Cline — the 67,000-star open-source coding agent — made two moves that lower the barrier to entry for AI-assisted coding. First, it shipped a standalone macOS desktop app that runs agent sessions independently of any IDE. Second, it added free coding models to every Cline surface: the desktop app, the VS Code extension, and the command-line interface. You do not need an API key to try Cline.
The desktop app launched as v0.0.4 on July 22, 2026 and has been updating almost every other day since. As of September 11 it is at v0.0.26 on both macOS and Windows, with the v0.0.24→v0.0.26 wave fixing Windows updates that silently failed, lost prompts on failed sends, and a session state that lied about whether the agent was still working — on top of the universal Mac download, token usage meter, and edit-any-message from August.
If you have been curious about coding agents but hesitated because of billing setup or vendor lock-in, this is the week to try one.
What Is Cline Desktop?
Cline started as a VS Code extension called “Claude Dev.” It grew into a CLI, a JetBrains plugin, an SDK, and now a desktop app. The VS Code extension still works — Cline Desktop does not replace it. Instead, it gives you a native macOS application for running and monitoring agent sessions outside of your editor.
For beginners: think of Cline Desktop as a control center for your coding agent. You can start a session, watch the agent work, review its changes, and manage multiple parallel sessions — all without VS Code open.
What Changed from v0.0.7 to v0.0.10
Three releases in eight days (v0.0.8 on August 2, v0.0.9 on August 5, v0.0.10 on August 7) turned the rough first week into a smoother experience. The highlights, in plain language:
One download for every Mac. Previously you had to pick between an Apple Silicon and an Intel build — a confusing choice if you are not sure which chip your Mac has. v0.0.9 ships a single universal DMG that runs on both, and existing installs migrate automatically on their next update.
A token usage ring in the composer. As you chat, a circular meter shows how much of the model’s context window you have used, alongside the cumulative cost. It changes color as you approach the limit. This is a quiet game-changer for cost anxiety: instead of discovering you hit a context ceiling mid-task, you see it coming and can start a fresh session on purpose.
Edit any earlier message. In v0.0.8, you can edit any message in the conversation, not just the latest one. The app forks the session at that point, rewinds the workspace to the checkpoint, and re-runs from your edited prompt. A failed restore is transactional — it will not leave your files half-rewound. For practical use: if the agent went down the wrong path two turns ago, you can correct course without killing the session and re-explaining everything.
Errors are no longer silent. Until v0.0.10, a turn could fail with barely any explanation. Now, sending a message with no model credentials — or losing a queued turn — shows an error right in the transcript, explains the underlying cause, and points you to Settings → Models. The welcome screen also shows a “Connect a model” button when no credentials exist, and it updates live as you add them.
Other useful fixes. MCP servers can now authenticate with OAuth straight from Settings → MCP (v0.0.10), and MCP errors appear on the individual server instead of as a page-level mystery. In v0.0.9, session history can be filtered by origin (desktop, CLI, extension, or scheduled), skills appear in the slash-command menu, and the agent header is draggable again — a long chat message no longer traps the window.
Free Coding Models: No API Key, No Billing
The headline feature across all Cline surfaces is free models. When you open the model picker in Cline Desktop, the VS Code extension, or the CLI, you now see certain models marked with a “(free)” label. These are models that Cline provides without requiring an API key or any payment.
You select a free model, ask the agent to do something — read a file, explain a function, write a test — and it works. The agent loop is identical to what you get with paid models: it reads your codebase, plans its approach, executes tool calls, and shows you the results.
The free tier has usage limits. When you hit one, Cline shows a dedicated error card with two pieces of information: you have reached the free-tier limit, and the exact time the limit resets. This is a much better experience than the vague “rate limit exceeded” messages that some other tools display. You know when you can try again, and you do not need to guess whether the problem is your setup or a temporary restriction.
For beginners, this means you can try Cline without:
- Setting up an Anthropic, OpenAI, or OpenRouter account
- Entering a credit card
- Configuring billing alerts
- Worrying about surprise costs
Pick a free model, ask the agent to do something, and see how it works. If you later want faster inference or more capable models like Claude Sonnet or GPT-5, you can switch to a paid provider at any time — and the token ring will show you exactly what each turn costs before you commit.
