User repository on GitHub
Go client for TypeSafe AI's System One API (Jev), with optional Langfuse instrumentation
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Repos & SDKs
Jevgo — fgn
Community Go client for the TypeSafe AI System One API covering the same surface as the official Python and TypeScript SDKs, using functional options, context cancellation, sentinel errors, `log/slog`, no dependencies outside the standard library, and optional Langfuse tracing.
- Stars
- 2
- Language
- Go
- License
- MIT
- Published by
- Community
- Author
- fgn
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is Jevgo — devbackend.
About this project
A Go client for TypeSafe AI's System One API and its flagship model, Jev. Send a state and typed questions; get typed answers and probabilities your code can act on directly. It covers the same API surface as the official Python and TypeScript SDKs with the same defaults, written the way a Go library should be: functional options, context cancellation, sentinel errors, log/slog, and no dependencies outside the standard library. See Compatibility for where behavior intentionally differs.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- Go
Highlights
- Response validation is strict: a missing answer or a distribution that does not sum to one becomes a `ResponseValidationError` rather than a silent zero.
- Answers are a sealed interface you can type switch on, and an answer kind newer than the SDK arrives as `UnknownAnswer` with its raw JSON intact.
- A compatibility table lists the differences: one-level Scores accepted, unknown answer kinds as raw JSON, `Retry-After` over 60s falls back to backoff.
- Optional Langfuse tracing lives in a separate `contrib/langfuse` module, recording one generation per call with model, input, output, usage and request ID.
Watch out
It is a community client with no affiliation, and its compatibility table records deliberate differences from the official SDKs, including a total call budget that is off by default where Python defaults to 30 seconds. Bodies over 16 MiB fail as `ErrResponseTooLarge` and are never retried, and while credential headers are redacted from logs, server messages and debug-level bodies can still contain sensitive content. 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.
Defaults match on timeouts, retries and headers, and the README carries a table of where it does not. A Score with a single level is accepted, missing or null response fields raise a validation error, an unknown answer kind becomes `UnknownAnswer` with raw JSON, `Retry-After` above 60 seconds falls back to backoff, and the total call budget is off by default where Python uses 30 seconds.
It validates against the API contract and against the questions as sent, including `Request.Extra` overrides. A missing answer, an option or level you never asked for, a null probability, a distribution that does not sum to one, or a choice that is not the most probable option all return a `*jev.ResponseValidationError` naming the field rather than a silent zero. `Request` and `Response` are not the wire format, the client encodes them.
There is a `Tracer` interface observing every `SystemOne` call that passes local validation, with the wire body and effective model at start and the decoded response, error and attempt count at end, and the context it returns parents the HTTP requests correctly. Pass `WithTracer(jevlangfuse.NewTracer(lf))` for one Langfuse generation per call, or supply your own `*http.Client` with a `RoundTripper` for transport-level instrumentation.
Errors are sentinels you can test with `errors.Is`: `ErrRateLimit` after retries, `ErrTimeout`, `ErrInvalidRequest` for requests rejected before sending, plus `ErrAuthentication`, `ErrUnprocessableEntity`, `ErrOverloaded` and `ErrInternalServer`. `WithTimeout` bounds each attempt while `RetryPolicy.TotalTimeout` bounds the whole call and reports as `ErrTimeout`, and your own context deadline always applies on top of both.
go get github.com/fgn/jevgo, then either pass the key with jev.NewClient(jev.WithAPIKey(...)) or set TYPESAFE_API_KEY and call jev.NewClient() with no options. Send client.SystemOne(ctx, jev.Request{...}) with a State and a Questions map keyed by names you choose, and read results through resp.Answers by type switch or through typed getters such as resp.Choice("department"). State, instructions and criteria all accept strings or structured JSON, so a []string of levels or a map[string]string of options works as is.
jev.Noul returns NoulAnswer with the probability of yes. jev.Choice returns ChoiceAnswer with the selected option in Choice, Probabilities per option and Confidence. jev.Score returns ScoreAnswer with a weighted position in Score, Probabilities and Legend per level, Confidence, and Level() for the most probable level, which is worth noting because Score is a weighted average rather than a level. An answer kind newer than the SDK arrives as UnknownAnswer with the raw JSON intact, and RawQuestion and Request.Extra let you send fields this version does not model.
Explicit options win over environment variables, which win over defaults, and an empty string means not set. WithBaseURL reads TYPESAFE_BASE_URL and defaults to https://api.typesafe.ai, WithDefaultModel reads TYPESAFE_DEFAULT_MODEL and defaults to jev-latest, WithLogger is off unless TYPESAFE_LOG_LEVEL is set, WithTimeout bounds each attempt at 10s, and retries default to 2 with 500ms to 5s backoff honoring Retry-After. Every option is also accepted per call as an override, and built options are safe to reuse across goroutines.
More Jev resources like this
45GitHub stars
Go SDK plus a pipeline-friendly CLI for typed decisions, covering single questions, batched question maps, retries, and line-oriented filtering over stdin.
Jev Repos & SDKs#community#jevbest#catalog-import
Community
mattn/go-jevGo
mattn/go-jevGo9GitHub stars
Go SDK for the TypeSafe API that asks typed Noul, Choice, and Score questions about a state and returns probability distributions with confidence and typed accessors.
Jev Repos & SDKs#community#jevbest#catalog-import
Community
Tangerg/typesafe-sdk-goGo
Tangerg/typesafe-sdk-goGo4GitHub stars
Community Go client for the TypeSafe System One evaluation API with raw structs and a builder that produce the same EvaluationRequest, defaulting to jev-latest.
Jev Repos & SDKs#community#jevbest#catalog-import
Community
kisshan13/typesafe-ai-goGo
kisshan13/typesafe-ai-goGo1GitHub stars
Unofficial, independently maintained Go client for the TypeSafe AI System One API with no third-party dependencies: choice, score and noul questions, typed accessors that error instead of returning zero, and per-call timeout, retry and header overrides.
Jev Repos & SDKs#community#jevbest#catalog-import
Community
cole-gillespie/typesafe-goGo
cole-gillespie/typesafe-goGo1GitHub stars
An unofficial Go client that keeps Jev answers in your own types, so a choice over your enum yields that enum and the compiler checks the switch you write on it.
Jev Repos & SDKs#community#jevbest#catalog-import
Community
anilsenay/jevGo
anilsenay/jevGo1GitHub stars
Unofficial Go SDK for TypeSafe AI's System One API, kept deliberately minimal: one client, a state map, a map of typed questions, and answers read back by question id.
Jev Repos & SDKs#community#jevbest#catalog-import
Community
guillemus/jev-goGo
guillemus/jev-goGoFrom the Jev guides
Original write-ups on what this entry is for.
Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.
Jev Repos & SDKs#open-models#research#local
Read article
From the Jev community
Posts from builders shipping with Jev right now.