Chatbox: connect OpenOxAI
Chatbox is a desktop AI client. This walkthrough configures an OpenOxAI Chat Completions provider for models that support that protocol, such as DeepSeek. GPT requires a separate Responses-capable configuration; choose the model service, key and protocol together.
What is Chatbox?
A desktop client with its own hosted service — and a fully usable BYOK mode.
Chatbox is a free AI desktop client from Chatbox AI (Mediocre, LLC). It runs on macOS, Windows and Linux, with companion mobile apps and a web build. Inside the app you can use Chatbox's own hosted models, or bring your own API key from a provider such as OpenOxAI.
This guide uses the second path. OpenOxAI exposes an OpenAI-compatible API, so Chatbox treats it like any other custom provider: one API host, one key, and a model list you control. No Chatbox AI subscription is required, and chats stay in the app on your machine.
Choose a Chat Completions model and a key for its service in this profile. Configure GPT separately through Responses and Claude through Messages. Each conversation must select a model from the matching provider; a shared model picker does not merge their protocols or key permissions.
Why pair Chatbox with OpenOxAI?
-
One key, many model families
Configure OpenOxAI once. Every model that key is allowed to call appears under the same provider in Chatbox's model picker.
-
Desktop workspace, not a chat box
Chat modes, file and image attachments, artifacts, copilots and multiple conversations live in one app.
-
Local-first by default
Custom providers need no Chatbox account. Chat history stays on your device; API requests go directly to OpenOxAI.
-
Switch models mid-project
Keep the conversation and change the model from the composer. Useful for comparing a fast model with a deeper one on the same prompt.
What you need
- Chatbox 1.23.5 (or a later desktop build) for macOS, Windows or Linux.
- An OpenOxAI account with credit on the plan you use.
- An OpenOxAI API key — create it in the dashboard, copy it once.
- At least one exact model ID from openoxai.com/models.
GPT requires a Responses configuration
This illustrated provider uses Chat Completions
OpenOxAI GPT models should use Responses API: base URL https://api.openoxai.com/v1, inference path /v1/responses, and a GPT-service key. The custom OpenAI API Compatible profile shown below uses /v1/chat/completions; it is the Chat recipe, not the default GPT connection.
If your Chatbox version provides a Responses provider, select that protocol explicitly and verify the final request in OpenOxAI Usage. If it does not, use a Responses-capable client such as Codex, OpenCode 2, Continue or Cherry Studio for GPT. Editing API Path alone does not add Responses request construction, streaming or response parsing.
Existing screenshots and test records describe their original model, version and protocol. They do not establish a new GPT Responses test. Claude also requires its matching Messages configuration and service key.
One endpoint, three layers
The provider stores the connection. The model list stores the IDs. The chat pins one model.
Chatbox does not need a special OpenOxAI integration. It sends an ordinary OpenAI Chat Completions request to the host you configure. OpenOxAI authenticates the key, checks that the key can use the requested model, routes the request, and records usage against your account.
That means three things must line up: the provider (host, key, path), the model entry (the exact ID), and the model selected in the chat. If any one is wrong, the request fails before a token is generated.
Click or tap any figure in this guide to open the full-resolution image. Press Esc to close it.
Endpoints for OpenOxAI
- API Host:
https://api.openoxai.com/v1 - API Path:
/chat/completions - Full request URL:
https://api.openoxai.com/v1/chat/completions - Chatbox sends an
Authorization: Bearer <key>header. The key is stored locally by the app.
Model count checked on 27 Oct 2026 at openoxai.com/models. The live model library is the source of truth; IDs and availability can change.
Install Chatbox
Download the desktop build from the official site. The walkthrough uses Chatbox 1.23.5 on macOS.
1. Open https://chatboxai.app/download and choose the macOS build
2. Open the downloaded .dmg
3. Drag Chatbox into Applications
4. Launch it from ApplicationsIf macOS blocks the first launch, allow it in System Settings → Privacy & Security.
1. Open https://chatboxai.app/download and choose Windows
2. Run the downloaded installer
3. Approve the User Account Control prompt
4. Launch Chatbox from the Start menuA portable build is available on the same page if you do not want a system-wide install.
1. Open https://chatboxai.app/download and choose Linux
2. Download the package for your distribution
3. Install it with your package manager, or make the AppImage executable
4. Launch Chatbox from your application menuDesktop integration varies by distribution; the app works without a tray icon.
Use the desktop app
The Chatbox web build runs inside the browser and cannot reliably call arbitrary OpenAI-compatible hosts because of CORS. OpenOxAI's API is designed for clients, so use the macOS, Windows or Linux desktop app for this setup. The mobile apps exist, but their menus differ from the desktop flow below.
