THE SOFT CAT WORKBENCH
Make a prompt, inspect an output or plan the next request. Start with the job. Leave with something you can use.
New here? Try the worked exampleA FIRST LAP / NO ACCOUNT NEEDED
A tiny extraction exercise. The source omits a due date. A useful answer keeps that gap visible.
Invoice INV-104 from Juniper Studio.
Amount due: GBP 129.50.
"due_date": nullAn example to check, not a recorded model response.Read the source, inspect the prompt and see the target result.
Open the same example as an editable draft. Nothing is saved or sent.
Find deliberate structural errors, then load the matching example.
These steps do not call a model. The schema checks the shape. You still need to check each value and quote against the source.
PICK THE JOB
Messages you can edit, compare and keep.
Build reusable prompts, inspect filled messages and export an OpenRouter request. Save versions locally and keep a backup.
TAKE AWAYFilled messages, an exportable request and optional saved versions.Open tool ↗Local editor · optional browser saves
Compare prompt words and whitespace, estimated token counts and signed input costs. Load saved Workbench versions and export the result.
TAKE AWAYA report of changed words, estimated tokens and input cost.Open tool ↗Local comparison · reads saved versions
Errors you can locate and fix.
A plan with visible limits and assumptions.
Budget instructions, history, documents, tools and the reply. Compare the same request against dated context and output limits.
TAKE AWAYA copyable plan with context and reply-limit warnings.Open tool ↗Local plan · dated model limits
Compare recorded model prices against an editable workload. Filter by model ID or provider, then export the costs and sources.
TAKE AWAYA CSV of workload estimates, sources and assumptions.Open tool ↗Local calculation · dated model rates
Estimate supplied text and repeated conversation history. Inspect parsed messages, distinguish missing replies and export the assumptions.
TAKE AWAYAn estimate that includes repeated conversation history.Open tool ↗Local text estimate · no save
A reasoned starting point for your task.
WHEN YOU WANT TO TRY A MODEL
Chat with models via OpenRouter using your own key. Compare responses and estimated costs.
OpenRouter request · your key · provider charges
Open Chat Playground Only sends when you press Send.HOW THE PIECES WORK
The static site, browser tools, checked GitHub deployments and cloud maintenance without the retired server.
↗What the Context Budget Planner counts, how it handles conflicting limits and what its result can establish.
↗Follow repeated conversation history, inspect parsed messages and distinguish missing output from a free reply.
↗