Originally published at terminalblog.com
Related articles
- Cline Now Has a Desktop App With Workspace-Free Chat and Claude Opus 5 Support
- Best Coding Agents 2026 — Decision Guide
- Claude Code Now Runs Code Reviews in the Background and Gets Opus 5
Cline now lets you use coding models for free. Desktop v0.0.7, released July 29, adds a "free" tier to the model picker — models labeled "(free)" cost nothing, and when you hit the usage limit the app shows a clear error card with the exact time the limit resets. The same feature landed in the SDK (v0.0.66) and the CLI extension (v4.0.12) on July 28, so every Cline surface now supports free models.
If you have been holding off on coding agents because API credits cost money, this is the release that removes that barrier.
Free Cline models: how they work
When you open the model picker in Cline Desktop, the extension, or the CLI, you now see certain models marked with a "(free)" label. These are Cline-provided models that do not require an API key or any payment. You select one, start coding, and the agent works exactly as it would with a paid model.
The free tier has usage limits. When you reach one, instead of crashing or producing a confusing error, Cline shows a dedicated error card that tells you two things: that you hit the free-tier limit, and when the limit resets so you can try again. This is a much better experience than the vague "rate limit exceeded" messages that some other tools display.
For beginners, this means you can try Cline without setting up any API keys, configuring billing, or worrying about surprise costs. Pick a free model, ask the agent to do something, and see how it works. If you later decide you want faster inference or access to more capable models, you can switch to a paid provider at any time.
The free models are available across all three Cline surfaces: the macOS desktop app, the VS Code extension, and the command-line interface.
System tray icon and session management
Desktop v0.0.7 adds a system tray icon that shows the app's current status and how many agent sessions are running. On macOS, this means you can glance at your menu bar and see whether Cline is active without switching to the app window. If you run multiple agent sessions in parallel — one refactoring a module while another writes tests — the tray icon counts them all.
Session history is now paginated. Instead of loading every session you have ever had at once, the app shows ten sessions at a time and loads older ones as you scroll to the bottom. This sounds like a minor performance fix, but it matters when you have been using Cline for weeks and have hundreds of sessions. Before this change, scrolling through long lists could lag. Now it stays responsive.
You can also mark sessions as favorites. Favorite sessions appear at the top of the list regardless of when they were last active, so your most important conversations are always one click away. Sessions are also now ordered by most recent activity rather than creation time, which matches how most people think about their work — the session you were just using should be at the top.
Subagent and teammate run visibility
If you use Cline's multi-agent features — where the main agent delegates tasks to subagents or teammates — you can now see those runs directly in the desktop app. Each subagent or teammate run shows its status (running, completed, failed) and its results. Before this change, subagent work happened invisibly in the background and you had no way to monitor progress without checking logs.
This is particularly useful when the main agent spawns a subagent to handle a large task like "run the full test suite and fix any failures." Instead of waiting and wondering, you can see the subagent's progress in real time.
Chat interface improvements
The desktop app's chat interface received several polish updates:
Tool-specific icons. Each tool call now shows an icon matching the tool type — file edits, terminal commands, and web searches each get a distinct visual. This makes it easier to scan a long conversation and find the moment where the agent made a specific change.
Elapsed thinking time. When the agent is reasoning about a task, the thinking section now shows how long it took. This helps you understand whether the agent is genuinely working on a complex problem or stuck in a loop.
Restyled reasoning sections. The visual presentation of the agent's reasoning has been cleaned up, with better spacing and clearer distinction between thinking and output.
Stable message layout. Message actions (like copy, retry, or edit) no longer shift the conversation layout while you are scrolling. The viewport stays anchored to where you are reading.
These changes add up. The desktop app went from feeling like an early-stage tool to something you could use as your primary coding interface.
Agentic compaction becomes the default
The Cline SDK (v0.0.66) makes agentic compaction the default context-compaction strategy. Context compaction is how Cline manages the conversation when it gets too long for the model's context window. Before this change, the default was basic compaction — a simple truncation that could lose important context. Agentic compaction is smarter: it uses a language model to summarize the conversation while preserving the key decisions, file references, and constraints that matter for ongoing work.
Two bugs were fixed in the compaction system:
OpenAI Compatible provider fallback. Agentic compaction was silently falling back to basic compaction on OpenAI-compatible providers because the summarizer tried to call api.openai.com instead of the configured provider endpoint. This is fixed — the summarizer now uses the correct base URL.
Tool-heavy transcript cut points. Agentic compaction previously could not find a good cut point in transcripts with many tool calls, causing it to skip compaction entirely and let context grow until it hit the model's limit. Now the compaction logic treats assistant messages as valid cut boundaries, so it can always find a place to compact.
Both fixes prevent situations where context grows uncontrolled and the agent starts producing worse outputs because it is working with a truncated or confused view of the conversation.
Provider catalog refresh
The built-in provider and model list is now generated from models.dev, a community-maintained directory of AI model providers. This means Cline's out-of-the-box provider coverage automatically includes new providers and models without waiting for a Cline release. If a new provider appears in the directory, it shows up in your model picker on the next update.
The SDK v0.0.66 also adds support for SAP AI Core as a provider and fixes a file-editing issue where CRLF line endings were being corrupted to mixed endings, which broke exact-match edit operations on Windows-originated files.
Bug fixes for regional users
Desktop v0.0.7 fixes an issue where the China/international endpoint toggles for Qwen, Moonshot, Z AI, and MiniMax were being silently ignored. Users in China or outside China were being routed to the wrong API endpoint, causing connection failures or slow responses. The toggles now work as expected, routing to the correct regional endpoint.
A separate fix addresses tool calls that failed when a model returned a line number as a string (for example, "insert_line": "3" instead of "insert_line": 3). The agent would fail, waste a round trip retrying, and sometimes produce an incorrect edit. This is a common case with smaller models that do not always produce perfectly typed responses.
Connector session persistence
Connector sessions — the persistent connections between Cline and external services like MCP servers — now survive daemon or hub restarts. Before this fix, restarting the Cline daemon or the hub process would disconnect all connectors and require manual reconnection. Now sessions persist across restarts and reconnect automatically.
This is especially important for teams using Cline in CI/CD pipelines or automated workflows where the daemon might restart between runs.
How to update
If you are using Cline Desktop, the app checks for updates on launch and every two hours in the background. It will download v0.0.7 and prompt you to restart. If you are using the VS Code extension, run the update from the Extensions panel or wait for the auto-update. For the CLI, update with your package manager.
The free models are available immediately after updating — no configuration changes needed. Open the model picker and look for models labeled "(free)."
Should you switch to the free tier?
If you are evaluating coding agents and want to try Cline without commitment, the free models are the obvious starting point. They give you a no-cost way to understand how Cline works, what its agent loop feels like, and whether the desktop app fits your workflow.
If you are already a paying user, the free tier is useful as a fallback. When you hit rate limits on your paid provider, switching to a free model keeps you working instead of waiting.
The free models are unlikely to match the speed or capability of premium options like Claude Opus 5 or GPT-4o for complex multi-file refactors. But for reading code, answering questions, writing tests, and making small edits, they are more than enough — and the price is hard to beat.
Top comments (0)