Changelog

For the full Chinese changelog with identical detail, see 更新日志.

v0.5.4 (2026-05) — Custom models & conversation stability

Custom Chat models — add OpenAI-compatible providers, API keys, and Chat model IDs in settings.
Grouped model picker — official models appear first; custom models are grouped by provider.
Provider/model diagnostics — tests show redacted curl, HTTP status, content type, and response preview when Base URL, key, or streaming protocol checks fail.
Conversation stability — empty general conversations are reused, placeholder titles are retried after the first turn, and server-side timeline sequence values keep streamed messages and tool results ordered.
Read-only subagents — explore_project, review_project, and research_context handle broad inspection, evidence review, and research context in separate sub-conversations.
Subagent reports — summaries, evidence paths, and sources can be saved under @system/@subagent/ for later reference.
Quieter tool display — dedicated tool cards suppress duplicate system fallback messages, and the internal finish_agent close-out tool stays silent.
AI export dialogs — AI share cards use internal scrolling with fixed footers for small windows. The Word generation workflow remains in code, but the main Export dialog currently hides that entry.
Deployment docs — production shared entry network and environment examples are aligned for the main, website, and one-api suites.

URL-to-Markdown — paste a web page and let the Agent turn it into usable writing context.
Research flow — @web, @base, and URL context fit better into writing and review.
Long-form rendering — Markdown block cache, streaming throttling, and lazy Mermaid rendering.
Tool-call resilience — display errors degrade locally instead of breaking the whole AI panel.
Quality guardrails — broader Agent scenario coverage and stability checks.

Persistent memory — remember stores user-scoped preferences and facts across sessions.
Scene-based models — pick defaults per scenario; account-available model lists update from the service and retired models are pruned.
Search tools — multi-folder grep, regex, structured tool results (function calling / tool_use).
Mobile layout — responsive three-column tweaks and touch-friendly controls.
Usage & billing UI — richer stats, formatting, refund/adjustment handling.
Performance — optional auto-summary for very long chats, batched tools, better parsing/error recovery.

Toggle models on/off in settings; cloud-account preferences apply across devices.
Heatmap and writing calendar for AI usage; per-model aggregates.

Markdown import hierarchy fixes; outline drag/drop ordering fixes; faster, safer save when switching files.

Windows desktop: outline editor, AI tools, Milkdown, Mermaid, KaTeX, Markdown ZIP export, local-first storage, bilingual UI.