What the Desktop App Can Do
Cline Desktop ships with several features that the VS Code extension does not offer:
System tray with session status. A system tray icon shows how many agent sessions are currently running. You get at-a-glance visibility without switching windows.
Session history with pagination. History is paginated in ten-session pages, with older sessions loaded on demand. You can favorite sessions, and they are sorted by most recent activity.
Subagent and teammate visualization. When a Cline session spawns subagents — specialized assistants for tasks like testing or code review — their runs appear in the desktop app with status and results. You can see what each subagent did without digging through logs.
Workspace-free chat. No project folder required for quick questions. Open the app, type a question, and get an answer.
File drag-and-drop. Drag files directly onto the chat to attach them; images display inline in the transcript.
Queued messages. While the agent is busy, you can queue additional messages, edit them, send them immediately, or delete them.
One-time routines. In addition to recurring schedules, you can schedule one-time routines that run once at a specific time.
Agentic Compaction as the Default
One change that affects all Cline users: agentic compaction is now the default context-compaction strategy. When a long conversation exceeds the model’s context window, Cline summarizes earlier messages to make room. The old “basic” strategy truncated at fixed points, often losing important context. Agentic compaction uses the model itself to generate a structured summary that preserves decisions, file paths, and task state.
This matters because long coding sessions — the kind where you iterate on a feature over an hour or more — previously lost coherence after compaction. With agentic compaction, the agent remembers what it did, what it decided, and what remains to be done. Combined with the new token ring, you get much fewer “surprise amnesia” moments.
September 2026 Update: What v0.0.11–v0.0.24 Fixed
Cline Desktop has shipped fourteen releases since the August 7 snapshot above. The biggest theme is trust — fixing the bugs that made users uncertain whether the app was working correctly. Here are the changes that matter most, in plain language:
Your chat no longer lies to you
In v0.0.24, the live chat stream used to show duplicated text. The sidecar architecture had two internal listeners receiving the same session events, and when a session was already running when you opened it, every message token appeared twice. The fix asks the observer listener directly whether it should render, instead of guessing from a timer. The result: what you see in the transcript is exactly what the agent said, nothing more.
Queued messages actually arrive
When you typed a follow-up message while the agent was still busy, the app queued it — but the queue could silently erase your message bubble. A race condition between the queue drain and the transcript refresh meant your text disappeared and the reply appeared with no user message attached. v0.0.24 fixes this by deferring the transcript replacement to the newer turn.
The agent asks before giving up
Before v0.0.24, when a model got stuck in a loop (five identical tool calls or six consecutive failures), the run ended silently and the composer went idle. You had no idea the agent had given up. Now you get a prompt: “Try a different approach” or “Stop this run.” The guidance is injected into the running turn so the model knows why it paused, instead of repeating the same broken call.
Editor errors finally say what went wrong
The editor tool now names the file, states whether old_text was null or omitted, and tells you how to recover. Previously, models like Kimi-k3 that fill optional parameters with null got a terse “old_text is required” and re-sent the same broken call until the loop detector killed the run.
Import sessions from Claude Code, Codex, and opencode
This is the migration feature for people who have been using other coding agents. Sessions imported from Claude Code, Codex, or opencode now show up in the chat with a clear source label. On the first resumed turn, the foreign history is summarized so the model understands the context. The original transcript stays intact — the summary is additive, not destructive.
Checkpoints no longer freeze large projects
Checkpoint creation previously rebuilt a throwaway git index on every message. If you had multiple gigabytes of untracked files — common with cloud-synced projects — each message could take 90 seconds or more. v0.0.24 keeps one snapshot index per session. From the second turn onward, checkpoint creation is near-instant, and the saved snapshots are byte-identical to before.
Background commands no longer hang
Commands that background a child process (cmd &, nohup, Git Bash equivalents) used to hang until the timeout because the inherited stdio pipes stayed open after the shell exited. Now they settle with the real exit code and a note that background output continues in the terminal.
