite's $3.50 for the light month, and only Claude Sonnet 5 exceeds Pro's $16.70 for the heavy one. The two sides are not like-for-like anyway, since a subscription also buys the licence-gated features listed above. Cheapest listed rate and lowest cost to finish your work are different questions; a model that needs a second attempt can cost more than a dearer one that does not.

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. Cache charges and external tools sit outside this example. The minimum Kunavo top-up is $10 in prepaid credit — a funding minimum, not a task fee or a subscription. See billing details.

Feature comparison

Two footnotes on that table. ChatWise's free-tier limits are thinner than they look: Pro advertises "Unlimited Tools (MCP)", which implies a cap on the free tier, but the pricing page does not publish the number, and neither it nor the custom-provider page says whether custom providers work on the free tier at all. And Chatbox's iOS app was removed from the US App Store in 2025 over a third party's trademark claim and later restored — that account comes from Chatbox's own issue, a party to the dispute, so read it as their statement rather than as an adjudicated fact.

If you are bringing a key to either one

Start from the client you have already chosen on platform grounds, then fund one key and run a bounded week before committing to a subscription or a one-time purchase. For Chatbox, the Chatbox integration guide has the exact field values including the API Host trap. Create a Kunavo account when you are ready to fund a key. If you are still shortlisting clients rather than these two, the AI agent API directory covers the wider set, and LibreChat vs Open WebUI compares two self-hosted alternatives with the same free-client, metered-tokens shape.

FAQ

Checked September 19, 2026 against the vendors' own live pages and public APIs: chatwise.app and its pricing page, docs.chatwise.app, the egoist/chatwise-releases feed, chatboxai.app pricing and guide pages, and the chatboxai/chatbox GitHub API. Re-checked September 21, 2026, when the Chatbox billing, plans, Claude setup, knowledge-base and service-FAQ pages were added as sources and the plan-gate rows were corrected against the pricing page's own comparison matrix. Neither client was installed or runtime-tested against Kunavo for this page. Kunavo token rates come from the live catalog, and every dollar example here is illustrative token arithmetic rather than a measured bill.