#Local LLMs with LM Studio

LM Studio can run a local model server that Sero treats like an OpenAI-compatible provider. Use this when you want to test local models, work with private local endpoints, or avoid sending a specific task to a hosted model provider.

Local models still need enough memory and GPU/CPU capacity on your machine. They may be slower or less capable than hosted models, and they are not automatically available inside every workspace container unless the URL is reachable from the process that needs it.

#Quick path

1. In LM Studio, download a chat/instruct model.
2. Load the model.
3. Start LM Studio's local server with OpenAI-compatible API enabled.
4. Open the model selector. Select its gear icon to open Model Manager.
5. Select Local, select Add Provider, and choose LM Studio.
6. Confirm the base URL is http://localhost:1234/v1.
7. Keep Authentication set to None unless your server requires a key.
8. Select Test, Fetch from server, and Add Provider. Then assign the profile model tiers.

#Set up LM Studio

In LM Studio:

1. Install or open LM Studio.
2. Download a model that supports chat.
3. Load the model into the runtime.
4. Open the local server panel.
5. Start the server.
6. Keep the server running while Sero uses it.

The Sero preset expects:

If your LM Studio server uses a different port, update the base URL before testing.

#Add the provider in Sero

1. Open Model Manager from the gear icon in the model selector.
2. Select Local.
3. Select Add Provider.
4. Choose LM Studio in Quick Setup.
5. Select Test.
6. Select Fetch from server to import model IDs from LM Studio's /models endpoint.
7. Select Add Provider.

Sero writes local provider configuration to <SERO_HOME>/agent/models.json and refreshes model availability after saving.

#Assign tiers

After you save the provider, open Admin and choose the LOW, MED, and HIGH defaults.

A practical local setup is:

Thinking levels only appear when Sero believes the selected model supports them. Many OpenAI-compatible local servers do not support reasoning-effort controls, so the LM Studio preset disables that compatibility flag.

If the model and server support thinking:

1. Edit the provider and select its thinking request format.
2. Expand the provider in Local and edit the model.
3. Turn on Thinking support.
4. Map each supported Sero thinking level to the provider value.

Sero hides levels that you set to Disabled from the ChatPanel. Sero does not infer thinking support from the model name.

#Host and container reachability

localhost means “this process's machine or network namespace.” That is usually fine for Sero desktop talking to LM Studio on your host machine. If a tool inside a workspace container must call the same local server directly, localhost from inside the container may point at the container, not the host.

If a containerized command cannot reach LM Studio:

• use a host-reachable URL instead of localhost when your setup supports it
• verify the LM Studio server binds to an address reachable from the container
• test from the same place that will make the request
• avoid exposing the server on untrusted networks

For most model selection and chat usage, configure the provider through Sero and let Sero manage model calls from the desktop process.

#Troubleshooting

#Related docs

• Models and Providers
• models.json Reference
• State and Folders