User repository on GitHub
Ask Jev over columns of data: closed-set questions, cached and joined back.
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Integrations & Adapters
Hunch — steven-shoemaker
Python library that maps Jev's Choice, Score and Noul onto verbs like classify, score, check, where, extract, verify and rank, keeping data shape in and out, batching one request per distinct row and caching answers.
- Stars
- 4
- Language
- Python
- License
- MIT
- Published by
- Community
- Author
- steven-shoemaker
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is Hunch — carldaws.
About this project
Judgment as a Python function. Ask a question about a string, a list, or a whole DataFrame column, and get an answer for every row that your code can branch on. Also available for TypeScript, with the same verbs over arrays. Using a coding agent? Give it the hunch skill so it uses these verbs instead of writing prompt-and-parse code:
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- Choice, Score and Noul map onto the verbs, and each verb is also a pandas accessor, so the DataFrame and Series forms are the same call.
- One request per distinct row: duplicates are asked once, several questions about a row batch together, requests run in parallel and answers are cached.
- detail=True returns the whole distribution plus a shape, sure when one option dominates, split when two are close, unsure when the evidence is flat.
- extract finds every candidate with code and lets Jev pick the one that answers the field, so values are copied from the text and unstated fields come back None.
Watch out
Every verb needs a TypeSafe key, and a DataFrame row is the unit of judgment, so a wide table means a wide prompt; pass columns= to narrow what Jev sees. Costs scale with the number of distinct values even though repeats are cached and requests are parallel. An LLM is optional and used only by generate, discover, refine and escalation. The README warns that predictions about behaviour sit near 0.3 to 0.4 when the row says nothing either way, so where is meant to be ranked with detail=True rather than filtered on a threshold. 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.
Because a chat model returns prose plus a number you cannot defend, and a different number on retry. Here each verb is a typed question, every answer keeps its probabilities, and detail=True adds a shape, sure, split or unsure, so your code decides which rows to trust, re-ask, or send to a person. Thresholds and weights stay in your code rather than in a prompt.
One per distinct row. Duplicate values are asked once, several questions about the same row go in a single request, requests run in parallel, and answers are cached so re-running costs nothing. Each row is the thing being judged and Jev sees every column unless you pass columns=, with anything in context= riding along on every row.
It copies them. Code finds every candidate, meaning every dollar amount, date or email, and Jev picks the one that answers the field while seeing the words around it. The value always comes from the text and a field the text does not state comes back None. Built-in finders are email, url, money, number, percent, phone and date; anything else is a regex or a function.
No. Jev makes every call and your code keeps the thresholds and weights; only generate, discover, refine and escalation touch an LLM, for fake data, category names, rewrites and second opinions on hard rows. If you do want one it can be Anthropic, OpenAI, Azure, OpenRouter, a local Ollama model or your own callable, and Jev itself can be reached through an OpenRouter or Vercel gateway.
It gives the whole distribution plus a shape: sure when one option dominates, split when two are close, and unsure when the evidence is flat. Shapes tell your code which rows to trust, re-ask, or send to a person.
It distinguishes the source says otherwise from the source does not say, and text a claim puts in quotes must appear in the source word for word, which catches made-up quotes. It returns supported, contradicted, not mentioned or misquoted.
No. rank scores every candidate on each dimension, normalizes and applies your weights, and changing the weights never re-runs inference. Adding query= turns it into a reranker for a specific target.
More Jev resources like this
25GitHub stars
Agent runtime with a terminal UI and a messaging gateway that keeps memory, self-built skills, session search, scheduling, and subagents across Telegram, Discord, Slack, WhatsApp, Signal, and CLI.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
wundercorp/lokiPython
wundercorp/lokiPython51GitHub stars
GraphRAG layer that runs a 4-phase ingest-to-answer pipeline over Neo4j, Memgraph, AGE or Kuzu using Laya or Jev decisions instead of per-hop LLM calls.
Jev Integrations & Adapters#community#jevbest#catalog-import
23GitHub stars
Memory-maintenance library that checks every stored memory against each new event with Jev yes-or-no votes, then marks stale facts superseded or contradicted instead of editing them.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
chopratejas/invalidatePython
chopratejas/invalidatePython14GitHub stars
Python library for schema-enforced JSON extraction into Pydantic models, with provider-native JSON mode fallback, auto-repair, token and cost tracking, plus agents, RAG, and eval helpers.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
jhd3197/PrompturePython
jhd3197/PrompturePython19GitHub stars
Ruby gem that models Jev as field predicates: feels and decide bind to model attributes with shared definitions, thresholds, confidence gates, batching, and test tapes.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
Qew7/jev-feelsRuby
Qew7/jev-feelsRuby5GitHub stars
Community Agent Zero plugin that exposes one typesafe_query tool for combined Choice, Noul, and Score questions over the same evidence, with probability cards in chat.
Jev Integrations & Adapters#community#jevbest#catalog-import
Community
3clyp50/a0-typesafe-aiPython
3clyp50/a0-typesafe-aiPythonFrom the Jev community
Posts from builders shipping with Jev right now.