Skip to content

2026.11.03.1

Choose a tag to compare

@github-actions github-actions released this 03 Nov 15:06
· 40 commits to main since this release
bc77ac7

Overview

This release adds three new provider presets (llm.buzz, CloudGPU, and CheapAIModel) and makes it possible to send several images to image-edit models in one request. It also cleans up ChatGPT sign-in, improves quota tracking for Ollama and ChatGPT Enterprise accounts, and fixes requests that failed because of conflicting output-token limits.

✨ New Features

  • New provider presets: llm.buzz, CloudGPU, and CheapAIModel are now in the "Start from a preset" picker. Choose one and paste your API key to finish setup. (#1027) @bigconnect (#1045) @mrhy7176 (#1050)
  • Partial configuration updates: Management and MCP configuration updates can now change only the fields you send and leave everything else alone. Add ?merge=deep to the request to turn this on. Nested settings are merged, and setting a value to null removes it. Without the option, updates still replace the whole configuration as before. (#1048)
  • Ollama balance tracking: Ollama quota checks now read your account balance directly from Ollama. They show your monthly allowance and purchased credits alongside the older usage limits, and no longer need a session cookie. (#1046)

🐛 Bug Fixes

  • Multi-image edits: Image-edit requests that included more than one image either failed or quietly used only the first image. Every image is now passed along, and single-image requests work exactly as before. (#1047) @glade-at-gigwell
  • Conflicting token limits: With Auto Compat turned on, some requests that set an output-token limit sent that limit to the provider twice under two different names. Providers that accept only one of them rejected the request. Plexus now sends just one. (#1049) @vannomad
  • Sign in with ChatGPT: The ChatGPT login no longer fails as soon as you start it. The provider list now shows a single "OpenAI (ChatGPT Plus/Pro)" entry, which supports inference, model discovery, and quota tracking. Starting a new login now cancels any earlier attempt that is still in progress. (#1040, #1043)
  • ChatGPT Enterprise quota: Quota checks for ChatGPT Enterprise and Business accounts no longer fail with "Check failed". These accounts now show a "Monthly" usage meter. (#1044)

🔧 Improvements

  • The legacy OpenAI Codex option is no longer offered when you add a new OAuth provider. Existing setups that use it keep working. (#1039)