First launch
Chatbox opens a short setup guide on a clean install. Close it and open Settings (the gear icon in the lower-left corner). Everything in this guide lives under Settings → Model Provider. You can reopen the welcome guide later from Help.
Get an OpenOxAI API key
One key powers Chatbox. Choose the right group when you create it — that choice affects routing and cost.
Sign in or create an OpenOxAI account
Go to openoxai.com/register and complete sign-up, or sign in at openoxai.com/login. Add credit to the plan you intend to use before sending production traffic.
Open the API Keys page
In the dashboard, open API Keys. The direct URL is openoxai.com/dashboard/keys. You should see the token list — not the wallet or model page.
Create the token and choose a group
Click New Token, give it a name such as chatbox-desktop, and pick a
group for your main workload. OpenOxAI uses groups to route Claude traffic; GPT traffic uses the
GPT exclusive group and is billed by input, output and cache reads.
| Group | Best for | What to expect |
|---|---|---|
| special | Daily Claude, first setup | Default starting point; lowest-cost routing for everyday work and stable batch calls. |
| reverse | Latency-sensitive chat | Prioritizes lower latency and higher throughput while keeping a cost advantage. |
| direct | Production and long context | More direct routing for stable behavior and consistency on long-context tasks. |
| GPT exclusive | GPT models | Used for GPT-family calls; billing follows input, output and cache-read token prices. |
Select the service that matches your model. Create separate keys and provider profiles for GPT, DeepSeek and Claude; GPT uses Responses and Claude uses Messages.
Copy the key once and store it safely
The full key is shown when the token is created. Copy it immediately into a password manager or the client's credential store. If you lose it, create a new token rather than reusing a key from a screenshot or chat message.
Never put a real key in a screenshot
No screenshot in this guide exposes a full key; Chatbox masks the credential field. The temporary key
used for the live hi verification was revoked after capture. Use the same discipline in
tickets, recordings and public repositories: a key is a bearer credential, and anyone who has it can
spend against your account until you revoke it.
Model catalogue
Check IDs, context length and live reference prices before you add a model in Chatbox.
Usage and rates
Pay-as-you-go billing is metered per token. Your dashboard shows requests, tokens and cost.
Add OpenOxAI to Chatbox
Chatbox has no built-in OpenOxAI preset, so you create a custom OpenAI-compatible provider.
Open Settings → Model Provider
Click the gear icon in the lower-left corner, choose Model Provider, then click Add at the bottom of the provider column. The provider chooser opens with a search field.
Type custom to jump to the custom section instead of scrolling through dozens of presets.
Choose “Add Custom Provider”
Select Add Custom Provider under Custom Additions. Do not pick the OpenAI
preset: it points at api.openai.com and would require you to overwrite OpenAI's own
configuration. A custom provider keeps OpenOxAI separate and easy to switch.
Name the provider and pick the protocol
Enter OpenOxAI as the name. Leave API Mode on OpenAI API Compatible — OpenOxAI's Chatbox integration uses the OpenAI Chat Completions protocol. Click Add.
Fill in the key, host and path
Chatbox saves the provider fields as you type. Paste the key from step 04, then enter the two URL fields exactly as shown below.
| Field | Value | Notes |
|---|---|---|
| API Key | sk-openox-… | The key you created. Use the paste button; avoid leading or trailing spaces. |
| API Host | https://api.openoxai.com/v1 | Includes /v1. No trailing slash is needed. |
| API Path | /chat/completions | Starts with a slash. Chatbox appends it to the host. |
Check the preview, not the field width
The API Host input is narrower than the URL, so the beginning of the value may scroll out of
view. Trust the line under the fields: it should read exactly
https://api.openoxai.com/v1/chat/completions.
/v1, one /chat/completions. A second /v1 is the most common 404.Confirm the provider is enabled
A green dot next to OpenOxAI in the provider column means the provider is active. There is no separate “Save” button on this page; Chatbox persists changes as you edit. If the dot is grey, click the provider row once and confirm the key and host are present.
Leave Improve Network Compatibility off unless you are on a network where requests stall. It can add latency, so enable it only for troubleshooting.
Add exact OpenOxAI model IDs
A provider with no models shows an empty picker. Add each ID you want to use — spelling matters.
Click “New” in the Model section
On the OpenOxAI provider page, scroll to Model and click New. Chatbox opens the model editor. The same editor is used to change an existing model later.
