nodecode-modular
Modular Cloud: MiniMax, DeepSeek and Gemma models, each sent the reasoning effort its API takes. A Nodecode rewrite of tommyjtl/hermes-modular-provider by tommyjtl.
Modular Cloud serves open models over an
OpenAI-compatible API at https://api.modular.com/v1. With this cell,
/connect offers Modular, /models lists the four models it describes, and a
turn on modular/<model> goes out the way Modular takes it:
- the reasoning control is a nested
reasoningobject, not the chat lane's top-levelreasoning_effort:{"effort": LEVEL}to think at a depth,{"enabled": false}to stop thinking,{"enabled": true}to just think; - a level the model does not offer is lowered to the nearest one it does,
never raised: DeepSeek V4.1 Flash takes
low,highandmax, sominimalbecomeslowandxhighbecomeshigh; MiniMax-M3.1-Flash takes five levels; minimax-m3 and gemma-4-31b-it take a toggle and no level at all, so any pick there means "think"; - turning reasoning off sends
reasoning.enabled = falsewhere the model has that switch, and its lightest level otherwise (MiniMax-M3.1-Flash always reasons, so it getslow); - with no level picked, a model with a switch is told to think and a model without one is sent no reasoning field, so its own default stands -- the original's rule.
Get it
Tell the agent "add the modular cell", or run nodecode add nc://modular.
Set it up
Make a key at https://docs.modular.com/models, save it with /connect, or
set MODULAR_API_KEY. The key rides to Modular and nowhere else.
The section is on unless it says "enabled": false:
{
"modular": {
// a proxy, or another base speaking Modular's wire
"base_url": "https://api.modular.com/v1"
}
}
Use it
Pick Modular in /models, then a model, and pick a thinking level with
/think if you want one. A model the table does not describe still works if
you name it: it is dialled at the same base and sent no reasoning object.
Not ported
- The
GET /v1/modelsread that narrowed the live list to the described models. Nodecode reads this cell's catalog row instead, which carries the same four models; a model Modular adds is one line inprovider.lisp. - Prices, which the original left to Modular's public catalog.
- The profile's default auxiliary model (
MiniMaxAI/MiniMax-M3.1-Flash): Nodecode has no per-provider auxiliary model to set.
Licensed under the Apache License 2.0; see LICENSE and NOTICE.