huncho
Dependency-free TypeScript SDK that turns Jev answers into named decisions with hysteresis thresholds, journaling and replay.
This summary is based on how curated lists describe the project; its README was not available to review.
Overview
huncho is a TypeScript SDK described in the yibie/awesome-jev list. It wraps Jev's Noul, Choice and Score answers into named decisions that carry enter and exit thresholds, giving hysteresis so a decision does not flip back and forth around a single cut-off. Nested decision trees can be settled in one call.
It also keeps a JSONL journal of answers, lets you replay a threshold change over recorded answers without running inference again, and computes Brier scores and reliability for calibration. It can talk to TypeSafe directly, OpenRouter or the Vercel AI Gateway. No README could be fetched, so installation and API details are not confirmed here.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Named yes/no decision with enter/exit thresholds(inferred) | yes · no |
| Choice | Nested decision tree branches settled in one call(inferred) | — |
How it works
- Define named decisions with enter and exit thresholds
- Ask Jev via TypeSafe, OpenRouter or Vercel AI Gateway
- Apply hysteresis to decide state changes
- Journal answers to JSONL
- Replay threshold changes and measure calibration offline
Strengths
- Hysteresis prevents flapping decisions near a threshold
- Replay tests threshold changes without new inference cost
- Built-in Brier and reliability checks for calibration
- No runtime dependencies
Caveats
- README not available; details come from a list summary only
- Maturity and tests unknown
At a glance
- Type
- Library
- Language
- TypeScript
- Runtime
- library
- Has code
- yes
Use cases and integrations
How curated lists describe it
TypeScript ecosystem: dependency-free SDK that turns Jev `Noul`, `Choice` and `Score` answers into named decisions with `enter`/`exit` thresholds (hysteresis), nested decision trees settled in one call, a JSONL journal, replay of a threshold change over recorded answers with no inference, and Brier/reliability calibration, over TypeSafe direct, OpenRouter or Vercel AI Gateway.
yibie/awesome-jev
Frequently asked questions
What is hysteresis here?
Separate thresholds for entering and leaving a decision state, so small probability changes do not flip the outcome.
Can I test a new threshold without paying for inference?
Per the list summary, yes: it replays recorded answers from its journal.
Which providers does it support?
TypeSafe direct, OpenRouter and Vercel AI Gateway.
Related projects
Rust CLI and library that asks Jev typed questions via OpenRouter and combines the probabilities with fuzzy-logic rules into decisions.
CLIbetaRustUnofficial standard-library Go client for Jev with generic typed Choice values and local confidence thresholds.
ProjectbetaGoRust CLI for Jev: ask yes/no, choice and rubric questions about text and branch on exit codes, JSON or MCP tools.
MCP serverbetaRustTypeScript library for typed semantic decisions via Zod schemas or native batches, defaulting to Jev, with confidence-gated policies.
LibrarybetaTypeScriptMonitoring proxy and client that records Jev answer probabilities and alerts on version changes, drift, edge crowding or accuracy drops.
ProxybetaPythonComposable Python toolkit for Jev: question builders, confidence gates, batch evaluator, calibration, CLI, MCP server and agent skill.
MCP serverbetaPython