Choice
Which option? You get the pick and a probability for every key you listed.
Depaza Edge · model d1
Decisions your code can trust.
Typed choices, scores, and gates — with probabilities. Schema in. Schema out. Confidence you can threshold. Chat is a product. This is infrastructure.
Score
Where on this rubric? A weighted index plus a distribution over levels.
Gate
Is this true? A number from 0 to 1 your software can act, review, or block on.
Playground
Run a decision
Pick a workflow. Edge returns typed answers. Your code applies the threshold.
Questions
Run Edge to get probabilities your code can branch on.
API
One endpoint
POST a state and named questions. Get typed answers. Code keeps control flow.
curl https://depaza.eu/v1/decide \
-H "Authorization: Bearer dpz_live_…" \
-H "Content-Type: application/json" \
-d '{
"state": {"ticket": "Payouts failed for 3 days. Refund the fee. ASAP."},
"questions": {
"team": {
"type": "choice",
"instructions": "Which team should own this?",
"criteria": {"billing": "Charges, payouts, refunds", "tech": "API, bugs", "sales": "Pricing"}
},
"urgent": {"type": "gate", "instructions": "Needs a reply today."}
}
}'What it will not do
Write the customer reply. Generate code. Run an agent loop. Invent a key that was not in your schema.
How you use it
If confidence is high, act. If it is not, ask a human. The threshold lives in your code — never in a paragraph.