Repository navigation
Releases: vinilana/jev-gateway
Releases · vinilana/jev-gateway
Release list
v0.5.1
v0.5.0
v0.4.3
v0.4.2
v0.4.1
0.4.1 (2026-09-20)
Fixed
- keep /health to "ok" once the gateway has a key (57651ff), closes #8
- keep hostile tool names out of hints, and pass malformed requests through instead of failing (40fd608)
- keep tool names that are not a single inert token away from Jev and the LLM (f306026), closes #8
- pass malformed requests through instead of answering 500 (174d84a), closes #8
- route Chat Completions requests that also carry built-in tools (f6fd00c), closes #8
v0.4.0
v0.3.1
First version published by the release workflow. It carries everything listed under 0.3.0, which
was tagged but never reached npm.
Changed
- Releases are automatic: pushing a
vX.Y.Ztag publishes to npm through trusted publishing
(OIDC), with provenance and without tokens, then creates the GitHub release. See
docs/releasing.md. - Pull requests and pushes to
mainrun the type check, the tests and the build in CI, on Node
22.15 and 24.
Fixed
- The Homebrew workflow checks for its token before anything else, instead of polling npm for half
an hour and then failing when no tap is configured.
Install or upgrade: npm install -g jev-gateway
v0.3.0
Added
- OpenCode support.
jev-opencoderuns stable OpenCode (v1) through a gateway on port 8791,
injecting a provider throughOPENCODE_CONFIG_CONTENTso nothing in your OpenCode config is
written. Checked against OpenCode 1.18.31. Thanks to @viniciosrab (#4). - Gemini API support. The gateway routes
POST /v1beta/models/<model>:generateContentand
:streamGenerateContent, forcing a tool throughtoolConfig.functionCallingConfig, and
jev-geminipoints the Gemini CLI at a gateway on port 8788. For clients that use a Gemini API
key. Unit-tested, not yet run against the real API. Thanks to @andrecodexvictor (#5). - Homebrew release automation. A workflow that updates a formula in
vinilana/homebrew-tap
after each release. It does nothing until the tap and its token exist; see
docs/homebrew-tap.md. Thanks to @viniciosrab (#2). - Token metering understands Gemini's
usageMetadata, streamed or not.
Fixed
- The proxy dropped the leading
/v1from any path that started with it, so/v1beta/...was
forwarded asbeta/.... Only the/v1segment is dropped now.
Install or upgrade: npm install -g jev-gateway
v0.2.2
Changed
- How each agent is pointed at a gateway now lives in
bin/clients.mjs, andbin/launcher.mjsexports what is needed to start a gateway process. The benchmark uses both. No change in behaviour. - The README links to the benchmark and its first results.
Includes the 0.2.1 fix: token metering works with Codex on a ChatGPT subscription.
Install or upgrade: npm install -g jev-gateway
v0.2.1
Fixed
- Token metering now works with Codex on a ChatGPT subscription. That backend streams its replies without a content-type header, so 0.2.0 read them as plain JSON and recorded no tokens at all.
Install or upgrade: npm install -g jev-gateway