Skip to content

v3.1.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 12 Oct 08:25
· 8 commits to main since this release
v3.1.0
4f31c89

PinkCollab v3.1.0: OMP RPC v2 support for large model catalogs

Highlights and fixes

  • Gateway negotiates RPC v2 when OMP advertises support and waits for successful negotiation before sending ordinary RPC requests. Indexed rpc_chunk responses are reassembled and validated, allowing model catalogs larger than the 1 MiB physical frame limit to reach the existing model picker API. This addresses issue #5, where large catalogs caused "Model catalog unavailable".
  • Chunk reassembly is bounded by local and advertised transport limits. Invalid, oversized, interrupted, or incomplete streams fail explicitly instead of leaving requests waiting indefinitely.
  • Startup RPCs no longer hold the feedback-ordering gate while awaiting responses, allowing ordinary events to drain through the bounded output queue without blocking startup or session reattachment.
  • Deployment documentation clarifies connection choices, background service management, pairing, upgrades, and troubleshooting.

Compatibility and upgrade

  • Legacy OMP RPC v1 remains supported, but it cannot transfer model responses above its physical frame limit. Large catalogs require an OMP build advertising RPC v2 support.
  • Android/Gateway Protocol 4 and the Gateway database schema are unchanged. Upgrade Android and the Gateway from the same release.
  • Install the signed Android APK over the existing app to preserve pairings and app data; do not uninstall the production app.
  • The npm launcher and all five platform packages use version 3.1.0. Update with npm install -g pinkcollab@3.1.0, then run pinkcollab setup using your existing transport configuration. External HTTPS deployments must retain their --transport external --public-url https://... settings.

Verification

  • Gateway regressions passed, including a 642-model catalog with a logical response larger than 1 MiB, tail-model selection, thinking-level changes, catalog refresh, and startup event bursts in v1 and v2.
  • A real installed OMP smoke run passed handshake, model catalog, thinking levels, and session resume without sending a model prompt.
  • The real OpenRouter large-catalog Android UI scenario has not been exercised in this validation; manual validation steps are documented in docs/development.md.
Assets 9
Loading