Configure Llm

Configure LLM

Manage the model that powers Jarvis via the settings protocol (§5.15). The server stores keys securely and hot-swaps the active provider; keys are never echoed back (key_set is a boolean only).

Steps

1. Read current config + catalog: request client.settings_get{section:"llm"}

→ server.settings.llm (current + providers).

1. Render a settings_panel (show_settings) with a provider select, a

model select, and a key entry/button — reflecting key_set and each provider's needs_key / needs_base_url / capabilities.

1. Apply changes with client.settings_update — send only what changed; omit

api_key to keep the existing key, include it to set/replace.

1. Confirm from the returned server.settings (don't read keys aloud).

Output

settings_panel (show_settings, props per registry.json):

{ "widget_type": "settings_panel",
  "props": { "title": "Model",
             "sections": [ { "title": "LLM",
               "settings": [
                 { "id": "provider", "label": "Provider", "type": "select",
                   "value": "openai", "options": ["openai","anthropic","ollama"] },
                 { "id": "model", "label": "Model", "type": "select",
                   "value": "gpt-4o", "options": ["gpt-4o","gpt-4o-mini"] },
                 { "id": "api_key", "label": "API key", "type": "button", "value": "set" } ] } ] },
  "interactions": ["tap","grab","resize","toggle","slider"] }

client.settings_update (client → server, §5.15):

{ "llm": { "provider": "ollama", "model": "llama3.1", "base_url": "http://localhost:11434" } }

agent.speech: { "text": "Switched to Llama 3.1 running locally.", "final": true }

Edge cases

• Missing key for a key-required provider → prompt for it; report

invalid_key from server.error clearly.

• Provider unavailable / offline → provider_unavailable; offer the mock or a

local provider as fallback.

• Capability mismatch (e.g. no vision) → warn if the user relies on perception

with a non-vision model.

• Never reveal keys — only key_set true/false; redact in speech and UI.
• Local provider needs base_url → require it when needs_base_url is true.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: sumitaich1998
• Source: sumitaich1998/jarvisvr
• License: MIT

Install and usage instructions live in the source repository linked above.