Are the reported runs real live outputs or projections?

Yes for the documented runs: on-topic approved and off-topic blocked in the guardrail, all three routes correct, and market-pulse source selection with verify passes. They are examples, not guarantees for your queries.

What stops Jev from inventing an endpoint or wasting money?

Jev choice can only return one of the three fixed Monid endpoints, and the guardrail blocks calls below 0.7 confidence before any money is spent. Async /v1/run polling handles execution.

Do relevance and budget gates interfere with each other?

Both gates stay independent: Growth hit its cap while Support and Data-Eng kept working, and an irrelevant query was blocked by relevance even when budget remained. Spend is attributed per team.

How do I handle Monid async runs and inconsistent cost fields?

Monid /v1/run returns 202 with a runId, so poll GET /v1/runs/{runId} until COMPLETED or FAILED. Read cost through the fallback chain from cost.value through billing.reportedCost to price.amount.value.

What are the four proof-of-concept scripts?

guardrail, a spend guardrail; router, deterministic routing to one of three Monid endpoints; market_pulse, a multi-source agent with a verify pass; and cost_governance, per-team monthly caps with a relevance gate and an independent budget gate.

How does the multi-source agent choose which sources to call?

It branches the question across three real data sources using three parallel boolean checks, calls only the needed sources, and then runs a second Jev pass to verify the fetched data actually covers what was asked. In the verified run a two-part question called two sources for $0.020 total.

CommunityJev Tools & CLIs

Jev Monid Poc

Q: How does the spend guardrail decide to skip a paid call?

Before calling a paid Monid endpoint, Jev scores whether the query is actually relevant. Below a 0.7 confidence threshold the Monid call is skipped and no money is spent. In the verified run an on-topic query was approved at 0.96 confidence and an off-topic one blocked at 0.01.

Four Node scripts pairing Jev relevance, routing, and verification decisions via Vercel AI Gateway with pay-per-call Monid execution and per-team budget gates.

See highlights

Open github.com

Stars: 1
Language: JavaScript
License: MIT
Published by: Community
Author: sprah
Last verified: Oct 2026

sprah/jev-monid-poc

User repository on GitHub

View on GitHub

Proof-of-concept combining TypeSafe AI's Jev with Monid's pay-per-call API marketplace

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Four small proof-of-concept scripts combining Jev (TypeSafe AI's typed decision model, accessed via Vercel AI Gateway) with Monid (a pay-per-call API marketplace). Jev makes a fast, typed decision; Monid executes the paid call. All results below were run against the live APIs and are real outputs, not projections. Before calling a paid Monid endpoint, Jev scores (boolean/probability) whether the query is actually relevant. Below a 0.7 confidence threshold, the Monid call is skipped and no money is spent.

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

Tagscommunity github

Works forJev for Classification

Used inJev for Software & DevTools

Languages: JavaScript

Highlights

Spend guardrail skips Monid calls below 0.7 Jev confidence, so off-topic queries cost nothing.
Deterministic router confines Jev choice to three fixed endpoints with no outside routes allowed.
Market-pulse agent fans out across three sources then runs a second Jev verify pass on coverage.
Multi-tenant governance enforces per-team caps on one shared key with chargeback-style attribution.

Watch out

Proof-of-concept scripts only, with polling-loop retries, no persistent storage, and minimal error handling. 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

using-typesafe-and-pinecone

Example collection pairing Pinecone retrieval with Jev judgments, including a reranking demo that narrows full-text candidates by natural-language criteria.