Connect Claude Code and other model providers
Sign in to Claude Code or connect OpenAI, Grok, Gemini, DeepSeek, GLM, Kimi, MiniMax, Qwen, and Xiaomi MiMo in Onevium. Choose and verify chat and image models.
Get startedChoose a connection method#
Start by connecting one service you already have access to. You can add other services and switch models in a conversation later.
| What you have | Where to connect | What to do next |
|---|---|---|
| Claude Code login or subscription authorization | Sign in to Claude Code in a terminal first (claude login), then choose I already use Claude Code in first-run setup or in the connect prompt that appears when you send | Choose a model under Claude Code in the chat model menu |
| A provider API key or Coding Plan credential | Settings → AI services → Add AI service, then choose the matching preset | Select that provider and model in the chat composer |
| A team gateway's address, key, and model IDs | Add AI service → Custom Provider | Select its API format; see custom connections |
Signing in to Onevium, activating a Onevium license, and obtaining model access are separate steps. A Claude subscription login does not create an API key for the provider form.
Prepare the right account and credential#
Check which service, region, and product your key belongs to: a general API account or a Coding Plan. Kimi Coding Plan and the Moonshot API are separate connections. GLM, Moonshot, MiniMax, Qwen, and MiMo Token Plan also have regional presets. The preset form's Get API key and Official setup guide links can help you check.
Enter secrets only in API Key, not in chat or screenshots. Presets select the authentication method for you. Even for Bearer authentication, enter the raw credential without a Bearer prefix.
Review chat and image models#
Open Settings → AI services → Add AI service. The tables below show the main choices in the catalog for the applicable version. They describe app presets, not a live check of your account's access.
Chat services
These connections can be selected in the chat composer. A preset supplies the address and API format; normally you only need your API key.
| Service in the interface | Account or connection | Example preset models |
|---|---|---|
| Anthropic | Anthropic API key; separate from Claude Code sign-in | Opus 5.5, Opus 5, Fable 5.2, Mythos 5.2, Sonnet 5.5, Haiku 5.5; Haiku 4.5 and older entries are also available |
| OpenAI | OpenAI API key; Responses API | GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Astra, GPT-6-Cyber, GPT-6 Rosalind, GPT-6 Sol, GPT-6 Luna, plus GPT-5.6 Sol, Terra, Luna |
| Grok | xAI API key; Responses API | Grok 4.7, Grok 4.6 |
| Google Gemini | Google API key; Gemini API | Gemini 3.8 Flash, Gemini 3.5 Flash-Lite, Gemini 4 Argon |
| DeepSeek | DeepSeek API key | DeepSeek V4.1 Flash (1M), DeepSeek V4.1 Flash, DeepSeek V4 Pro (1M) |
| GLM (CN) / z.ai (Global) | GLM Code Plan for the matching region | GLM-5.3 (1M), GLM-5.3-Flash (1M) |
| Kimi Coding Plan | Kimi Code coding plan | Kimi K3 (256K), Kimi K3.1 Pro, Kimi K3.1 Pro Highspeed, Kimi K2.8 Preview, Kimi K2.7 Code Highspeed |
| Moonshot (CN / Global) | Moonshot API account for the matching region | Kimi K3 (1M), Kimi K3.1 Pro (1M), Kimi K3.1 Pro Highspeed (1M), Kimi K2.7 Code; older K2 entries are also available |
| MiniMax (CN / Global) | MiniMax Code Plan for the matching region | MiniMax-M3.1 Flash (1M), MiniMax-M3 (1M) |
| Qwen (CN / Global) | Alibaba Cloud Model Studio API key for the matching region (pay as you go) | Qwen3.8 Max (1M), Qwen3.8 Flash (1M), Qwen3.7 Plus (1M); open-weight Qwen3.8 2.4T-A95B and Qwen3.8 27B |
| Qwen Coding Plan (CN / Global) | Model Studio Coding Plan for the matching region | Qwen3.7 Plus (1M) |
| Xiaomi MiMo | Xiaomi MiMo API key (pay as you go) | MiMo V2.6 Pro (1M), MiMo V2.6 Flash (1M), MiMo V2.6 Pro UltraSpeed (1M) |
| MiMo Token Plan (CN / Global) | MiMo Token Plan for the matching region | MiMo V2.6 Pro (1M), MiMo V2.6 Flash (1M) |
The table shows the 1.3.3 catalog, synced on 2026-11-10; earlier versions list fewer entries. The Qwen and Xiaomi MiMo presets are available from 1.3.2. Kimi K3.1 Pro needs a Kimi Code Plus (Moderato) plan or higher, so the Kimi Coding Plan roles stay on K3. Retired MiniMax Preview and M2.7 models switch to M3.1 Flash after the upgrade.
OpenAI, Grok, and Google Gemini can use their native APIs in chat. The Anthropic, DeepSeek, GLM, Kimi, Moonshot, MiniMax, Qwen, and MiMo presets use Anthropic-compatible APIs. For a gateway, select its matching connection type.
Pay-as-you-go API keys and plan keys (Coding Plan / Token Plan) are not interchangeable. Qwen and MiMo each have separate presets for them, and a mismatched key usually returns 401 or 403. With the OpenAI preset, replies show GPT's reasoning summary. If your organization has not completed OpenAI verification, the app retries once without the summary and the conversation continues.
