Operator drafts
Suggested replies for the person who just stepped in.
The drafts endpoint asks your configured model for a few short reply options, built from the recent conversation and the user's current state. Operators can steer them with a line of guidance, then use, edit or ignore them.
Takeover · operator drafts
- #1
I’m really glad you told me. You don’t have to sort this out alone tonight.
Use - #2
That sounds exhausting. Would it help to talk through what happened today?
Use - #3
I’m here with you. Is there someone nearby you’d feel OK reaching out to?
Use
POST /api/sdk/takeover/user_42/drafts · count: 3
A starting point that already reads the room
Someone joining a hard conversation needs to catch up fast. Drafts give them options that already reflect how the user is doing. Nothing is sent until the operator chooses to send it.
- One to six options per request, three by default
- Optional guidance, such as “keep it brief” or “ask about their week”
- Works before a takeover starts, so operators can decide whether to step in
- If the model call fails, the endpoint returns an empty list instead of an error
Conversation · user_88
I can’t do this anymore.
Hi, I’m Priya, a real person on the team. I’m here with you. Can you tell me where you are right now?
AI paused · user history and state open for the operator
Ask for drafts
drafts = requests.post(
API + "/sdk/takeover/user_88/drafts",
headers={"X-API-Key": os.environ["HUMANE_API_KEY"]},
json={"count": 3, "guidance": "keep it brief and warm"},
).json()["drafts"] # ["I'm really glad you told me...", ...]Guardrails on the co-pilot
- Its own stricter rate limit: 6 requests per minute per key by default
- Uses your tenant's own configured model and key
- Drafts aren't stored as conversation messages
- Operators send through the messages endpoint, so every sent turn is recorded
Works well with
Human takeover
When the signals say someone needs a person, an operator opens a takeover for that user. The AI pauses, the operator's messages reach the user in the same conversation, and the AI resumes when the operator releases it or goes idle.
ExploreTone and context
Before your model runs, Humane assembles its prompt from the user's state, relationship, recent conversation, relevant memories and open tasks, scoring and trimming each section to fit a budget. It also sets the tone, empathy, formality and length the reply should take.
ExploreModel adapters
Humane talks to models through four adapters: OpenAI-compatible, Anthropic, Amazon Bedrock and Replicate. Change providers without touching the behavioural layer.
ExploreEvery AI should know who it's talking to.
The safety and human-oversight layer for emotional AI: user state across sessions, risk alerts, safety gates and human takeover, with every decision explained.
