mirror of
https://github.com/cogwheel0/conduit.git
synced 2026-08-29 13:31:44 +00:00
0976724146
* Add Open WebUI 0.10 compatibility Update Conduit for newer Open WebUI API and streaming behavior while preserving legacy fallbacks for older servers. * Fix Open WebUI compatibility review findings Address PR review edge cases around idempotent toggles, knowledge fallbacks, structured output merging, and task-socket recovery semantics. * Address follow-up review edge cases Handle legacy knowledge prefetch failures, bound pagination, harden code-interpreter fences, and keep structured streaming plain text synchronized. * Preserve structured output on assistant versions Keep parsed structured output attached to alternate assistant versions when serializing chat history back to Open WebUI. * Tighten structured output recovery edge cases Refine recovery completion inference, stale content handling, detail-only rendering order, and pending tool-call state from review feedback. * Harden remaining Open WebUI edge cases Avoid retrying stateless toggles, bound tag pagination, narrow legacy fallbacks, and support custom tool-call outputs from structured responses. * Assert preserved version output payload Tighten the structured-output regression test to verify the full assistant-version payload. * Avoid provider refresh during chat build Lazy-read voice input outside initState and avoid footer subscriptions that can schedule provider refreshes while assistant rows are building. * Address remaining review edge cases Handle wrapped chat toggle state, avoid stale rendered-details reuse, preserve newer structured text, and harden structured-output parsing for reasoning and terminal statuses. * Preserve plain content during detail cleanup Restrict semantic detail stripping to renderer-owned blocks and keep modern knowledge-file text from nested data payloads. * Clean up latest review edge cases Recognize more upload ID shapes, clear stale semantic details for plain outputs, and seed rendered tool-call keys to prevent duplicate placeholders. * Avoid duplicate knowledge uploads Stop falling back to legacy uploads after ambiguous modern upload success, narrow attach 400 fallbacks, and preserve nonsemantic content when removing stale generated details. * Tighten stale detail and upload edge handling Expand modern upload ID extraction and keep newer plain snapshots from being overwritten by stale stripped streaming text. * Handle final knowledge API edge cases Fallback on non-JSON knowledge file responses and constrain upload file-id extraction to file-shaped payloads. * Tighten final API review fixes Avoid ambiguous upload ID extraction and surface unconfirmed conversation toggles instead of reporting success. * Fail fast on unconfirmed API outcomes Preserve audio defaults, reject failed knowledge deletes, and avoid accepting ambiguous upload IDs or mismatched toggle confirmations. * Preserve API config defaults during audio merge * Tighten knowledge fallback and upload IDs |
||
|---|---|---|
| .. | ||
| models | ||
| providers | ||
| services | ||
| utils | ||
| views | ||
| voice_call/presentation | ||
| voice_mode | ||
| widgets | ||