ReleasesPublished 2026-11-05v0.85.0

v0.85.0

This is the release note as published, not a rewrite of it. Release notes are published in the language they were written in.

A run whose child process keeps its output pipe open no longer hangs. Streaming a command's output used to wait without limit for the pipe to close, so a grandchild that inherited it (a test runner's worker, a background server) kept the run, its cancel button and the test suite waiting forever. The reader now runs on its own thread with a five-second grace after the child exits, cancelling kills the whole process group on POSIX and the process tree on Windows, and the test suite dumps every thread's stack if a single test runs past fifteen minutes instead of freezing silently. (#1196)
The settings search finds the pull-request and close-the-loop switches. Searching "pull request", "PR", "merge", "close the loop" (or "ciclo" in Portuguese) now lands on and focuses those two switches, and "autonomia" finds the card that holds them, in every language the app offers. Before, the search found nothing and the new close-the-loop switch could only be reached by scrolling. (#1193)

Design artboards and their exports load images from data: only. The content security policy of an artboard, and of the HTML it exports, no longer allows blob: images; every other directive is unchanged, so the policy only got stricter. Nothing the app builds used blob: (assets already arrive as data: URLs), and no existing design referenced one. (#1194)

A browser tab can become the owner's screen by pairing with a one-time code. The desktop app shows an eight-character code (five minutes, five attempts); typing it into the browser grants that tab an HttpOnly, same-site cookie for up to twelve hours, revocable from the desktop. Only the desktop issues codes — the bridge and the chat channels cannot — and a paired tab cannot turn on the bridge, full control or "Ignore permissions". Every state-changing request from it must come from the same origin. (#1198)
Discord and Telegram bots can hear voice messages and answer aloud, locally. Voice notes are transcribed on this machine with faster-whisper (pinned model revision, duration checked before decoding, a per-user rate limit). Spoken replies are off by default and use Piper only when the owner points CHIMERA_CHAT_VOICE_MODEL at a voice .onnx they chose; Chimera ships no voice, and piper-tts (GPL-3.0) lives in a separate voice extra outside [full]. (#1197)
Named OpenAI-compatible endpoints from Settings. Settings › Connections now lists your OpenAI-compatible endpoints (a local vLLM, LM Studio or llama.cpp server, or a hosted one) and lets you add, test and remove them; the test makes one bounded GET /models call to that endpoint only. Only the owner's screen can change or test them, the key is never shown, and an endpoint's key must be its own *_API_KEY variable, never another provider's credential or a CHIMERA_ setting. Closes #1133. (#1195)