Settings

Per-agent behaviour — model choice, skip rules, escalation, and human takeover.

Settings

Settings are where an agent’s behaviour is configured: which model it runs, when it stays quiet, and when it hands off to a human. This is the resource you tune most often once an agent is live.

Model

Field	Type	Notes
model_provider	string	Which provider to call
model_name	string	The model
model_type	string	Model category
temperature	float	Sampling temperature
max_tokens	integer	Response ceiling

Lower temperature for agents that quote policy or amounts, where you want the same answer every time. Raise it only for conversational tone.

When the agent stays quiet

Skip rules are the cheapest safety mechanism available — the agent never engages rather than engaging badly.

Field	Type	Effect
skip_emails	boolean	Skip email channel entirely
skip_channels	array	Channels to skip
skip_tags	array	Skip tickets carrying these tags
skip_subjects	array	Skip on subject match
skip_instagram_messages	boolean	Skip Instagram DMs
allowed_intents	array	Allow-list — only act on these intents
time_threshold	integer	Age threshold before the agent declines
out_of_office_keywords	array	Treat as out-of-office and skip

allowed_intents is an allow-list, not a blocklist. Leaving it narrow is the safest posture for a new agent — it acts only on intents you have explicitly approved.

Escalation and human takeover

Field	Type	Effect
agent_take_over_tag	boolean	Whether a tag triggers takeover
agent_takeover_phrases	array	Phrases that hand control to a human
health_concern_keywords	array	Health-related language that forces escalation
escalation_team_id	integer	Team that receives escalations
escalation_tag_name	string	Tag applied on escalation
assignee_user_id	integer	Default assignee
agent_emails	array	Addresses the agent sends from

health_concern_keywords exists so an agent never improvises on a medical or safety matter. Keep it populated — an unset list means those conversations are handled like any other.

Endpoints

Method	Path
GET	/v1/settings	List
GET	/v1/settings/:id	Get
POST	/v1/settings	Create
PUT	/v1/settings/:id	Update
DELETE	/v1/settings/:id	Delete

Example

curl https://api.stateset.com/v1/settings \
  -H "Authorization: Bearer $STATESET_API_KEY"

Tighten an agent while you build confidence in it:

curl -X PUT https://api.stateset.com/v1/settings/42 \
  -H "Authorization: Bearer $STATESET_API_KEY" \
  -H 'content-type: application/json' \
  -d '{
    "temperature": 0.2,
    "allowed_intents": ["order_status", "shipping_address_change"],
    "agent_takeover_phrases": ["speak to a human", "this is urgent"],
    "escalation_tag_name": "needs-human"
  }'

The same controls elsewhere

These settings are the per-agent view. The equivalent controls at the workflow layer are in Workflow Studio — skip rules, escalation patterns, and the review gate — and in the Temporal engine’s AutomationConfig.

Related

• Rules — named policy constraints
• Response Public API — configure settings programmatically