Entries whose name ends in (1M) run with a 1M context window in Claude Code. OpenAI, Grok, and Gemini connections do not get that window automatically; see 1M context window for the model entry to add.
These names help you find menu choices; they are not a list of IDs to paste. For example, DeepSeek V4.1 Flash uses deepseek-flash, the Kimi Coding Plan K3 main model uses k3-256k, and Moonshot K3 uses kimi-k3[1m]. Do not swap IDs between these connections. Saved custom settings may also differ from a new preset. Check your account's access with an actual request.
Image generation
To generate images, connect a service under Image generation in the same add-service window:
| Image service | Preset model choices | Credential |
|---|---|---|
| OpenAI (Image) | GPT Image 2.6 Flare, GPT Image 2.6 Sunburst | OpenAI API key |
| Google Gemini (Image) | Nano Banana 2.1, Nano Banana 2, Nano Banana 2 Lite, Nano Banana Pro | Google API key |
After saving, click Image settings beside that provider entry:
- Choose the Default image model.
- Set the resolution, aspect ratio, quality, and output format supported by that model. Requests that omit these options use these defaults; available controls vary by model.
- With several image connections, choose Use for image generation, then check the default marker in the list.
- Enable Settings → Tools → Image creation and send an actual request using the image generation tutorial.
Image services do not appear in the chat model picker. Selecting a GPT or Gemini chat connection does not configure image generation. After changing defaults, check for save errors and verify an actual generated result.
Example: connect the DeepSeek preset#
- Open Settings → AI services → Add AI service.
- Under Chat services, find DeepSeek and click Connect.
- Paste your DeepSeek account's API key. The preset supplies
https://api.deepseek.com/anthropicand Bearer authentication. - Review Visible models, keeping at least one visible. Leave advanced settings at their defaults for your first connection.
- Click Connect. Once you see the saved provider entry, return to chat.
- Find the DeepSeek group in the composer's model menu. Select DeepSeek V4.1 Flash or the corresponding option your account can use.
Follow the same process for OpenAI, Grok, Gemini, and other presets, using each service's own key. Saving an entry does not test the upstream connection or switch every conversation.
The retained screenshot below is from an earlier version. Field positions and model names may differ; follow the steps above and the current app menu.
Real 1.2.2 connection dialog showing the current DeepSeek preset models. The key is empty and no connection was made; verify a real request with your own account.
Manage models and default reasoning effort#
Check the provider and model in the composer before starting a task. Identically named models can belong to different connections. Use the menu selection, not the model's own description of itself in a reply.
In Settings → AI services, click Manage Models beside a saved connection to show or hide models, or add a real ID confirmed by the service. Keep at least one visible model per chat connection. See model entries and role mappings for the fields.
In the same window, Default reasoning effort lets you set a default per provider and model. Changes save immediately and supply the default for new conversations. Existing conversations retain their own choice; use the chat composer to adjust the current task. These settings appear only for models with a choice of supported effort levels.
Provider connections are shared within the app, while conversations retain their own selections. Channels and automations also have separate settings; changing the current chat does not change all of them.
Disconnect deletes the provider entry. For an ordinary model switch, use the composer. Before deleting a connection, move conversations, channels, and automations that still reference it to another working connection.
Check configuration, then verify a real request#
Click the stethoscope icon, Check configuration, beside a saved entry. It checks local addresses, credentials, and model records. It does not make an upstream request.
Select your intended model and send a small request:
Reply with one sentence confirming you received this message.
Do not read files or run commands.
After a successful text response, verify a tool in a test project containing README.md:
Read only README.md in the project root and summarize its run instructions.
If it does not exist, say so. Do not create or modify files or run commands.
Check for the actual file-reading tool call and its result. If you need image generation, subagents, channels, or automations, verify those entry points separately. A text response does not establish that all those capabilities work.
Troubleshoot a connection#
| Symptom | Check first |
|---|---|
| 401 or authentication failure | Does the key match the region and product? Did you add a Bearer prefix? Is a custom connection using the right authentication method? |
| Model not found or access denied | Check the provider, real model ID, and account permissions. Catalog entries do not grant access |
| OpenAI, Grok, or Gemini chat is missing | Check the app version and look under Add AI service → Chat services. Do not select an image connection |
| A custom connection still lists Claude models | Check its connection type and add real model entries. For Anthropic-compatible connections, also check role mappings |
| Timeout or connection failure | Check API address, network access, and Settings → Network proxy. A proxy address is not a model API address |
| Text works but tools fail | Inspect the tool and upstream error. Confirm that the API and model support the capability you need |
| The key cannot be decrypted after moving computers | Edit the provider, re-enter the original account's API key, and save |
| A channel or automation still uses the old model | Check the selection in that channel or automation's own settings |
New model controls
Fable 5.2 has five thinking-effort levels, from low to max, with medium as the default. Supported Claude routes offer one-turn Ultrathink and an independent collaboration control. GPT-6.1 Terra uses Responses for tool calls. Mythos 5.2, GPT-6-Cyber and GPT-6 Rosalind require provider approval and are disabled by default; a catalog entry does not grant account access. Existing model mappings and disabled states are preserved. New Moonshot connections use Kimi K3.1 Pro for fast and subagent roles.
Next steps#
Once connected, complete your first task. For a team gateway, see custom providers. To understand which actions a task can take, read permissions.