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.

Anthropic's Claude 4 models are here.

Opus 4 and Sonnet 4 both show strong (& improved) coding abilities, with Sonnet 4 at 72.7% on SWE-bench and Opus 4 at 72.5%

What does this mean for developers using Cline? 🧵 Claude Opus 4 is engineered for complex, long-running agentic workflows. It's been trained to create and update 'memory files' (akin to context files in Cline), enhancing long-term task awareness. This power comes at a steep API cost ($15/$75 per 1M tokens).

The playbook for AI engineering is being written as we speak. Success isn't magic, it's method. For dev teams looking to build better software, faster with AI, here are 5 foundational pillars to consider. 🧵 Pillar 1: Understand What AI Engineering Really Is.
It's evolved beyond copy-pasting. True AI engineering is deep, IDE-native collaboration, blending human insight with AI capabilities. Think of AI as a brilliant, fast, but forgetful pair-programmer.

If you're an engineer who's feeling hesitant or overwhelmed by the innovation pace of AI coding, this thread is for you.

Here's the 10% of fundamentals that will put you in the 90th percentile of AI engineers.

🧵/many First, a crucial mindset shift: stop treating AI like a vending machine for code. Effective AI Engineering is IDE-native collaboration. It's a strategic partnership blending your insight with AI's capabilities.

Think of AI as a highly skilled (but forgetful) pair programmer.

Want to try Cline for free? @OpenRouter has free models that show a peek into the future of commodified inference.

Four models that are worth a try:

deepseek/deepseek-chat-v3-0324:free
meta-llama/llama-4-maverick:free
deepseek/deepseek-r1:free
qwen/qwen3-235b-a22b:free

🧵

https://twitter.com/1875995486110593024/status/1917640360274321885

Here's how you can get started using Cline with a free model from OpenRouter 👇

AI coding performance often dips when context windows exceed ~50% fullness, leading to errors or sluggishness.

Cline uses built-in context awareness + a customizable .clinerule to automatically trigger the new_task tool, keeping performance optimal.

Here's how it works: 🧵 As you work with Cline, the context window fills up -- with your prompts, Cline's responses, file contents, tool outputs, etc. Think of it like RAM. More context can be good, but too much can overwhelm.

Just pushed Cline v3.15! Task timeline, Gemini Implicit Caching, Open-source docs, & much more

Here's what we've got for ya ↓ We've created a Task Timeline -- see Cline's workflow as a visual "storyboard" in your task header. Understand tool calls, file edits & more at a glance. Hover for instant summaries. Clarity for complex tasks, and built by one of our community contributors (thanks eomcaleb!)

Here's how you can use the @firecrawl_dev MCP server to turn online docs into a functional .clinerule in minutes.

This makes Cline a pro in any library you're using. The core idea is simple:
1. Provide the URL of the documentation you want to process.
2. The @firecrawl_dev MCP scrapes and extracts the key information.

The upgraded Gemini 2.5 Pro (03-25 --> 05-06) has been available for over 24 hours now, and we're already seeing some exciting feedback from our users putting it to the test.

So, what's the word on how it's actually performing in Cline?

(hint: very well) ↓ One of the first things users are noticing is a smoother experience with file edits. If you've wrestled with "diff edit" errors with complex changes or large files before, the new version seems to navigate these much more reliably. We're hearing it's "actually doing well with larger files" now.

We nearly decided against creating Plan & Act modes, but now we see planning as a cornerstone feature of Cline.

Effective AI coding isn't prescriptive -- it demands a shared strategy transitioning from one-shot Hail Marys to context-riche, agent-driven success:🧵 Here's what we noticed: LLMs can be a bit antsy. They're often quick to generate code, sometimes before fully grasping the nuanced context of your project. This can lead to that frustrating cycle of near-misses. We saw a need for a more deliberate approach.

Cline for Research?

Combine MCP servers like @perplexity_ai and @firecrawl_dev with .clinerules to create specialized research workflows you can toggle on/off.

Here's how you can combine MCP Servers & .clinerules to make specialized workflows 🧵 ↓ Our rule guides the process:
1. Refine topic? (Yes/No)
2. Choose method? (AI Search/Deep Crawl/etc)
3. Pick output? (Chat/MD/JSON)
4. Cline executes via the chosen MCP

You just need to click the right buttons. cline-for-research.md

Cline v3.14 is live.

Here's what's new (mega 🧵) ↓ First up, significant improvements to Gemini caching & cost transparency:
- Refined caching logic for Gemini/Vertex for better reliability & cost savings.
- Added Cache UI for OpenRouter/Cline providers.
- Enabled pricing calculations for Gemini/Vertex.

Top 10 Most Popular MCP Servers in the Cline 1. GitHub: Manage repos, files, issues & PRs directly using Cline. Essential for integrating version control workflows seamlessly into your development process.

Wow. A Cline user has evolved their Cline Recursive Chain-of-Thought (CRCT) system for Cline with v7.7. This is like Memory Bank on steroids.

The advanced context management for large codebases just got even better. 🧵 The challenge with AI on big projects: context limits mean constantly re-explaining the codebase. CRCT v7.7 solves this not just with memory, but by building a deep, persistent *understanding* of your project's structure and dependencies.

Cline's 'Memory Bank' = a methodology for persistent context using custom instructions & docs, not a fixed feature. Here's how our power users are creating unique variations with visual prompts 🧵 AI assistants like Cline operate with context limits. The 'Memory Bank' methodology tackles this using structured documentation (simple markdown files in your project) that Cline reads via custom instructions or `.clinerules` to rebuild project understanding across sessions.

Plan Mode minimizes rework and enhances code quality by providing the right context before implementation. Here's why we believe planning is essential in any AI coding workflow: 🧵

https://x.com/nickbaumann_/status/1917404271110414507

We built Plan and Act as distinct modes recognizing that context-gathering deserves dedicated focus -- just like the top developers who always explore requirements before writing a single line of code.

Cline contributors and our dev team have been COOKING in the past few weeks. Here's a recap of our favorite features in case you missed any: 🧵 Fix with Cline code actions make error handling lightning fast. When you see a lightbulb, select 'Fix with Cline' to send code + errors directly to Cline.

Cline v3.13.3 is here with new ways to manage context and costs:

- /smol slash command to compress long conversations 🤏
- Gemini 2.5 Pro prompt caching (Cline, OpenRouter)
- MCP server download counts
- UI tooltips for easier navigation

Here's the breakdown ↓ Running into context limits or high costs in a long Cline task? Use the new /smol command (or /compact) to compress the conversation history *within the current task*.

It summarizes the chat, reducing token usage for subsequent turns so you can keep working.

LLMs often use old training data. Add `use context7` in Cline prompts to inject current docs via the Context7 MCP.

But how does Cline actually use this MCP? It's not just magic. 🧵 When you add `use context7`, Cline leverages tools provided by the Context7 server @upstash in a two-step process to ensure accuracy:

Just pushed Cline v3.13.2 -- this patch release improves the Gemini provider performance with caching, adds thinking budget support, and includes the Gemini 2.5 Flash model.

Here's the breakdown: 🧵 Some improvements:
- Support for `!include .file` directive in `.clineignore`
- Improved slash command functionality
- Better prompting for the new task tool (better summaries of your context!)