Cline

• Ollama's response-start timeout now defaults to 5 minutes instead of 30 seconds, so large models that cold-load no longer fail before they finish loading — unreachable servers still fail immediately, requests are still cancelable, and an explicit requestTimeoutMs is still honored
• Ollama turns that come back completely empty (no text, reasoning, or tool call) are now retried at the model boundary instead of failing the task with "Model returned empty response"; non-empty turns stream through with no added latency, and turns that error or hit the token limit pass through unchanged
• Checkpoints are created again in VS Code and the CLI — run-boundary detection assumed the run's prompt arrives as run input, but both hosts seed it into the initial messages, so no checkpoints were ever recorded. Detection now also survives process restarts and compaction
• Checkpoint restore is now a true workspace rewind: files Cline created during a task are captured in the snapshot and restored to their checkpoint-time content, and files created after the checkpoint are removed. .gitignored paths (build output, node_modules, .env) are left untouched, and a pre-restore recovery snapshot

• Provider errors forwarded through the Vercel AI Gateway now surface the real upstream message (e.g. "This model's maximum context length is 40960 tokens...") instead of a raw Zod issue dump, and opaque object errors no longer render as [object Object]
• fetchClineRecommendedModels now returns display-ready model names, resolved through the model catalog (including Vercel and OpenRouter id aliases) under a single shared timeout budget, so hosts no longer have to map ids to names themselves
• Cline free models now resolve their OpenRouter display names in the catalog
• The live catalog no longer drops the video input capability
• On Windows, PowerShell commands now travel over UTF-8 stdin instead of the command line, so non-ASCII commands survive the active code page and long commands are not capped by the Windows command-line limit — getShellInvocation() replaces the now-deprecated getShellArgs(), and a stdin write failure surfaces as a command error instead of hanging
• Fixed sessions rooted at the filesystem root (/) failing to run any command: basename("/") produced an empty workspace hint that schema validation rejected, so every command threw

What's in 4.1.0

4.1.0 converts the stable extension to a combined A/B package: one VSIX containing both the current (legacy) extension and the new SDK-based extension, plus a small loader that activates exactly one per window, controlled by a staged remote rollout.

For nearly everyone, nothing changes — the loader activates the same extension you were running in 4.0.12. A small percentage of users (starting at 1%) are gradually opted into the new SDK-based extension, which will become the default for everyone as the rollout completes.

Details

• The rollout is staged and remotely controlled; assignments apply on window reload, never mid-session.
• If the new extension fails to activate for any reason, the loader automatically falls back to the current extension in the same window — no user action needed.
• Settings, credentials, and preferences are shared: nothing is migrated or lost when a machine switches between the two.
• Tasks created on the new extension aren't visible while on the old one (they return if re-promoted).

Legacy bundle: legacy-extension @ 3fdc186f1 (4.0.12 + free-model button wording). Next bundle: main @ 0746ea72bf6.

• cline history now opens inside the existing TUI, with resume and delete actions, instead of rendering a second view in the same process
• Connector threads (Slack, Discord, Telegram, Linear, Google Chat, WhatsApp) now recover when the session they were bound to is gone — the stale binding is dropped and the turn replays against a new session, instead of failing with "session not found" until threads.json is edited by hand
• cline --help now reports the real default --config and --data-dir paths
• The per-server timeout in cline_mcp_settings.json is now honored for initialize, tools/list, and tools/call, so slow MCP servers no longer fail against a hardcoded 5s limit (from SDK v0.0.67)
• Reasoning controls are now routed from the models.dev catalog across providers, with clamped budgets and correct per-provider encoding (from SDK v0.0.67)
• OpenRouter now defaults to anthropic/claude-sonnet-5 (from SDK v0.0.67)
• Fixed the China and international endpoint toggles being ignored for Qwen, Moonshot, and Z AI (from SDK v0.0.67)
• Legacy API keys are now migrated for every secret-backed provider (from SDK v0.0.67)

• Reasoning controls (effort, budget, on/off) are now driven by the models.dev catalog and normalized once before provider encoding, so requests match what each provider actually advertises; Anthropic's mandatory and impossible thinking modes are handled explicitly, and out-of-range budgets are clamped
• OpenRouter now defaults to anthropic/claude-sonnet-5
• The per-server timeout in cline_mcp_settings.json is now honored by the SDK's MCP clients for initialize, tools/list, and tools/call instead of hardcoded 1.5s and 5s limits — it defaults to 60 seconds and is clamped to 1–3600 seconds
• Fixed the China and international endpoint toggles being ignored for Qwen, Moonshot, and Z AI
• Legacy API keys are now migrated for every secret-backed provider instead of a subset
• Legacy OpenAI Compatible model-info overrides are now carried into the seeded models.json instead of being dropped
• Removed the "Enable R1 messages format" option from the OpenAI Compatible provider
• Fixed checkpoint restores across session resumes

• New system tray icon showing app status and how many agent sessions are currently running.
• Session history is now paginated in ten-session pages, fetching older history only when you reach the end.
• You can favorite sessions, and sessions are now ordered by most recent activity with consistent status dot colors across views.
• Subagent and teammate runs from a session now show up in the app with their status and results.
• Chat polish: tool-specific icons on tool disclosures, elapsed thinking time and restyled reasoning sections, aligned timestamps, and message actions that no longer shift the layout while scrolling stays anchored to the conversation viewport.
• Free Cline models are now supported and labeled "(free)" in model pickers, with a clear message — including reset time — when you hit the free-tier limit.
• Fixed the China/international endpoint toggles for Qwen, Moonshot, Z AI, and MiniMax being ignored, which silently routed regional users to the wrong host.
• Fixed tool calls failing when a model emitted a line number as a string (e.g. insert_line: "3"), forcing the agent to waste a round trip retrying.
• Refreshed the bundled provider and model catalog.

