- unwind ai
- Posts
- Free AI Coding Agent in Your IDE
Free AI Coding Agent in Your IDE
PLUS: Enterprise-grade RAG on-prem, DeepSeek's opensource O1-level model
Shubham Saboo & Gargi Gupta
January 22, 2025
In partnership with
Today’s top AI Highlights:
Enterprise-grade RAG (ingest, store, search and eval) in a few lines of code
ByteDance releases AI code editor with free access to GPT-4o and Claude 3.5 Sonnet
DeepSeek R1 delivers OpenAI O1 performance for free
Bring real-time search to your apps with Perplexity's new API
AI agent to test all your pull requests
& so much more!
Read time: 3 mins
AI Tutorials
Game development demands handling a daunting array of specialized skills - a compelling narrative and storylines, intricate mechanics, visual aesthetics, technical architecture, and more. It’s a struggle synchronizing these - scope creep, misaligned creative visions, and technical bottlenecks.
In this tutorial, we'll build an AI Game Design Agent Team that coordinates multiple specialized AI agents - each focusing on their domain expertise - to generate cohesive game concepts where narrative, gameplay, visuals, and technical specifications work in harmony.
The entire process is automated so developers can quickly iterate on ideas and ensure all crucial aspects of game design are considered.
We share hands-on tutorials like this 2-3 times a week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.
Latest Developments
AI dev startup Eyelevel.ai is tackling a major pain point for developers working with RAG: inconsistent results. They've released GroundX, a platform focusing on accuracy and security, addressing the "garbage in, garbage out" issue that plagues many RAG setups.
GroundX processes complex enterprise documents like forms, diagrams, and tables with 50% higher accuracy compared to other RAG frameworks, using a vision model trained on over a million pages. It focuses on the quality of ingestion as much as on the retrieval. GroundX has both air-gapped deployment and a modular architecture that can run fully offline in its own namespace.
Key Highlights:
Document Processing - Their fine-tuned vision model understands and processes tables, forms, and diagrams that often trip up LLMs, converting them into structured semantic objects with narrative descriptions. The system maintains context between page elements to reduce hallucination risks and supports custom fine-tuning for specific document types.
Enterprise-Ready Storage - Built on battle-tested open source tech (OpenSearch, MySQL, Redis), the storage layer encrypts data at rest and in transit. You get full control over encryption keys and can scale to millions of documents while maintaining fast retrieval times - all with zero storage costs.
Hybrid Search - It blends text, vector, and micro graph search, allowing developers to fine-tune the results for optimal relevance. A custom-built reranker model refines the results, ensuring the best possible answers from your data. It showed 6x less accuracy degradation compared to Pinecone when scaling up data volume.
Free Data Storage and Evaluation - GroundX offers free storage, regardless of data volume. For debugging and improving results, GroundX provides an "X-Ray" tool that visually shows how documents are parsed. Along with a retrieval viewer to inspect search results and a chat interface to check completions and sources.
Deployment - The platform is built on Kubernetes and can be deployed on-premise, in air-gapped environments, or through their cloud platform. For on-premise setups, GroundX provides an open-source version and a commercial version (with better performance and support) both configurable with Helm. Cloud deployments are subscription-based.
An entirely new way to present ideas
Gamma’s AI creates beautiful presentations, websites, and more. No design or coding skills required. Try it free today.
ByteDance has launched Trae, a new AI code editor directly competing with tools like Cursor and VS Code, built upon the familiar VS Code framework. It is already gaining attention with a free AI agent powered by state-of-the-art LLMs GPT-4o and Claude 3.5 Sonnet.
Trae IDE comes with features like real-time code suggestions, project-level code generation, and multilingual AI chat assistance. With support for both English and Chinese interfaces, Trae helps with project development from zero to deployment through its Builder mode. The editor currently runs on macOS, with a Windows version in development.
Key Highlights:
Free LLM access - Trae comes with an AI agent powered by either GPT-4o or Claude 3.5 Sonnet, both of which are excellent in coding tasks. You can use this agent for code generation, explanations, error fixing, and commenting—all without any cost.
