shipwithjev

Catalog / Research & data

0360GitHub

jevcache

Local decision cache keyed on (model, schema, state), with redaction and canonicalisation before hashing, for cheaper repeats and deterministic replay in CI.

Local decision cache keyed on (model, schema, state), with redaction and canonicalisation before hashing, for cheaper repeats and deterministic replay in CI. No license file at the time of writing.

hyperspaceai/jevcacheREADME ↗
# jevcache

**The decision ledger for Jev-class models.** A Jev decision is (approximately) a pure
function of `(model, schema, state)` — so it's memoizable. jevcache caches those
decisions locally: same decisions, fewer bills, and deterministic replay in CI. It's
backend-agnostic — point it at any local decision endpoint, a hosted Jev provider, or
a mock — and it never proxies or resells inference (your key stays on your machine).

- **Local-first & offline.** An embedded, zero-dependency store (in-memory map +
  append-only log). A point lookup is a map access — instant.
- **Private by construction.** State is redacted and canonicalized *before* it's
  hashed, so the fingerprint is over decision-relevant content only; PII (emails,
  phones, long digit runs) and volatile fields (ids, timestamps) never enter the key.
  For sensitive schemas, a per-schema `salt` keeps outsiders from enumerating decided states.
- **Deterministic CI.** Publish fixtures, `replay` them, catch model drift.


> **This is the distribution repo** — the `jevcache` CLI binary and docs.
> Install with `curl -fsSL jevcache.sh/install | sh`, or download a binary from [Releases](https://github.com/hyperspaceai/jevcache/releases).

## Install

```bash
curl -fsSL jevcache.sh/install | sh
```

Installs a single static binary (~2–3 MB, no runtime) for macOS/Linux, arm64 or x64.
The installer verifies a SHA-256 checksum before installing. `recall()` checks the local
ledger only (never a backend); `decide()` recalls, then on a miss routes to your configured
backend (`JEVCACHE_BACKEND=local|jev|mock`) and caches the result.

The CLI is written in Rust; its fingerprint is byte-identical to this repo's TypeScript
core, so a Rust ledger, the TS internals, and the hosted commons all share the same keys.


## CLI

Also filed under Research & data

  1. 0620

    Jev Score: rubrics, scores and confidence

    A worked guide to Jev's Score primitive: writing a request, defining rubric levels, reading recorded probabilities and the weighted-score math.

    Jev Trader · Research & data

  2. 0614

    Bot journey classification in WebDecoy

    Sends a detected bot's last 48 request paths to Jev, which picks what it's after (prices, articles…) and how it crawls (pagination, IDs…), or unknown.

    WebDecoy · Research & data

  3. 0610

    Fake / Real: link fact-checker with Jev as judge

    Paste an article or post link. Fake / Real extracts its claims, finds outside evidence, and has Jev judge whether the evidence supports or contradicts each one.

    @DansiDanutz · Research & data

  4. 0609

    AutoRubric: rubric-based evaluation

    Combines rubric science and LLM-as-a-judge research to grade outputs with AI judges: LLMs, decision models like Jev, or both.

    @deliprao · Research & data