Skip to content

RELEASE.md

Latest commit

 

History

History
29 lines (18 loc) · 1.99 KB

RELEASE.md

File metadata and controls

29 lines (18 loc) · 1.99 KB
Raw

Release and distribution

Before treating any artifact as GA-ready, review and resolve the blocker list in GA_RELEASE_READINESS.md. The current repository can pass source-level MVP checks, but the GA path still requires the product shell, real worker, runtime packaging, signing, and strict release-machine E2E work called out there.

Reproducible build

  1. npm install
  2. . .open-code/toolchain/env.sh if you are using the sandboxed toolchain.
  3. cargo fmt --check and cargo clippy --workspace --all-targets -- -D warnings with rustfmt and clippy installed.
  4. OPEN_CODE_MEMORYD_PROBE_STRICT=1 npm run test:memoryd:probe on a host that can bind localhost.
  5. npm run test:e2e on a Mac/Linux machine with Ollama and the target local Gemma model ready.
  6. npm run build
  7. npm run package

Open VSX

Publish the generated .vsix to Open VSX under your namespace after you have an access token. This repo uses publisher open-code in the extension package.json; change it to a namespace you own before publishing.

Desktop product

For a full branded Electron app, merge this extension into a Code-OSS fork and follow FORK_AND_RELEASE.md: reproducible build docs, signed binaries where required, changelog and attribution, extension strategy (Open VSX first), credential policy.

Model packaging

Default open model: Gemma 3 4B on localhost (Ollama-compatible or your bundled llama.cpp sidecar). The extension default is gemma3:4b for the MVP bootstrap, and the current MVP manifest lives at runtime/model-manifest.json. Pin exact model artifacts and checksums in your server manifest before a consumer desktop release. First-run model download should be only from your updater, not a separate end-user Ollama install, when you ship a consumer build.

Credentials

Store API keys in the OS keychain; memory stores metadata only. See SECURITY.md.