Use Xiaomi MiMo V2.6 with Koder CLI

MiMo Token Plan has known integration bugs and is temporarily unavailable. Its setup entry is gray, and calls are blocked. Existing configuration is kept. Token Plan configuration below is historical reference; do not follow it to enable this entry. MiMo pay-as-you-go is separate and keeps its existing status.

The two products use different keys and endpoints.

The Token Plan console is authoritative for your cluster. Singapore uses token-plan-sgp.xiaomimimo.com; Europe uses token-plan-ams.xiaomimimo.com. The setup screen lets you enter the console address. Explicit custom proxy URLs remain available through the custom-provider option.

Configure without putting your key in command history

Run koder setup, choose either MiMo · pay as you go or MiMo · Token Plan, choose mimo-v2.6-pro or mimo-v2.6-flash, and paste the key at the masked prompt. In the Chinese UI the preset names are 小米 MiMo · 按量付费 and 小米 MiMo · Token Plan 套餐. Token Plan also asks for the console Base URL.

Alternatively, run one of these in a normal interactive terminal. The command asks for the key without displaying it:

koder model set --preset mimo-token-plan --model mimo-v2.6-pro
koder model test
koder ask "Reply with the single word ready."

For Flash, change the model ID to mimo-v2.6-flash. For pay as you go, change the preset to mimo. An existing environment-variable credential can be selected with --key-env MIMO_API_KEY; do not put the actual key in that option.

The API request uses /chat/completions, the exact selected model ID and max_completion_tokens. Tool-call IDs and reasoning history are retained through subsequent calls. Private reasoning is not printed as the answer. The preset uses a 1,048,576-token context setting and an 8,192-token output ceiling; explicit smaller endpoint/account limits take precedence. Model tests and tasks use your provider quota.

Use the built-in engine

MiMo connects through Koder directly. The internal backup integration is disabled and gray by default. It never activates automatically after a connection failure. Keep the built-in Koder engine selected while checking the exact Token Plan region, endpoint, model ID and key class. A bare official origin is expanded to the selected protocol path; a path copied from a different protocol is rejected locally.

Check a failure

Run koder engine list, koder model show and koder model test. The model test checks the saved Koder API configuration; a Hermes task additionally checks the actual external engine route. A 401 can indicate the wrong key/endpoint class; 403 can indicate model or region access; 404 can indicate a URL or model-ID error; 429 can indicate rate limits or exhausted quota. Share only redacted output.

The implementation and local protocol checks do not establish access on your Xiaomi account. Complete a live koder model test with your own configured key before relying on a task. No membership, learning, sync or computer permission is granted by choosing a model.

Sources: Xiaomi model IDs and capabilities, Token Plan endpoints and credentials, Xiaomi's Hermes configuration guide.