Blog

Mika API Key Setup: Add TeamoRouter as a Custom Provider and Pick the Main, Vision and Light Models

Short answer

Mika doesn't come with a model; you add your own provider. To use TeamoRouter:

  1. Create an API key in the TeamoRouter console (it starts with sk-teamo-) and top up your balance.
  2. In Mika, open Settings › Models (模型), tap Add provider and pick Custom (自定义).
  3. Interface type: OpenAI-compatible. URL: https://api.teamorouter.cn/v1. Paste your key.
  4. Tap Test and save (测试并保存). If the model you want isn't in the list, tap Enter a model name yourself and type its ID.
  5. Pick the roles:
    • Main model (must support tool calling): deepseek-flash-1017, or kimi-k3.1-pro for harder multi-step tasks.
    • Vision model (reads screenshots): glm-5.3-flash. Open the model and turn on Can see images.
    • Light model (small, cheap jobs): glm-5.3-flash.

The in-app steps follow Mika's model guide as of November 2, 2026. We sent the same kinds of requests to the TeamoRouter gateway, but we did not run the Mika app ourselves (see "What we tested" below).

New accounts start with a $0 balance — top up before your first request. Top up from $5. "Test and save" already makes a small model call, so it needs a balance too. If a request returns 402 (400 on the OpenAI format) with "wallet balance is insufficient", your balance is empty — your URL and key are fine. Top up and retry.

What Mika is

Mika (wangzan101/Mika, Apache-2.0) is an on-device AI agent for Android: you give it a task and it operates apps and the web on your phone, reading the screen and tapping for you. The repository was created on October 29, 2026. There is no Mika server: your messages and screenshots go only to the model provider you set up, and API keys are stored encrypted on the phone.

Mika splits the work across roles:

Role What it does Requirement
Main model Thinks and acts: plans the task and calls Mika's tools Must support tool calling
Vision model Describes the UI and finds buttons in screenshots Must accept images; off by default, or "use main model"
Light model Small jobs: judging whether a task worked, extracting memories, understanding intent Anything cheap and fast; defaults to the main model

A fourth role, speech to text, uses the phone's built-in recognition by default.

Add TeamoRouter step by step

  1. Open Settings › Models. On first launch Mika walks you through this.
  2. Tap Add provider and choose Custom at the end of the preset list. Custom accepts four interface types: OpenAI-compatible, Anthropic, Gemini and OpenAI Responses.
  3. Choose OpenAI-compatible, set the URL to https://api.teamorouter.cn/v1 and paste your sk-teamo- key. Mika stores the key only on the phone, encrypted.
  4. Tap Test and save. Mika calls the provider's model-list endpoint once, then gives the model that would become your main model a test tool and asks it to call it (a few dozen tokens). The provider is saved only if both work; otherwise Mika shows the server's error message.
  5. If a model isn't in the list or the test picks the wrong one, tap Enter a model name yourself, type the ID (for example deepseek-flash-1017) and tap Test and save again. Only that model gets the test call.
  6. Under Roles, set the main, vision and light models. For the vision model, tap it on the provider's page and turn on Can see images: for a custom provider Mika assumes a model can't see images unless you say so.

Which models to pick

Role Model ID on TeamoRouter Why
Main deepseek-flash-1017 DeepSeek V4.1 Flash: tool calling, 1M-token context, low price. DeepSeek's official price for V4.1 Flash is $0.30 / $1.20 per million input / output tokens at peak and half that off-peak
Main, harder tasks kimi-k3.1-pro Moonshot AI's Kimi K3.1 Pro, for long multi-step tasks; costs more per token than DeepSeek V4.1 Flash
Vision glm-5.3-flash Z.ai lists video, image, text and file input and a 1M-token context
Light glm-5.3-flash Fast and inexpensive; fine for yes/no checks and short summaries

Official prices are from DeepSeek's and OpenAI's pricing pages on November 2, 2026. TeamoRouter's current prices are on the pricing page. The same key works for every model in the table, so you can mix them freely.

Screenshots cost tokens. Each screenshot sent to the vision model is billed as image input on top of the text. A cheaper vision model keeps a long task affordable. If your main model can see images, turn on Can see images for it or pick it as the vision model: Mika then sends screenshots straight to the main model instead of making a separate vision call.

What we tested

On November 2, 2026 we sent the requests Mika makes to the TeamoRouter gateway with a placeholder key:

  • the model list, GET /v1/models;
  • a chat completion with a tool to deepseek-flash-1017;
  • a chat completion with an image to glm-5.3-flash.

All returned 401 for the placeholder key, which shows the URL and request formats reach the gateway. We did not run a paid end-to-end call or the Mika app itself; the in-app steps are based on Mika's documentation.

Paying for usage

No subscription: you top up a balance from $5 and pay only for the tokens you use, by card.

FAQ

Which interface type do I choose for TeamoRouter?

OpenAI-compatible, with the URL https://api.teamorouter.cn/v1. That covers DeepSeek, Kimi, GLM and the other models in the table.

"Test and save" fails. What now?

Read the error Mika shows; it is the server's own message. 401 means the key is wrong or incomplete; 402 (or 400 with "wallet balance is insufficient") means the balance is empty; "model is not available" means the model ID is misspelled.

Does Mika send my screen to TeamoRouter?

Mika has no server of its own. Your messages and screenshots go to the provider you configure, here TeamoRouter, which passes them to the model you picked.

Can I use one model for all three roles?

Yes, if it supports tool calling and can see images; set it as the main model and choose "use main model" for the other roles. Splitting roles usually costs less, because the light model handles the many small checks.

Related

Ready to connect?Log in · top up · create an API key — three steps to start.