nodecode-aistupidlevel

AI Stupid Level: the Smart Router, where each request goes to the model measuring best right now, as a Nodecode provider. A Nodecode rewrite of StudioPlatforms/aistupidlevel-sdk by StudioPlatforms.

With it, /connect offers AI Stupid Level, /models lists its routing modes, and a turn on aistupidlevel/auto-tooling goes to https://aistupidlevel.info/v1 over the chat lane. A model id here is not one model but a routing rule:

auto-tooling sends each request to the model scoring best on the tool-use benchmark (the mode to pick for an agent)
auto-coding, auto-reasoning, auto-best and auto-value route on the coding, multi-turn reasoning, combined and score-per-dollar benchmarks
auto-task picks one of those from the request itself
an id such as claude-opus-5-5 pins that one model

The router calls the model with the provider keys saved on the account (OpenAI, Anthropic, Google, DeepSeek, Kimi, GLM), so billing stays on those accounts. It takes text and tool calls only and refuses image input.

The live catalog is the router's own GET /v1/models, which Nodecode's listing reads with the key once it is saved: the pinned models and any mode added later appear in /models without a change here. The seven rows above (the original's fallback list and the fast mode it names for titles and compression) stand in until that listing answers.

See NOTICE.

Get it

Tell the agent "add the aistupidlevel cell", or run nodecode add nc://aistupidlevel.

Set it up

Make a router key at https://aistupidlevel.info/router/keys (it starts aism_), and add the provider keys the router should call with in the router settings on the same site. Save the router key with /connect, or set ASL_ROUTER_KEY; a saved key wins over the variable.

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

{
  "aistupidlevel": {
    // where the router is served; the chat lane appends /chat/completions
    "base_url": "https://aistupidlevel.info/v1"
  }
}

The original's profile also names an ASL_BASE_URL variable for the same address; here that setting is the section's base_url.

Use it

Pick a routing mode in /models and talk to it as usual. auto-tooling is the original's own recommendation for an agent loop: the router keeps a tool-calling conversation on one model, so a loop does not switch models mid-task.

MIT licensed.