cline

• A new Changes button in the session header shows the session's line counts and opens a pane listing the files changed in the last turn or across the whole session, with a diff of each. When a turn finishes, a summary under it lists the files it changed; click it to open the pane on that turn
• A project files panel (folder button in the session header) shows a file tree beside the conversation, with a read-only tabbed viewer, git status letters, and a marker on files the session changed. Files the agent edited get a File/Diff toggle. Works on local and SSH environments
• An integrated terminal panel opens at the bottom of the window (toggle with Ctrl+backtick)
• Pull request status moved from the bar above the composer into a header button. Its menu shows the PR title, merge status and checks (failing checks first), and lets you open the PR on GitHub, create a PR, copy the branch name, or refresh
• A new Local Models settings page manages Ollama and LM Studio models: setup guidance, model details and storage, and a composer chip showing how much memory the loaded model uses
• Keybindings are configurable in Settings (also reachable from the app and Help menus). Quit can require holding the shortcut, 1 second by default on macOS, so it's harder to quit by accident
• Slash commands and @file mentions show as chips in the composer and in sent messages
• If the Hub fails to start, the app now shows a startup screen with Retry and Export diagnostics, instead of failing every action with "desktop backend endpoint not ready"
• General settings shows the Hub's status and has a Restart Hub button
• Opening a task that's already running in another Cline app now attaches to that run instead of starting a second one that couldn't be stopped. Requests that were waiting on an app that closed mid-task now fail with a clear message instead of hanging
• Sessions with a very large transcript (over about 8 MB) can be resumed again. Before, the transcript was sent to the Hub twice, which went over its message size limit
• Long lines in the file and diff viewers wrap, and rows always span the full width when scrolled
• New Cloud chats start on your local Cline or ClinePass model. If Cloud doesn't offer that model, ClinePass subscribers get a ClinePass model and everyone else gets Claude Sonnet 5.5
• A pasted Cline API key now replaces an old browser sign-in instead of being ignored. Provider settings fields stack properly in the narrow side panel
• The Sessions/Projects sort choice in the sidebar is remembered across restarts
• On Linux desktops without a tray (stock GNOME on Fedora or Debian), closing the window quits the app instead of leaving it running with no way to quit
• Internal messages meant only for the model no longer show in the transcript, and live session titles no longer show the raw user_input wrapper
• Schedule status stays current and failed runs show the full error
• Checking for changes no longer spawns thousands of git processes in workspaces with many untracked files
• SSH environments behind a ProxyJump alias now connect
• The update toast's Restart button sits under the message text
• A turn whose connection drops after the model has streamed only reasoning is now retried instead of ending the run
• Model requests retry at most 3 times instead of 5
• Kimi K3 now gets max reasoning for Extra High and Max, instead of an xhigh setting it rejects
• Gemini API keys (the gemini provider) no longer send native Google Search. Free-tier keys failed every turn with a 429 quota error on Gemini 3+, and on Gemini 2.x every Cline tool was dropped
• Auto-compaction no longer runs for models with an unknown context limit, and it no longer runs again every few tool calls after the first compaction
• ask_question: the model keeps going after you pick an option, empty answers are rejected, and an answer with only an attachment counts
• run_commands with an empty command list now returns an error instead o

• Opening a task that's already running in another Cline app now attaches to that run instead of starting a second one that couldn't be stopped. If an app closes mid-task, requests waiting on it now fail with a clear message instead of hanging
• cline hub status and cline doctor no longer hang on a Hub that has frozen. Its health check now times out after 3s and the Hub is reported as not running
• A turn whose connection drops after the model has streamed only reasoning (for example, "The socket connection was closed unexpectedly") is now retried instead of ending the run
• Model requests retry at most 3 times instead of 5
• Kimi K3 now gets max reasoning for Extra High and Max, instead of an xhigh setting it rejects
• Gemini API keys (the gemini provider) no longer send native Google Search. Free-tier keys failed every turn with a 429 quota error on Gemini 3+, and on Gemini 2.x every Cline tool was dropped
• Auto-compaction no longer runs for models with an unknown context limit, and it no longer runs again every few tool calls after the first compaction
• ask_question: the model keeps going after you pick an option, empty answers are rejected, and an answer with only an attachment counts
• run_commands with an empty command list now returns an error instead of a silent success. On PowerShell, the tool tells the model how quote escaping works and to confirm recursive deletes first
• Plugin slash commands run in the active session
• Compiled CLI binaries load plugins and serve the dashboard from resources embedded in the binary
• ACP: the provider and account options now appear under the right config categories
• The skills tool call label no longer shows null arguments
• Refreshed the model catalog. Adds the Liquid Inference and Munito AI providers and drops CrofAI, iFlow and Llama. The Cline free list adds Reflection Beam, Nemotron 3.5 Lightning and Laguna S 2.1 and drops Mimo V2.6 Flash and Muse Spark 1.3 Contributor. ClinePass now defaults to Kimi K3.1 Pro and adds MiniMax M3.1 Flash and DeepSeek V4.1 Flash 1017. Default models change for 56 other providers, mostly to Kimi K3.1 Pro, MiniMax M3.1 Flash or DeepSeek V4.1 Flash 1017

