Declare and switch local model servers

What it is

A local model server runs on your own machine, for example llama-server, Strata or PolyStrata. Loomcoding declares each one once, in a global registry, and every project can use it.

What it is for

Using a model that runs on your machine, without a cloud key.
Switching between a cloud provider and a local server during a session.
Adding another server that speaks the OpenAI-compatible API.

How

Start the server on the port that its entry declares.
Start Loomcoding. It checks each server once, in the background.
Type /provider to choose a server, then /model <id> to choose its model.

Example

# <profile>/.loomcoding/providers.toml
default = "llama"

[[provider]]
id = "llama"
name = "llama-server"
kind = "openai_compat"
base_url = "http://127.0.0.1:8081/v1"
model = "your-model-name"

Then, in Loomcoding:

> /provider llama
> /model your-model-name

Where the registry lives

<profile> is your home folder, or the folder in LOOMCODING_HOME when that variable is set. Loomcoding creates providers.toml on the first start, with the three servers below. It never overwrites a file that already exists.

id	Server	Default port	Key
`llama`	llama-server (llama.cpp)	8081	none
`strata`	Strata	8080	variable `STRATA_API_KEY`
`polystrata`	PolyStrata	8082	none

The key is never written in the file. api_key_env holds the name of the variable that contains it. A server on the loopback address (localhost or 127.0.0.1) can leave api_key_env out.

Settings and defaults

Name	Default	Requires restart
`default` in `providers.toml`	`"llama"`	Yes
`[[provider]] base_url`	`http://127.0.0.1:8081/v1` for `llama`	Yes
`[[provider]] api_key_env`	not set (allowed only on loopback)	Yes
`[[provider]] context_window`	not set	Yes

Command	What it does	Requires restart
`/provider [id]`	Switches provider and model. Without an id, opens the picker with the status of each server.	No
`/model [id]`	Changes the model of the current provider. Accepts any id, even one that is not listed.	No
`/settings`	Opens the full form: provider, URL, key variable, model and effort.	No

/provider and /model save the choice in the loomcoding.toml of the project.

Server status

Status	Meaning
`up (N)`	The server answered with N models.
`down`	The connection was refused, the wait ran out, or the answer is not a model list.
`sin acceso (clave)`	The server answered 401 or 403: the key is missing or wrong.

A server that is down keeps its fallback model, so you can select it before you start it.

Result

When it works, /provider lists each server with its status. The next prompt goes to the server and model you chose.

If it does not work

Symptom	Fix
`/provider` says "Providers are still loading".	The first check has not finished. Wait a moment.
A running server shows `down`.	Check that the port and the `/v1` path in `base_url` match the server.
A server shows `sin acceso (clave)`.	Define the variable named in `api_key_env` before you start Loomcoding.
A project still uses the old provider.	Its `loomcoding.toml` wins. Change it with `/provider`.

To declare reasoning effort levels for a local model, see the technical note.

Example​

Where the registry lives​

Settings and defaults​

Server status​

Result​

If it does not work​

Next steps​