Windows gets a proper title bar
v0.0.24 adds a custom title bar on Windows with caption controls that follow the compact title-bar height in narrow windows. The taskbar icon was updated as well. Small quality-of-life change, but it makes the app feel native instead of like an Electron port.
Token counts for every session
The sessions view previously only hydrated the four most recent rows with token counts and costs — every other row showed “-”. Now every visible session is hydrated on demand, with reads capped and re-run when status changes. You can see the actual cost of a session from three days ago without opening it.
Agent Plugins from your filesystem
In v0.0.23, packages under ~/.agents/plugins are discovered and run by the shared Hub. Each plugin’s plugin.json is validated, its skills become available to the agent, and its MCP servers start automatically. This is separate from Cline Plugins — Agent Plugins are a community-level extension point that works across Cline surfaces.
Hub update dialog no longer loops
A build-mismatch dialog used to appear on every launch if you had both the desktop app and CLI installed from different commits. It offered “Update and restart” which led to “no app update available” — an infinite loop. Now the dialog only appears when an actual update is staged, and “Later” sticks across session switches.
More fixes worth noting
- Sign-in confirmation code now appears in the app while you wait for the browser, so you can match it against what the browser shows you.
- Voice input failures from provider setup issues now take you straight to voice settings instead of showing a toast you cannot act on.
- Session history with many children no longer renders empty — child rows sort after their root, so a busy session cannot hide itself and every older session from the sidebar.
- Cline Pass model selection is preserved when starting a new chat — the catalog is treated as discovery data, not validation, so models missing from a partial catalog are not silently swapped.
- One wedged MCP server no longer blocks all others from shutting down.
Is It Worth Updating?
If you tried Cline Desktop in August and bounced off bugs — duplicated text, messages disappearing, the app silently going idle — the v0.0.24 release is the reason to come back. These are not cosmetic tweaks. They fix the interactions that determined whether you trusted the tool.
The Latest Wave: What v0.0.25 and v0.0.26 Do (September 10–11)
Less than 48 hours after v0.0.24, Cline shipped v0.0.25 and v0.0.26. The headline feature — a live PR card in the composer — is the kind of thing that makes a coding agent feel like a teammate instead of a terminal. Here’s the wave in plain language.
Your pull request is now one glance away
In v0.0.26, the composer shows your current branch’s GitHub PR — number, merge status, changed-line totals, and CI checks. It refreshes every 30 seconds while visible. If there’s no PR yet, a Create PR button opens GitHub’s comparison form. Click anywhere on the card to inspect individual CI checks right in the app.
One honest caveat: Cline does the watching, not the doing. It needs GitHub CLI (gh) installed and signed in, and it never pushes commits or submits the PR for you. It’s a dashboard, not a replacement for your review workflow — but for anyone who has alt-tabbed between a terminal and GitHub five times a day, it kills a whole class of context switching.
Windows updates finally work
The dreaded “Error opening file for writing” on Windows is gone. The root cause was a classic: the compiled sidecar re-executed itself as a detached background daemon that outlives the app, the NSIS installer only killed the main binary, and the daemon silently held code-sidecar.exe so the update stalled until you killed it by hand. The installer now stops it first — matched on the full path, so updating one channel won’t kill a side-by-side Cline Beta’s sessions.
Your prompt is no longer eaten by a failed send
If you hit send and the turn failed before starting — say, an OAuth refresh threw — the old behavior wiped your typed message and you retyped everything. Now the text and attachments return to the composer, merged with anything you added while the send was pending. If you’d already started typing something else, it leaves your new draft alone. Small fix, huge sanity saver.
“Please log in” now tells you where
Providers that log in through a local CLI — Claude Code, Codex CLI, OpenCode — used to show a confusing browser sign-in button that could not do anything, or dumped you at a generic settings page. Now:
- Claude Code, Codex CLI, and OpenCode sessions can start without an API key, using the credentials their own CLIs store — OpenCode is correctly treated as a local-CLI provider instead of an OAuth one.
- Auth failures (“OAuth session expired”, “Not logged in · Please run /login”) now point you at the actual CLI to fix, not Settings → Models.
