Appearance
Settings API
Manage agent settings and configuration.
Endpoints
| Method | Endpoint | Description |
|---|---|---|
| GET | /api/agents/{id}/settings | Get settings |
| PUT | /api/agents/{id}/settings | Update settings (deep-merged) |
Get Settings
bash
GET /api/agents/{id}/settingsResponse
Settings are a nested object (AgentSettings in src/lib/types.ts). Secret-bearing fields (API keys, wallet keys) are redacted in the response.
json
{
"success": true,
"data": {
"llm": {
"provider": "workers-ai",
"model": "@cf/meta/llama-4-scout-17b-16e-instruct",
"temperature": 0.7,
"maxTokens": 2048,
"systemPrompt": "You are a helpful assistant."
},
"calendar": { "timezone": "UTC" },
"workflow": { "maxConcurrent": 5 },
"browser": {
"headless": true,
"viewportWidth": 1280,
"viewportHeight": 720,
"defaultTimeout": 30000,
"recording": false,
"hitlProvider": "cf"
},
"toolRouting": { "jev": true, "threshold": 0.6 },
"memory": { "provider": "local", "verify": true },
"knowledge": { "aiSearchInstance": "" }
}
}Update Settings
bash
PUT /api/agents/{id}/settings
Content-Type: application/jsonSend only the keys you want to change; nested objects are merged.
json
{
"llm": { "provider": "openai", "model": "gpt-4o", "temperature": 0.5 }
}Settings Categories
llm
| Setting | Type | Description |
|---|---|---|
provider | string | AI provider |
model | string | Model name |
temperature | number | Creativity (0-1) |
maxTokens | number | Max response length |
systemPrompt | string | System instructions |
visionModel | string | Model used for image inputs (auto-fallback chain when unset) |
toolRouting
Controls the Jev tool-category router. Also exposed in Prompt Studio → Tool Categories.
| Setting | Type | Description |
|---|---|---|
jev | boolean | Default true. false = regex triggers only, and no partner-model decisions anywhere for this agent |
threshold | number | Probability a category must reach to be loaded (default 0.6) |
memory
See Agent Memory → Memory providers. Exposed in Settings → Runtime.
| Setting | Type | Description |
|---|---|---|
provider | "local" | "cloudflare" | local (default): agent_state + Vectorize + FTS. cloudflare: the Agent Memory binding (private beta); falls back to local without the binding |
verify | boolean | Default true. Jev checks each new memory against similar existing ones and marks replaced ones superseded |
knowledge
| Setting | Type | Description |
|---|---|---|
aiSearchInstance | string | Name of a Cloudflare AI Search instance indexing the agents' R2 bucket; searched alongside the local index (Knowledge → Retrieval) |
browser
| Setting | Type | Description |
|---|---|---|
headless | boolean | Headless mode |
viewportWidth / viewportHeight | number | Viewport in pixels |
userAgent | string | Custom UA |
defaultTimeout | number | Milliseconds |
locale / timezone | string | Browser locale and timezone |
recording | boolean | Default false. Launch Cloudflare Browser Run sessions with recording on (30-day retention); Settings → Runtime |
hitlProvider | "steel" | "cf" | Backend for human-in-the-loop live-view sessions. Headless automation always uses the CF provider |
calendar, workflow, contacts, skills, payments, integrations, prompts
See the respective feature pages: Calendar, Workflows, Contacts, Payments.