Enter the ID, nickname and limits
Paste the model ID exactly as it appears on the OpenOxAI model page. The nickname is only a label inside Chatbox; it never reaches the API.
| Field | Example | Why it matters |
|---|---|---|
| Model ID | claude-sonnet-5-5 | Sent to OpenOxAI. Aliases, display names and provider prefixes are not interchangeable. |
| Nickname | Claude Sonnet 5.5 | Shown in Chatbox only. Use a label you can scan quickly. |
| Model Type | Chat | Keep Chat for text and multimodal chat models. |
| Capabilities | Off by default | Enable Vision, Reasoning or Tool use only when the selected model actually supports them. |
| Context Window | 1000000 | Client-side budget. Use the model's documented context length. |
| Max Output Tokens | 64000 | Client-side cap for one reply. This is not the model's full context window. |
Context is not output
A 1M context window describes everything the model can read in one request. Max Output Tokens caps only the reply. Setting the output cap to the full context size is wrong and can cause provider-side errors or surprising cost. The 64,000 figure above is the client cap used in this walkthrough; lower it if a model rejects the request, and check the model page before raising it.
Add two or three models, then save
A single model is enough to start, but adding two makes the picker and the model switch workflow obvious. Click Save after each entry. The model appears in the provider's Model list with its context and output limits.
Fetch can query /v1/models when the key is valid and populate the
list automatically. If Fetch returns an error, add IDs manually — the rest of the setup still works.
Recommended IDs
Captured from openoxai.com/models on 27 Oct 2026. Context values are listed on each model card.
| Model | Model ID | Context | Good for |
|---|---|---|---|
| Claude Sonnet 5.5 | claude-sonnet-5-5 | 1M | Balanced default for coding, writing and analysis. |
| Claude Opus 5.5 | claude-opus-5-5 | 1M | Harder reasoning and long-form work where quality matters more than speed. |
| GPT-6.1 Sol | gpt-6.1-sol | See model library | Use a separate Responses provider, not this Chat Completions profile. |
| Gemini 3.8 Flash | gemini-3.8-flash | 1.05M | Fast long-context work and high-volume chat. |
| DeepSeek V4.1 Flash | deepseek-v4.1-flash | 1M | Cost-efficient reasoning and coding. |
| GLM 5.3 | glm-5.3 | 1M | Agent-style workflows and coding. |
| Kimi K3 | kimi-k3 | 1.05M | Long documents, research and codebase reading. |
| Grok 4.7 | grok-4.7 | 500K | Reasoning and conversational workloads. |
The ID is the contract
claude-sonnet-5-5, Claude Sonnet 5.5 and
anthropic/claude-sonnet-5.5 are three different strings. Only the first is the ID in this
example. If the ID is wrong, OpenOxAI returns a model-not-found error even when the key is perfect.
Pick a model and start
The provider and model list are ready. Now pin a model to a conversation.
Create a new chat
Close Settings and click New Chat. The composer opens with the last-used model or a Select Model button. Either way, open the model selector in the composer.
Find OpenOxAI under “More Providers”
Scroll the picker past Chatbox AI and the other presets to the More Providers section. OpenOxAI appears with a green dot and the number of models you added. Expand it and choose a model.
DeepSeek V4.1 Flash.Send hi and wait for a reply
Type exactly hi and press Enter. This is the smallest end-to-end test: one request
exercises the host, path, key, key group, model ID and model permission. Do not move on until an
assistant reply appears.
The model tag under the reply should show OpenOxAI (model nickname).
hi sent. ② The assistant replies. ③ The OpenOxAI model label and token count. This is the pass condition: a normal reply, not a saved provider and not an error response.
The captured run uses deepseek-v4.1-flash, a low-cost test model. Any ID from section 06
follows the same request path. The English system prompt at the top of the chat is only for this
walkthrough; it is not required to connect.
Verification rule
A saved provider, a green dot and a model in the picker do not prove that the
connection works. An authentication error, an empty bubble or an error panel is a failure. The only pass
condition is a normal assistant reply to hi.
Pass condition: a reply to hi
A correct setup returns a normal assistant message. Check all four signals: a response bubble instead of an error panel, the OpenOxAI provider label under the message, the model nickname you created, and token counts in the message footer. On the OpenOxAI side, the request appears in Usage with the model and completion status.
If any of those are missing — especially if you see an authentication error — the setup is not connected. Fix it with the troubleshooting section before publishing, sharing or relying on the configuration.
Switch models at any time from the same selector. Chatbox keeps the conversation history, so you can ask the same prompt of Sonnet and Opus and compare the answers without starting over.
Fix the request before changing the client
Check the full preview URL first, then the key, then the model ID. Most failures are one of those three.
404 or “not found” on every request
The path is wrong, or /v1 appears twice. Look at the live preview under the provider fields.
- Correct:
https://api.openoxai.com/v1/chat/completions - Wrong:
https://api.openoxai.com/v1/v1/chat/completions— the host already includes/v1. - Wrong:
https://api.openoxai.com/chat/completions— the host is missing/v1. - Wrong:
https://api.openoxai.com/v1/chat/completions/with a trailing slash in the path — remove it if your build appends it literally.
