JevStation

Run typed decision models with structured answers and probabilities.

0 views this week0 upvotes

About JevStation

JevStation is an advanced AI developer platform tailored for running typed decision models such as Jev and TypeSafe's System One model. Unlike typical language models that generate prose, JevStation specializes in delivering precise, structured decisions with calibrated confidence scores and full probability distributions. This makes it uniquely suited for applications where making accurate, interpretable decisions is critical, streamlining the integration into software by removing the need for text parsing.

Users interact with JevStation by submitting queries—such as support tickets, product reviews, or specifications—and posing up to 12 typed questions per evaluation. The platform handles credit management transparently and allows reuse of context with pinned references, making repeated or contextual evaluations efficient. With a focus on privacy, JevStation does not use user data for training, providing a secure environment with clear cost controls and automatic credit refunds on failures.

JevStation supports three question types: Choice (categorical), Score (numeric), and Noul (null/unknown), returning typed answers that can be seamlessly used in decision-making code. It also offers an open submission system to share question sets or tools with the community, fostering collaboration.

Use Cases

Support ticket triage: Paste batches of customer support tickets and ask Noul questions to determine urgency and Score questions to assess complexity, enabling automated prioritization.
Review sentiment analysis: Analyze product or service reviews by asking Choice questions to classify sentiment as positive, neutral, or negative with confidence scores, improving moderation workflows.
Quality rubric scoring: Evaluate sections of lengthy documents against defined rubrics by scoring each part numerically to maintain consistent quality assessments.
Churn risk prediction: Given interaction logs or customer feedback, predict the likelihood of churn with probability distributions and confidence to guide retention strategies.
Team collaboration via playbooks: Publish proven question sets or decision criteria as reusable playbooks for teams to standardize evaluation processes.
Automated tool-call validation: Evaluate conditions before executing critical API calls by sending typed questions and directly receiving validated, type-safe decisions.

Key Features

Typed answers with probabilities and confidence values
Supports three question types: Choice, Score, Noul
Single evaluation can include up to 12 typed questions
Credit-based pricing with automatic refunds on provider errors
Ability to pin reference state for reusable context

Pricing

JevStation offers a freemium pricing model. Upon creating an account, users receive 50 free credits, which are sufficient for about 50 standard evaluations. Each standard evaluation—consisting of up to 8,000 characters of state and up to 5 typed questions—costs 1 credit. Larger evaluations, either exceeding 8,000 characters or involving more than 5 questions (up to 12), cost 3 credits. Credits are only deducted upon successful completion of an evaluation; any failures or provider errors result in automatic credit refunds, ensuring cost control and fairness. Detailed credit balance and usage information can be managed within the user's account dashboard.

Pros & Cons

Pros

+ Provides typed, structured answers with probability distributions and calibrated confidence scores, enabling precise decision-making.
+ Eliminates need to parse natural language output by returning direct coded decisions, simplifying integration with software.
+ Supports multiple question types (Choice, Score, Noul) for versatile evaluation needs within a single API call.
+ Features reusable pinned context states, enhancing efficiency for repeated or context-dependent evaluations.
+ Offers clear cost controls with credit-based pricing and automatic refunds on evaluation failures, alongside privacy by default.
+ Enables community sharing via open submissions to a shared catalog, fostering collaboration and reuse.

Cons

- Limited to decision modeling; does not support natural language generation or free-form text outputs.
- Maximum of 12 typed questions per evaluation may restrict very complex queries requiring broader analysis.
- Relatively specialized tool; users unfamiliar with typed decision contracts may face learning curve.
- Freemium tier credits are limited; heavy usage requires purchasing additional credits.

Frequently Asked Questions

What types of questions can I ask JevStation?

JevStation supports three typed question formats: Choice (categorical selections), Score (numerical values), and Noul (null or unknown). Up to 12 such questions can be included in a single evaluation.

How does JevStation charge for evaluations?

Each evaluation costs credits based on size and complexity—1 credit for up to 8,000 characters and 5 or fewer questions, and 3 credits for larger evaluations. Credits are billed only on successful runs and refunded automatically if errors occur.

Is my data used to train JevStation models?

No. JevStation is privacy-oriented and does not use your submitted data for training purposes. Your evaluations remain private by default.

Can I reuse evaluation context across multiple queries?

Yes. JevStation allows you to pin reference states or standing instructions to a thread, enabling reusable context to inform subsequent evaluations efficiently.

How do JevStation's outputs differ from regular language models?

Unlike text-generating LLMs that output prose requiring parsing, JevStation returns direct typed decisions with associated probabilities and confidence scores, suitable for immediate programmatic use without parsing.

Is there a sandbox or playground to test JevStation without signup?

Yes. JevStation offers a browser-based playground accessible without signup where you can run evaluations and experiment with question sets using free credits.