Question 1

Is a cached decision safe to reuse, or does the model drift under me?

Accepted Answer

The assumption is stated as approximate: a decision is treated as roughly a pure function of model, schema and state, which is what makes it memoizable at all. recall only checks the local ledger and never calls a backend, exiting 3 on a miss so CI can branch. decide recalls first and falls back to your configured backend on a miss, so you decide how much to trust a hit.

Question 2

Does the cache key end up containing my users' personal data?

Accepted Answer

Not by design. State is redacted and canonicalized before hashing, so the fingerprint covers decision-relevant content only; emails, phones, long digit runs, ids and timestamps never enter the key. For sensitive schemas a per-schema salt keeps outsiders from enumerating decided states. Published bundles carry fingerprints and answers only, never raw state.

Question 3

Can I hand a cache to someone with no API key at all?

Accepted Answer

Yes. publish bundles the caches already in your ledger into one file per schema, and add merges someone else's bundle while registering the schema so recall and decide resolve immediately. A bundle is plain JSON, so you can inspect it before adding. Reads are keyless; the hosted index mints a free write key on first remote publish so nobody can overwrite another publisher's cache.

Question 4

Where does it run, and does my API key ever reach it?

Accepted Answer

It ships as one static binary of roughly 2 to 3 MB for macOS and Linux on arm64 or x64, with an embedded in-memory map plus an append-only log, so a point lookup is a map access. It never proxies or resells inference and your key stays on your machine. On serverless or edge you host the daemon once and call it over HTTPS behind a bearer token.

Question 5

How do I install the jevcache CLI?

Accepted Answer

curl -fsSL jevcache.sh/install | sh installs a single static binary of roughly 2 to 3 MB for macOS or Linux on arm64 or x64, verifying a SHA-256 checksum before installing.

Question 6

How does jevcache serve decisions over HTTP?

Accepted Answer

jevcache serve exposes POST /decide, which caches on a miss, and POST /recall, which never calls a backend; when JEVCACHE_SERVE_TOKEN is set the instance requires a bearer token.

Question 7

Does the Rust CLI produce the same cache keys as the TypeScript core?

Accepted Answer

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

Jevcache — hyperspaceai

About this project

Highlights

Questions about this entry

Other listings of this project

From the Jev guides

From the Jev AI model community