Is this an official Camunda or TypeSafe product?

No. The README labels it a community connector. It packages Jev calls as connector operations with retries, trace fields, and secret-based auth, but support and versioning come from the repo maintainer.

What happens with catalogs larger than 255 options?

Choice questions accept up to 255 options each, so the connector runs an automatic tournament across groups with token-aware batching and parallel calls. Larger catalogs simply take more passes.

How do I route uncertain decisions to a human?

Each response carries confidence plus a needsReview flag for low-confidence outcomes, alongside process key, element id, model version, token usage, duration, and request id for audit.

Does it run on Camunda 8.8, or is 8.9 Self-Managed required?

Camunda 8.9 Self-Managed or a compatible connectors runtime, plus Java 21 and Maven, since mvn clean package produces jev-ai-decision-connector-1.2.0.jar for the connectors runtime and a standalone exec jar for local testing. On 8.9 only SECRET_-prefixed variables become connector secrets, so the key goes in connector-secrets.txt next to docker-compose.yaml.

What three operations does the connector offer?

Choice: pick the best option, returning jev.selection.id and jev.selection.top; Noul: find all matching options, returning jev.filter.matches; and Custom questions, returning your own noul, choice and score answers under jev.answers. .

When should I not point a process at Jev?

When the answer has to be written. Jev returns decisions, not reasons, so emails, summaries, reports, code generation, chat and open-ended questions are out of scope; a common pattern is Jev routing first, then an LLM drafting the reply.

What does it cost at volume?

TypeSafe billed $0.042 per million input tokens with output free as of the README. Routing a million tickets of about 500 tokens each is about $21, against roughly $580 for an LLM at $1 per million input plus 20 output tokens at $4 per million.

CommunityJev Tools & CLIs

Camunda Jev AI Decision Connector

Community Camunda 8 connector that calls Jev from BPMN to pick the best option, find every match, or ask custom noul, choice, and score questions gateways can route on.

See highlights

Open github.com

Stars: 1
Language: Java
License: APACHE-2.0
Published by: Community
Author: rajeshponna
Last verified: Oct 2026

rajeshponna/camunda-jev-ai-decision-connector

User repository on GitHub

View on GitHub

Camunda 8 connector for Jev by TypeSafe AI: fast, typed AI decisions in BPMN. Pick the best option, find all matches, or ask noul/choice/score questions over lists of any size.

Forks: 1
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Make fast, typed AI decisions inside your BPMN processes with Jev by TypeSafe AI. Route requests, pick the best option from a list, find every matching item, or score and classify anything, in milliseconds and for a fraction of what a chat LLM costs. The result is structured data your gateways and ad-hoc sub-processes can use directly, with no prompt parsing and no JSON repair. What is Jev and TypeSafe AI. Why it costs so much less. When to use it, and when not.

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Scoring

Used inJev for Software & DevTools

Languages: Java

Highlights

Offers three operations: pick the best option, find all matches, or ask custom typed questions in one flow.
Runs automatic tournaments across groups so lists of hundreds or thousands of options still resolve.
Flags low-confidence outcomes with needsReview and returns duration, token usage, and request id.
Keeps the API key in Camunda secrets and builds to plain and standalone runtime jars via Maven.

Watch out

Community connector, not an official Camunda or TypeSafe product; needs Camunda 8.9, Java 21, and Maven to build. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

brier

Open local decision model on Qwen3-4B with LoRA adapters that answers choice, score, and noul questions with probabilities, calibrated confidence, and conformal sets, offline with no per-call cost.