Skip to main content

Inception Mercury Decide

Mercury Decide evaluates named questions against shared state and returns typed, calibrated answers (noul, choice, score) in one call. OpenTryOn exposes it on the decide service with MercuryDecideAdapter.

Unlike Mercury 3 chat, Decide uses POST /v1/decisions with model mercury-decide — no streaming, no tool calling. Only input tokens are billed; output tokens appear in usage but are free.

Registry idmercury-decide
AdapterMercuryDecideAdapter
AuthINCEPTION_API_KEY (same key as Mercury 3)
Endpointhttps://api.inceptionlabs.ai/v1/decisions

Question schema​

Same noul / choice / score shape as other decide models (see Typesafe Jev). OpenTryOn validates questions with load_questions — each question needs a non-empty instructions field even though the upstream API allows null.

CLI​

export INCEPTION_API_KEY="..."

opentryon decide --model mercury-decide \
  --questions questions.json \
  --state "Customer message: duplicate charge on my card." \
  --dry-run

opentryon decide --model mercury-decide \
  --questions questions.json \
  --state "Screenshot of checkout after Pay." \
  --image checkout.png

--image paths are converted to base64 data URLs (PNG/JPEG/WebP/GIF). Remote http(s) image URLs are not supported by Mercury Decide — use local files.

Python​

from tryon.api.inception import MercuryDecideAdapter

adapter = MercuryDecideAdapter()
out = adapter.decide(
    state={"message": "Please refund the duplicate charge."},
    questions={
        "billing": {"type": "noul", "instructions": "Is this about billing?"},
        "team": {
            "type": "choice",
            "instructions": "Which team owns this?",
            "criteria": {"billing": "payments", "support": "other"},
        },
    },
)
print(out["answers"])

MCP​

Tool: decide_mercury_decide. Restart MCP after upgrading OpenTryOn.

See also​