• Free Cline models are now supported end to end: free models show as "(free)", and hitting the free limit renders a dedicated card with the reset time (from SDK v0.0.66)
• /settings general toggles (plan/act mode, tool auto-approve, compaction mode) now persist across restarts
• Upgraded the TUI stack from opentui 0.1.102 to 0.4.3
• Fixed a grey panel left behind on screen after closing a dialog (model picker, help, command palette) — a leftover from the opentui upgrade
• Fixed a React duplicate-key warning when read_files listed the same path more than once
• Aborting a task no longer risks killing the shared hub daemon
• Connector status delivery failures are no longer fatal to the turn
• Agentic compaction is now the default context-compaction strategy, with fixes for it silently falling back to basic compaction and for tool-heavy transcripts that could never find a cut point (from SDK v0.0.66)
• Editor edits preserve a file's existing line endings, fixing failed exact-match edits on CRLF files (from SDK v0.0.66)
• Broader built-in provider coverage, now generated from models.dev (from SDK v0.0.66)
• Updated the bundled model catalog (from SDK v0.0.66)

• Support for free Cline models (cline-free): free models are labeled "(free)", priced at zero, and hitting the free tier now raises a dedicated limit error that includes the reset time
• Agentic compaction is now the default context-compaction strategy
• Fixed agentic compaction silently falling back to basic compaction on OpenAI Compatible providers (the summarizer built its handler without a base URL and hit api.openai.com), and manual compaction budgeting against a 64k fallback instead of the model's real context window
• Fixed agentic compaction never finding a cut point in tool-heavy transcripts, which produced endless "auto-compaction skipped" while context kept growing — assistant messages are now valid cut boundaries
• Connector sessions now persist and automatically reconnect after a daemon or hub restart
• Plan/act mode, tool auto-approve, and compaction mode are now persisted in global settings, with cross-process-safe writes so two hosts no longer clobber each other's changes
• The built-in provider list is now generated from models.dev, broadening out-of-the-box provider coverage

• Queued messages now appear in a collapsible list above the composer with a count — expand it to edit, send-now, or delete individual queued turns.
• New sidebar update indicator: once an update has been downloaded, an accent-colored icon stays in the sidebar showing the new version with a one-click restart, so the update is still reachable after you dismiss the toast (and restart failures are now surfaced instead of silently doing nothing).
• No more appearance flash on launch — the app paints in your saved (or system) light/dark theme before the first frame.
• The header and sidebar now show the full workspace name and git branch, and lay out correctly on narrow windows; a transient git lookup no longer wipes a valid branch name back to "no git".
• Cleaner collapsed-sidebar settings layout: compact width, left-aligned navigation, stacked account details.
• Clarified the auto-update setting — it's now "Keep CLI up to date" and explains that it governs the cline terminal command, not the app itself (the app updates separately).
• Toggle switches now use a solid accent color when on, for clearer contrast.

• Major performance overhaul: the app now feels snappy end-to-end. The animated background renders at a locked 60fps instead of ~10fps, typing in the composer no longer stutters (245 slow keystrokes → 3), streaming responses coalesce updates instead of re-rendering the whole chat per token, and app boot fetches the provider catalog once instead of three times.
• The native folder picker and command execution no longer freeze the app while the sidecar writes session logs or discovers your editor.
• Fixed the composer getting stuck on "Agent is working..." after queued turns finished.
• Added a Cline API key path to onboarding, and you can now cancel a pending browser sign-in instead of being stuck waiting for it.
• Fixed window dragging.
• MCP server cards are now consistent across marketplace views, with a single uninstall action and setup guidance shown on installed servers.
• Fixed agentic compaction silently falling back to basic compaction for OpenAI-Compatible providers, and manual /compact never actually reaching the model when auto-compaction was off.

Full Changelog: https://github.com/cline/cline/compare/desktop-v0.0.4...desktop-v0.0.5

• Start chatting without opening a project folder — the app now supports workspace-free chat sessions.
• New first-run onboarding flow to get you set up on launch.
• Drag and drop files directly onto the chat to attach them.
• Image attachments now display inline in the chat transcript.
• Schedule one-time routines (not just recurring ones), with navigation to jump to a routine's run.
• New custom overlay title bar with in-app navigation.
• Redesigned channel setup as expandable cards.
• Added a setting to replay the new-user experience.
• Cleaner chat markdown rendering, and external links now open correctly in your browser.
• Agent sessions now use agentic compaction by default, keeping long conversations within context more intelligently.
• Fixed the agent not finding gh and other CLI tools by resolving your login shell's PATH.
• Headless routines now default to YOLO mode so they can run unattended.
• Fixed request metering for the SAP AI Core provider.

Full Changelog: https://github.com/cline/cline/compare/desktop-v0.0.3...desktop-v0.0.4