⫻ JevFlowOpen source ↗

Semantic if-statements
for your software.

Try a decision. Challenge the inputs. Find your threshold.
A hands-on workbench by Pasindu Suraweera.

Open the workbench ↗
Built by PasinduBring your own key7 runnable recipes
→→→

02 / Focused questions. Structured answers. No text to parse.

01 / YOUR WORKBENCH

Put the decision to the test.

Choose a recipe, change its context, and inspect the outcome.

Checking backend
REAL DECISIONS / YOUR API KEY
No key set

Your key passes through this server to TypeSafe. The app does not store or log it. Live runs send your context to TypeSafe and may incur charges. Use a trusted host or self-host.

Below this threshold: human review. Actions are previews only.
JevFlow original
THE QUESTION

Which team should handle this ticket?

Inspect the question sent to Jev
⌘ / Ctrl + Enter to run
 Decision outputAWAITING JEV
CHOICERun with your key
No result yet
Provider-reported confidence
↳
ACTION PREVIEW
Awaiting result
✓
View response JSON { }
Jev API · action preview
Every run calls Jev with your key. Context goes to TypeSafe and provider charges may apply. Changing the threshold uses the last response and makes no new API call.





THE COMMUNITY LAB
Good ideas deserve a test drive.
Original JevFlow adaptations of patterns from community projects.
3 runnable recipes




01 / ROUTING
Pick the right model.
Give the router a task. See whether it calls for speed, reasoning, coding, or a general model.
task→code→review?
Inspired by BunsDev / TypeSafeAI ↗Selection only. No downstream model call.


02 / BROWSER AGENTS
Find the next click.
Describe a page and its available actions. Ask which step gets closer to the goal.
goal→open_docs
Inspired by browser-use / Jev Ultrafast ↗Snapshot evaluation. No browser is controlled.


03 / GAME LOGIC
Give a character a choice.
Change hunger, energy, and money. See how a character weighs eating, resting, working, or exploring.
hunger: 85→eat
Inspired by jammaru / Hundred ↗One decision. The full town lives in the original project.


Built here from scratch. Original creators are credited for the ideas. These recipes do not embed their applications or imply endorsement.


02 / A SMALL LAYER. A BIG DIFFERENCE.
Your workflow. Your threshold.
Your call.
01
Bring the context
Pass a ticket, a document, an agent action, or your current application state.
state: { ...yourContext }
02
Ask a focused question
Define what you need to know and the answer space your application expects.
choice · score · yes / no
03
Keep control of the action
Route the result, set a threshold, or hand uncertain cases to a human.
if (decision) { act(); }


03 / MADE FOR YOUR CODEBASE
Feels like software.
Because it is.
Explicit inputs. Typed outputs.
A decision boundary you can read.
One question.
A useful answer.
JevFlow keeps the recipe, model result, and action policy separate. Inspect the question, tune a threshold, then take the pattern into your own application.
Read the official SDK docs ↗
Server-side example · TYPESAFE_API_KEY required


04 / SMALL DECISIONS, EVERYWHERE
Start with a useful decision.


05 / UNDER THE HOOD
A clear boundary.
From context to action.
Jev evaluates. Your application decides what happens next.
Your application Context + focused question
↓
YOUR SERVER / DECISION LAYER
Jev System OneTyped answer + probability
→
Policy checkThreshold + validation
↓
Execute actionAbove your threshold
Human reviewUncertain or restricted


OPEN BY DESIGN
Build on JevFlow.
Inspect the code. Change the rules. Build your own decision layer.
Download source ↓JevFlow on GitHub ↗
◇ MIT licensed
Built by Pasindu Suraweera. Independent of TypeSafe AI.
Jev supplies the predictions. JevFlow supplies the recipes and decision policy.