LLM Access (`llm`)

Requires: LLM

Sends text or multi-turn dialog requests through the user's configured LLM profiles (Ollama, Gemini, Groq, Cerebras, Puter, and other providers). API keys stay in the main process and are never exposed to addons.

Setup

Add "LLM" to permissions in manifest.json (the user approves it at install time).
Check at runtime with permissions.has(AddonsPermission.LLM).
Optionally call llm.getSettings() / llm.listProfiles() before llm.chat.

{
  "id": "my-addon",
  "permissions": ["LLM"]
}

How it works

Uses the same profiles and routing as Settings → LLM Access.
defaultProfile may be auto (host picks among enabled free-capable profiles) or a specific profile id.
options.profile is honored only when the user enabled Allow addons to choose a profile.
freeOnly (global or per-request) skips paid profiles / paid-looking models.
On rate limits (HTTP 429 / Retry-After), the host cools down that profile and may fall back to another.

`llm.listProfiles()`

Returns public profile metadata (no API keys), including soft quota snapshots when tracked.

const { success, profiles, message } = await llm.listProfiles();
// profiles[].id, name, provider, model, params, enabled, isPaid,
// hasApiKey, contextWindow?, quota?

`llm.getSettings()`

const {
  success,
  enabled,
  defaultProfile,
  allowAddonProfileSelect,
  freeOnly,
  message,
} = await llm.getSettings();

Field	Description
`enabled`	Master switch; when `false`, `chat` fails
`defaultProfile`	`'auto'` or a profile id
`allowAddonProfileSelect`	Whether `options.profile` overrides are allowed
`freeOnly`	Prefer / restrict to free profiles

`llm.chat(input, options?)`

Parameter	Required	Description
`input`	yes	Plain string or `{ role, content }[]` (`system` / `user` / `assistant`)
`options.profile`	no	Profile id or `'auto'` (needs `allowAddonProfileSelect` when overriding default)
`options.temperature`	no	Override profile default
`options.topP`	no	Override
`options.maxTokens`	no	Override
`options.frequencyPenalty`	no	Override
`options.presencePenalty`	no	Override
`options.stop`	no	Stop sequences
`options.freeOnly`	no	Force free-only for this call
`options.timeoutMs`	no	Request timeout

await llm.chat('Summarize this donation message');

await llm.chat(
  [
    { role: 'system', content: 'Reply briefly.' },
    { role: 'user', content: 'Hello' },
  ],
  { profile: 'auto', temperature: 0.4, maxTokens: 256 }
);

Returns:

{
  success: boolean,
  content?: string,
  profileId?: string,
  provider?: string,
  model?: string,
  usage?: { promptTokens?, completionTokens?, totalTokens? },
  message?: string,
  triedProfiles?: string[],
}

Example

if (!permissions.has(AddonsPermission.LLM)) {
  console.warn('LLM not granted');
  return;
}

const { enabled } = await llm.getSettings();
if (!enabled) return;

const res = await llm.chat('Write a short thank-you for a donation');
if (res.success) {
  console.log(res.content);
}

LLM Access (llm)