Set API Host to https://api.openoxai.com/v1 and API Path to /chat/completions. Do not paste the full URL into either field.
Model not found, or the key cannot use it
The ID is wrong, the model is not available to your key's group, or the model was retired.
- Copy the exact ID from openoxai.com/models. Display names and provider prefixes are not IDs.
- Case and punctuation matter:
claude-sonnet-5-5is notclaude-sonnet-5.5. - Check the model card's availability and your key group. A key created for one product line may not unlock every family.
- If a model was retired, delete it from Chatbox and add its replacement ID.
The OpenOxAI group is empty in the picker
The provider exists but has no model entries, or Chatbox has not refreshed the provider.
- Reopen Settings → Model Provider → OpenOxAI and confirm at least one model is listed.
- Click New and add an ID manually if Fetch failed.
- Confirm the green enabled dot is next to the provider name.
- Close and reopen the model picker, or restart Chatbox if the list is still stale.
Check fails, requests stall, or the web build cannot connect
Separate a credential problem from a transport problem before editing the provider.
- Check / Fetch fail: the key is the problem, not the network.
- Requests time out: enable Improve Network Compatibility, then retry. Disable it again if latency worsens.
- Browser version fails: use the desktop app. CORS can block custom OpenAI-compatible hosts in the web build.
- Large responses fail: lower Max Output Tokens for a test, then raise it gradually.
If a specific model fails while others work, the issue is model-side, not the provider. Try another ID from the table in section 06.
A 60-second diagnostic order
- 1. Preview URL exactly
https://api.openoxai.com/v1/chat/completions? - 2. Key copied without whitespace, active, funded?
- 3. Model ID copied from the live model page?
- 4. Provider green and at least one model listed?
- 5. New chat, model selected,
hisent, normal reply received?
Questions before you commit
Short answers to the things people ask after the first successful request.
Is Chatbox free to use with OpenOxAI?
The Chatbox desktop app is free to download. Using it with your own OpenOxAI key does not require a Chatbox AI subscription. You pay OpenOxAI for the tokens you consume; Chatbox's own hosted service is a separate, optional product.
Do I need a Chatbox account?
No. A custom provider works without signing in to Chatbox AI. The app stores the provider settings and chat history locally. Signing in to Chatbox AI is only needed if you want its hosted models, sync or paid features.
Can I reuse this Chat provider for GPT and Claude?
Use a separate provider and a matching service-scoped key for each family. GPT should use Responses API, Claude uses Anthropic Messages, and the illustrated Chat Completions provider only applies to models that support that protocol. Adding a model ID does not change the protocol.
Where is my API key stored?
In Chatbox's local provider configuration on your device. It is sent as a bearer token to
api.openoxai.com with each request. Treat the device profile as a credential: lock the
machine, avoid shared accounts, and revoke the key in the OpenOxAI dashboard if it is exposed.
Can I reuse the same key in CC Switch or Cherry Studio?
Yes, subject to the client's protocol. Chatbox uses the OpenAI-compatible host
https://api.openoxai.com/v1. CC Switch manages Claude Code and other tools, where
Anthropic Messages uses the root https://api.openoxai.com instead. The key can be the
same; the endpoint format is what changes.
Why does the model reply look different from the Chatbox AI models?
You are talking to the model you added, through OpenOxAI, not to Chatbox's hosted service. The model family, system prompt, temperature and context settings all affect the answer. Check the provider label under the message and the model nickname in the composer.
Can I send images or files?
Only if the selected model supports the input type. The model cards in the OpenOxAI library list capabilities; most of the recommended text models are text-only. If you enable Vision in the Chatbox model editor but the model cannot accept images, the request will fail at the provider.
Does the mobile app work with this setup?
Chatbox's mobile apps support custom providers, but the menus differ from the desktop flow in this
guide. The provider fields are the same: OpenAI API Compatible, API Host
https://api.openoxai.com/v1, API Path /chat/completions. The web build is
not recommended because of CORS.
Keep going
The pages you will need after the first successful reply.
Models & pricing
Exact IDs, context length and live reference prices.
CC Switch
Manage Claude Code, Codex and other coding-tool profiles from one desktop app.
Chatbox help center
Official documentation for Chatbox features, accounts and settings.
One guide. A matching key. A verified request.
If this saved you a configuration loop, post the guide with the preview card and the verified hi → reply proof. The card and the Open Graph metadata are already set up for X / Twitter, Discord and Slack unfurls.
Post this guide on XNeed a hand?
OpenOxAI support can check the key group, account balance and model access. Include the model ID, the timestamp of the request and the error text — never the key itself. Contact support →