v0.3.3 — web dashboard render fix

Fixed

• Web dashboard 500 on fresh installs (community fix, thanks @alsldlflglhljlk, #5). All TemplateResponse calls now use the modern Starlette (request, name, context) signature, and the fastapi floor is raised to >=0.110 (Starlette new enough for that signature). Only affects the grid-inference-worker web control dashboard; streaming/job handling is unchanged.

v0.3.1 — streaming-only + token cap fix

Two operator-facing fixes.

Fixed

• Streaming-only. The legacy /v2 HTTP poll loop is retired (the grid's /v2 API is gone and returns 410). GRID_STREAMING now defaults to true, every start path uses the multi-backend supervisor, and the dead /v2 client is removed. Workers that were silently polling /v2 now connect over WebSocket. Update to fix the 410 Gone errors.
• No more token cap. The worker was clamping every request's max_tokens down to GRID_MAX_LENGTH (4096) — cutting long outputs off mid-sentence and starving multi-step tool calls. It now honors the client's requested max_tokens as-is and only falls back to a default when none is given. Defaults raised to 32768 output / 131072 context.

Notes

• Multi-backend: every backend in GRID_BACKENDS is served (set up via the quickstart, each validated live).
• To allow even larger single outputs, set GRID_MAX_LENGTH. Existing single-backend .env configs keep working.

v0.3.0 — multi-backend quickstart + supervisor

Multi-backend setup wizard, and the worker now serves every configured backend (not just the first).

Added

• Multi-backend quickstart. The setup wizard configures several backends in one run (an "add another backend?" loop). Each backend is validated live: the model is picked from the backend's real served list, then a real completion is run against (url, key, model) before it goes on the grid. Stale keys and wrong model names are caught at setup instead of failing silently.
• Streaming is now the default connection mode. Legacy HTTP polling is opt-in (the /v2 queue is retired).

Fixed

• run_worker.py and the headless streaming path now launch the multi-backend supervisor (run_workers()), which keeps one connection alive per backend and restarts dead ones. Previously only the first backend in GRID_BACKENDS started, so a 2nd/3rd model never came online.

Notes

• Existing single-backend .env configs keep working unchanged.

v0.2.1 — fix dashboard on modern Starlette

Fixes

• Cannot compile/run from source (#2) — the web dashboard used the old
  TemplateResponse(name, context) signature, which breaks on current Starlette
  (seen as a jinja2 unhashable type: dict error on first request). All dashboard
  routes now use the modern TemplateResponse(request, name, context) form, and the
  fastapi floor is raised to >=0.110 (bundles starlette>=0.36) so the signature
  is always present on a fresh pip install.

Upgrade: git pull && uv pip install -e . (or reinstall from this release).

v0.1.0 - Initial Release

First early release of the Grid inference worker. Run a local model, connect to the Grid, and start earning.