Project-Centric 'Builder' Mode - Trae has a special "Builder" mode that helps you develop projects from scratch, taking care of project setup, task breakdown, code generation, running commands, and offering previews. It works with the AI to actively build your project, making sure you have all steps covered.
Contextually Aware AI - The AI in Trae doesn’t just work in isolation. It can read your code, files, and even terminal output to give better suggestions and answers, giving you more relevant and helpful AI interactions. You can even use images as context in the chat, so if you have designs, error screens, or other reference materials, Trae will use this to better interpret your instructions.
VSCode Migration - Coming from VSCode? Trae automatically imports your extensions, settings, and keybindings. It maintains compatibility with VSCode's extension marketplace and supports the familiar Electron-based architecture, letting you keep your existing workflow while gaining AI capabilities.
Quick Bites
DeepSeek has released the R1 series of reasoning models, similar to OpenAI’s O1 models. The family includes two large MoE models, DeepSeek-R1-Zero and DeepSeek-R1 with 38B active parameters, both designed for maximum reasoning capabilities. Additionally, they've made 6 smaller, distilled models available, fine-tuned on open-source models, for different resource constraints. All models, including the distilled versions, are open-source.
DeepSeek-R1 achieves performance comparable to OpenAI's o1 on reasoning-based tasks, excelling in code, math, and scientific reasoning.
The models were trained using large-scale reinforcement learning with a focus on self-evolution of reasoning, and without the need for initial supervised fine-tuning.
All the models are opensourced under MIT license. DeepSeek R1 is also available via API for: $0.14 / million input tokens (cache hit), $0.55 / million input tokens (cache miss), and $2.19 / million output tokens.
Perplexity has launched Sonar, a search API to integrate real-time, citation-backed search capabilities into your applications. The API comes in two tiers: Sonar and Sonar Pro, with the Pro version offering advanced querying, deeper context windows, and generating twice as many citations on average. Notably competitive on pricing ($5 per 1000 searches for both tiers), both versions provide self-serve API access and do not use customer data for LLM training.
Google has released an updated Gemini 2.0 Flash Thinking model, now available as the exp-01-21 variant in AI Studio and via API. This version boasts a 1 million token context window, native code execution support, and longer output generation. The model demonstrates strong reasoning on math and science benchmarks. API is currently free, but it has a rate limit of 10 RPM. Also, the thinking process is now disabled by default in the API, but can be enabled using config = {'thinking_config': {'include_thoughts': True}}.
Tools of the Trade
Fume: AI agent that manually tests your pull requests by running your app on its own computer. Unlike traditional tools that only analyze static code, Fume can spin up browsers to test UI changes, execute custom test scripts, and provide video recordings of its testing process.
Firecrawl /extract: New endpoint to scrape web data using simple English prompts instead of writing web scrapers. It can crawl entire websites to extract specific information (like contact details or emails). It integrates with Zapier, and currently has 500K free tokens in open beta.
Mermaid to Excalidraw: Converts Mermaid syntax diagrams into Excalidraw drawings. It allows you to create diagrams using text and then convert them to a hand-drawn style for further editing.
Notate: Open-source desktop research assistant that combines local LLM capabilities (via Ollama), vector search (using ChromaDB), and multi-modal document analysis to process, organize, and analyze data privately. Built with TypeScript, React, Python, and FastAPI; works with major AI providers (OpenAI, Anthropic, Google, Ollama); offers features like document analysis, YouTube/webpage processing, and semantic search across your knowledge base.
Awesome LLM Apps: Build awesome LLM apps with RAG, AI agents, and more to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos, and automate complex work.
Hot Takes
AI is eating UI. ~
Naval RavikantThere is no moat. We've hit terminal race condition. Every AI lab in America and China is converging on the same training paradigms. Singularity is inevitable at this point.
We're caught in the gravity well, and you don't realize it, but we've already crossed the event horizon. The attractor state is clear.
ASI is on our doorstep and everyone is running as fast as they can to get there. Blink, and you fall behind. ~
David Shapiro
That’s all for today! See you tomorrow with more such AI-filled content.
Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!
PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉
Reply