The Cline CLI (Preview) isn't just Cline in your terminal.

It's your scriptable coding agent and a force multiplier for VS Code and JetBrains Cline, allowing you to wield the IDE-level Cline as an orchestrator for Cline subagents.

Here's how you can get started 🧵 Start any task with `cline` or `cline "insert your prompt" to launch the Cline CLI and start in Plan mode.

You're not locked into any single LLM provider with Cline.

Model-agnostic design means you can route to:
- Direct providers
- Aggregators
- Local models

Here's how to strategically choose your routing: Direct Provider Connections:

Straight to Anthropic, OpenAI, Google, xAI.

Benefits:
- Latest model versions instantly
- No intermediary delays
- Provider-specific features
- Simplest communication path

Trade-off: Multiple accounts and API keys to manage.

LLMs need focus, because attention isn't enough.

(thread on the Focus Chain, Cline's link to persistent context) The greatest challenge in building coding agents is producing quality results over (1) long task horizons with (2) large amounts of context.

However, the most complex engineering challenges require both.

"How do you make money?"

"Why?"

We've been asked this question for months. It highlights our contrarian business model: aligning our incentives with our users to build the best coding agent.

(and ensuring long-term profitability)

🧵

https://twitter.com/262484988/status/1950972156873216121

The question reveals the trap every AI coding tool falls into.

They all make money the same way: reselling AI inference at a markup. Buy wholesale, sell retail. The gas station that also built your car.

Imagine if gas stations didn't tell you how many gallons you were getting. The station built your car, but you can't look under the hood.

That's the inference subscription dilemma in one image, and the reality of selling commodities on subscription models.

🧵

https://twitter.com/1353836358901501952/status/1949898502688903593

The pattern is predictable:
1. launch with "unlimited" access
2. power users take you at your word
3. economics hit hard
4. add hidden limits
5. users discover limits when work stops

Subscription models for AI inference don't work. The math always wins.

Here's how you can use the @huggingface provider in Cline 🤗

(thread) Follow the link to create your Access Token (create an account if you don't have one yet)

What we're seeing from coding agent companies:

They're abandoning RAG (via vector embeddings) for code exploration.

Why?

Code doesn't think in chunks, and it confuses the agent.

& codebase context that's gathered like a senior developer would leads to better outcomes. 🧵 When you embed code, you're literally tearing apart its logic. A function call in chunk 47, its definition in chunk 892, the context that explains why it exists scattered across a dozen fragments.

When your AI coding agent goes off track, your instinct is to course-correct. To explain what you really meant. To clarify, rephrase, add constraints.

But you're swimming upstream. Each correction adds more context pollution.

There's a better way -- using message checkpoints 🧵 Research from Microsoft and Salesforce just confirmed what we've been seeing: LLMs experience a 39% performance drop when you try to guide them through multiple conversation turns versus giving complete context upfront.

This fundamentally changes how we should think about AI interactions.

5 MCP Servers We Recommend

Our new guide to the MCP starter pack demonstrates how to integrate web search, live documentation, and browser automation directly into your Cline workflow. Equip Cline with the right context at the right time. 🧵 1. Supercharge your AI's knowledge with the Perplexity MCP. Give Cline access to the entire web for up-to-date research, so you never have to leave your editor to find answers.

Claude Max subscribers:

You can use your subscription in Cline instead of paying per-token API pricing.

Here's how you can set it up 🧵 Setup is simple:

1. Install Claude Code following Anthropic's guide
2. In Cline: Settings API Configuration > Select "Claude Code"
3. Set the path to your path to Claude Code CLI (can be just "claude")

This chart from Artificial Analysis is the only thing you need to see to understand the state of AI in 2025.

The "best" model is a constantly moving target. 🧵 This is why we're seeing so much developer tool fatigue.

If your AI coding tool is locked into a single vendor, you're guaranteed to be falling behind the state-of-the-art for months at a time.

We actually didn't want to build a 'plan mode' in Cline. It went against our core principle of simplicity. But then we saw how our power users were using Cline, and it became clear we had to.

Some behind-the-scenes on Plan mode & why it's a critical paradigm in AI coding 🧵 Internally, and with our earliest users, we noticed a pattern. As the AI got more capable, people would instinctively say "wait, don't code yet" or "let me see a plan first." They needed a brake pedal for an AI that was too eager to help.

"Intelligence too cheap to meter is well within grasp"

Here's why we've been building Cline for this exact future -- where inference abundance, not scarcity, defines how developers work.

🧵

https://twitter.com/1605/status/1932547247243505924

We made a deliberate choice:

While others architect around inference scarcity (caps, throttling, hidden operations), we built for abundance.

Why?

Just pushed Cline v3.17.12

This release includes free Grok 3 access for 2 weeks, collapsible MCP panels, better file context, and more reliable diff edits.

Full details below. 🧵 For the next two weeks, we're offering free Grok 3 access in partnership with xAI.

https://twitter.com/1875995486110593024/status/1932513639015329822

We turned a 50-question PDF on LLMs into a 10-episode lecture series, with Cline orchestrating the entire process.

Here’s a look at the workflow that made it possible, using @GoogleDeepMind 2.5 Pro to process the PDF and @elevenlabsio MCP to generate the lectures. 🧵 It all started with this great resource on LLM basics shared by @omarsar0.

https://twitter.com/3448284313/status/1930984834454712537

Just shipped Cline v3.17.9 👇

More Claude 4 optimizations, task timeline navigation, and CSV/XLSX support 🧵 We've been fine-tuning how Cline works with Claude 4, focusing on search/replace operations. The latest optimizations use improved delimiter handling that's showing great results in our testing.

LLMs have static knowledge cutoffs. They don't know about library updates, new APIs, or breaking changes that happened after training. Context7 by @upstash bridges this gap by injecting real-time docs into Cline. 🧵 The reality: open-source libraries continue evolving past the knowledge cutoff date of even the latest frontier models. If you're using a library that updated in the last year, the AI is working with outdated information.

quick guide to cline's slash commands:

/newtask = branch your work with preserved context
/newrule = teach Cline your coding preferences
/smol = compress context without losing progress
/workflow = automate repetitive tasks
/reportbug = submit issues with proper formatting

🧵 /newtask is like a perfect developer handoff.

When your context window fills up mid-project, it packages what matters -- your plan, progress, relevant files, and next steps -- into a fresh task.

Cline doesn't index your codebase. No RAG, no embeddings, no vector databases.

This isn't a limitation -- it's a deliberate design choice. As context windows increase, this approach enhances Cline's ability to understand your code.

Here's why.

🧵 The industry default: chunk your codebase, create embeddings, store in vector databases, retrieve "relevant" pieces.

But code doesn't work in chunks. A function call in chunk 47, its definition in chunk 892, the context that explains why? Scattered everywhere.

ICYMI: Here's what's new in Cline over the past couple weeks 🧵

(& a cline wallpaper if you're so inclined) First up -- Task Timeline in v3.15. Now you can see exactly what Cline is doing with a visual "storyboard" right in your task header. Every tool call, every file edit, all laid out chronologically. Hover for details.

🌎 Global Workflows & more improvements in Cline v3.17

Here's what's new 🧵 Global Workflows are here!

Your best workflows can now travel with you across all projects. Store them in ~/Documents/Cline/Workflows and they're available everywhere.

Local workflows still take precedence, so project-specific customizations remain intact.