Full Changelog: https://github.com/cline/cline/compare/cli-v3.0.73...cli-v3.0.74

• Starting a session that already has a run in progress now reattaches to the live session instead of replacing its runtime. Before, the original run kept going but couldn't be reached, and its transcript writes landed on a stale snapshot. A start that includes a prompt while a run is in progress is rejected
• The hub records every client that registered a capability for a session, including clients that join a task that's already running, and sends each request to the first one still connected. If none is left, the request fails quickly with a message the user can act on, instead of waiting forever
• A turn whose connection drops after the model has streamed only reasoning is now retried, and the partial reasoning is discarded. Turns with text, media or tool calls are still never retried
• Model requests retry at most 3 times after the first attempt (down from 5). Failed attempts are recorded in Langfuse traces, and cancelled requests are no longer reported as Langfuse errors
• AI SDK updated to ai 7.0.136 / @ai-sdk/provider-utils 5.0.58. max reasoning now reaches the provider: the gateway no longer turns it into xhigh, and Kimi K3 gets max for Extra High and Max. Empty Anthropic turns are still retried despite the new anthropic.* custom stream parts
• The gemini provider no longer sends native Google Search. On Gemini 3+, free-tier keys failed every turn with a 429 quota error, and on Gemini 2.x it dropped every Cline tool. Vertex keeps native search
• Auto-compaction no longer runs when the model's input limit is unknown. Manual and overflow-recovery compaction still work
• After the first compaction, the size estimate now uses the messages as they are actually sent, with tool results truncated. Before, it overestimated about 7x and compaction ran again every few tool calls
• Checkpoint diffs read file contents with a single git cat-file --batch process instead of up to two git show processes per file. Workspaces with many untracked files used to spawn thousands of git processes
• SSH remotes work behind a ProxyJump alias. The connect budget goes from 10s to 30s, the per-command limit to 60s, and tunnel readiness to 40s. When ssh exits mid-upload, its real error is reported instead of EPIPE, and a LocalForward on the alias no longer collides with the tunnel
• ask_question answers are framed so the model keeps going after a bare choice. Empty answers are rejected, and an answer with only an attachment counts as answered
• run_commands rejects an empty commands list with a tool error, and its schema now advertises minItems: 1
• On PowerShell, the run_commands description explains backtick escaping, single-quoting literal paths, preferring cmdlets over cmd /c, and confirming recursive deletes first
• Plugin commands run in the active session runtime and the hub. Plugin watches start working again when discovery directories appear
• Compiled binaries load embedded plugin and dashboard resources straight from the binary, and every declared SDK runtime export is packaged
• A configured CLINE_PLUGIN_RUNTIME_DIR is resolved to an absolute path
• When OpenRouter has no match for a Cline recommended model, its capabilities are filled in from the Vercel catalog
• Desktop cloud session orchestration moved into the hub. session.list reports activeCloudCommands, and the hub isn't replaced while cloud commands are still running
• Core adds conversation summaries and cloud continuation support
• Run stall telemetry: agent.run-finished and agent.run-failed carry stall signals from a new RunHealthObserver
• @cline/ui now ships with the SDK packages on the shared version. It adds a Kbd component and token chips for slash commands and file mentions in AgentComposer
• Refreshed the model catalog (7,041 → 6,457 models). Adds the Liquid Inference, Munito AI and Munito AI (EU) providers and drops CrofAI, iFlow and Llama. The Cline free list adds Reflection Beam, Nemotron 3.5 Lightning and Laguna

• OpenAI-compatible requests for GPT-6 models (gpt-6-sol, gpt-6.1-sol, gpt-6-1-sol, and later gpt-5 through gpt-9 ids) send max_completion_tokens instead of max_tokens, so OpenAI and Azure no longer reject them. Azure's gpt-35-turbo keeps max_tokens
• The hub daemon records which client started it (CLINE_HUB_LAUNCHER_CLIENT, CLINE_HUB_LAUNCHER_VERSION, CLINE_HUB_LAUNCHER_TRANSPORT) and publishes that as OTel resource attributes. SSH remote hubs report the local client with transport ssh. @cline/shared adds withHubLauncherIdentity, claimHubLauncherIdentity and readHubLauncherIdentity
• The telemetry self-check report includes traces_exporter
• Refreshed the model catalog (6,952 → 7,041 models). Adds the CommandCode and Reflection providers. The Cline recommended list leads with Claude Sonnet 5.5 again, so the cline provider's bundled default goes back to Claude Sonnet 5.5 from Gemini 4 Argon. Default models change for 302.AI (Gemini 4 Argon), Eden AI and Prime Inference (Muse Spark Open), CoreWeave (Reflection Beam 1.0), NanoGPT, Pioneer (Nemotron 3.5 Super VL), and Scaleway (DeepSeek V4.1 Flash). OpenCode Go keeps its https://opencode.ai/inference/go/v1 base URL, now pinned, because models.dev moved its provider-level URL to an OpenAI-only path

Full Changelog: https://github.com/cline/cline/compare/sdk/sdk/v0.0.94...sdk/sdk/v0.0.95