- Signing out of ChatGPT (Codex) actually sticks — the legacy stored credentials used to silently re-import you on the next action. Now they’re cleared too.
That last one was a security-adjacent paper cut: you clicked “sign out” and Cline signed you back in behind your back.
ChatGPT (Codex) model list is now honest
The shared OpenAI catalog was leaking models you couldn’t actually use into the picker — retired GPT-5.4 models, GPT-4o/4.1, chatgpt-image-latest — and the runtime was inheriting 1.05M token limits the backend doesn’t give ChatGPT accounts. Now ChatGPT subscribers see only what their plan can use, with the real budgets (400K/272K/128K). The default moves to gpt-5.6-terra.
Model defaults changed for 36 providers — check what you’re pinned to
v0.0.25 refreshed the model catalog: NaN joined, and the resolved default model changed for 36 providers including Bedrock, Vertex, OpenRouter, Kilo, GitHub Copilot, Gemini, Cerebras, Fireworks, Requesty, and Vercel AI Gateway. Several moved off Claude Fable 5.1 to GPT-6 Astra, Vertex goes to Gemini 3.8 Flash, OpenRouter/Kilo to Inception Mercury 2.5. If you use one of these without pinning a model, your next session may silently start on a different model — worth a glance at Settings → Models.
Free models now show zero cost — and other fixes
- Cline Pass and free models report $0 cost instead of the upstream provider’s price.
- A forked session no longer comes back stuck on “Thinking…” — restored sessions were being persisted as “running” when they were idle, so resuming showed a session that was never going to finish.
- Deleting a queued prompt no longer leaves a ghost message in the transcript.
- Nested PowerShell commands no longer flood errors —
$_was being interpolated away by the outer parser before the inner shell saw it, soWhere-Object { $_.Name ... }errored per item over large trees while still exiting 0. - Scheduled sessions no longer stall out — a poller bug could silently burn through retries; recurring schedules now also default to your local timezone instead of UTC.
- Invalid image formats (HEIC, TIFF, SVG, BMP, ICO) are rejected at attach time instead of failing mid-turn.
- Prompt telemetry to Cline’s tracing backend is limited to Cline/Cline Pass — turns on your own provider keys are no longer traced, a real privacy win.
Is the new wave worth it?
If you’re on Windows and updates have been failing, this is the update that fixes updates. If you run agents against Claude Code/Codex/OpenCode, the local-CLI auth and honest auth errors remove the most confusing first-run friction. And if you live in GitHub PRs, the composer card is a genuinely new reason to open Cline instead of your browser.
If you are a Claude Code user considering switching, the session import feature means you do not lose your existing work. Bring a session over, and the agent picks up where the last one left off.
If you are brand new to coding agents, none of these fixes matter to you yet — you will just get a working app. Start with a free model, ask it to explain a function in your codebase, and see how the agent loop works.
How to Get Started
Desktop app: Download the installer from the Cline GitHub Releases page. macOS users get a single universal build (no architecture choice needed); Windows users get a native app with a custom title bar. Open the installer, drag the app to Applications (Mac) or run the setup (Windows), and future updates arrive automatically.
VS Code extension: Open the Extensions panel in VS Code, search for “Cline,” and install or update. Free models appear in the model picker after updating.
CLI: Install with your package manager. It shares the same agent core as the desktop app and extension, so free models, agentic compaction, and everything else work identically.
In all three cases, open the model picker, look for “(free),” and start coding. Nothing else to configure.
Should You Care?
If you are evaluating coding agents, Cline’s free tier removes the biggest friction point: cost. You can compare Cline’s agent loop — how it plans, executes, and recovers from errors — against Claude Code, Codex, Goose, or Kilo without spending a dollar.
If you are already a Cline user, the v0.0.8–v0.0.10 releases are worth updating for: the token ring alone changes how sessions feel when you approach a context limit, and edit-any-message saves you from throwaway sessions.
If you have never tried a coding agent at all, Cline’s free models are the lowest-risk start. The free models will not match the speed or capability of Claude Opus 5 on complex multi-file refactors. But for reading code, answering questions, writing tests, and small edits, they are more than enough — and the price is hard to beat.