nc://xiaomi · Nodecode

nodecode-xiaomi

A Nodecode cell for Xiaomi MiMo, Xiaomi's model platform. With it, /connect offers Xiaomi MiMo, /models lists the MiMo models, and a turn on xiaomi/<model> goes over the OpenAI chat wire to wherever your key is served: https://api.xiaomimimo.com/v1 for a pay-as-you-go key, the Token Plan's regional cluster for a Token Plan key. A reasoning model is asked to think the way MiMo takes it.

Ported from oh-my-pi's xiaomi provider. See NOTICE.

What omp's login does, and how this cell carries it

omp's login for this provider is a custom one, but it signs nothing in. It opens the console's key page, asks for a key, and checks it with a one-token chat request to mimo-v2.5. It takes two kinds of key:

a pay-as-you-go key (sk-...), checked at https://api.xiaomimimo.com/v1
a Token Plan key (tp-...), which only the plan's regional clusters serve. omp tries Singapore, then Amsterdam, then China, and its model discovery does the same, so the models it lists carry the cluster that took the key and a turn goes there.

Here the key is a plain key, and the cluster rule is the cell's:

Key. Save it with /connect, or set XIAOMI_API_KEY. A saved key wins over the variable.
A tp- key's cluster. The section's token_plan_region. At "auto" (the default) the cell does what omp does: the first of token-plan-sgp, token-plan-ams and token-plan-cn whose /models takes the key serves it, for as long as the process runs. A listing finds it as it lists; a round that comes first asks the clusters itself. "sgp", "ams" or "cn" pins one and asks nothing. If no cluster takes the key, the round goes to Singapore and shows its refusal.
A sk- key goes to base_url, https://api.xiaomimimo.com/v1 unless the section says otherwise. A tp- key never does, as in omp.

Where to make a key: <https://platform.xiaomimimo.com/#/console/api-keys>; a Token Plan key is on the plan page, <https://platform.xiaomimimo.com/console/plan-manage>.

Thinking

omp's rule for the MiMo family is the zai dialect. For a reasoning model, an effort sends thinking: {"type": "enabled"} and off sends thinking: {"type": "disabled"}; reasoning_effort is never sent, since MiMo does not take it. With no effort set, nothing is sent, so the platform's default applies, as the chat lane does for every provider.

Install

Copy this folder into ~/.nodecode/cells/nodecode-xiaomi/ and restart Nodecode, or run (restart-cells).

Configure

Optional. The section is on unless it says "enabled": false:

{
  "xiaomi": {
    // where a pay-as-you-go (sk-) key is served
    "base_url": "https://api.xiaomimimo.com/v1",
    // where a Token Plan (tp-) key is served: auto, sgp, ams or cn
    "token_plan_region": "auto"
  }
}

Not ported

The login's key check against mimo-v2.5. /connect checks a key against the base's /models listing instead, which for a tp- key is the cluster the listing hook finds.
omp's replay rules for MiMo (reasoning_content on every assistant turn, no synthetic placeholder). The chat lane already replays the reasoning_content a model streamed, on every assistant turn once one carries it.

Test

From this repository's root:

tools/test-cell.sh nodecode-xiaomi

Every round is a stubbed dex:post, every cluster probe a stubbed dex:get, every listing a stubbed nle::http-fetch: nothing reaches Xiaomi.

MIT licensed.