Turn unreliable model output into validated objects.
Loading providers…
⚠ That doesn't look like valid JSON yet — keep typing, or switch to a simple {"field": "type"} map.
Groups runs so you can compare two configurations later.
Ctrl + Enter to run
pipeline
confidence
Result
Waiting for the model
The provider is generating an answer. Wassitai will then extract, repair and validate it against your schema.
Nothing has run yet. When it does, this panel shows:
The validated object — your schema's fields, with the right types, ready to use.
The pipeline trace — which stages touched the answer, and what they changed.
A confidence score — lower when the answer had to be repaired or re-asked.
The raw answer — the model's own words, exactly as they arrived, so you can see what the pipeline had to work with.
New to this? The beginner's guide explains every field on this page in plain language.
The parsing pipeline never ran — this failed before the answer came back, which is why every stage is neutral.
The pipeline ran to the end — every one of its stages completed, and this failed after them.
How far the run got isn't known — the server answered, but not with anything this page can place on the pipeline.
No per-stage detail came back with this failure, which is why the stages are blank.
The field that could not be rendered:
The answer itself validated. Here is what came back:
Which fields failed
()
Fields that failed validation, with the value the model returned
Field
Problem
Model returned
Wassitai asked the model again time(s) before this answer fitted your schema — which is why confidence is below 100.
Diagnostics
Usage
prompt completion total
What the provider returned, before Wassitai extracted, repaired or validated anything. A [tool call: …] line marks an answer the model returned as a function call rather than as text — everything after it is the provider's own.