Free AI Toolkit (1/3): Run AI on Your Own Laptop and Build With It

You do not need a paid subscription to learn AI. A large part of the best tooling is open source, which means you can run it yourself, read the code and change it. This is the first of three posts that list free tools I would try, with plain-language notes and the licence of each. Tool names are plain text; searching the name will find the official page.

Run language models locally

• Ollama (MIT): no API budget? Download open models and chat with them offline. It supports many open models such as Qwen, Gemma, DeepSeek and gpt-oss.
• llama.cpp (MIT): old laptop with no GPU? It gives fast inference on a CPU, written in C and C++.
• Open WebUI (custom licence): a ChatGPT-style interface for your local models, which can run entirely offline and connect to Ollama or OpenAI-compatible APIs.

If you are new to language models, the video above shows practical, everyday ways to use them, with an honest view of what they get wrong.

Coding agents

• OpenHands (MIT): an AI developer that writes, runs and fixes code, described as a self-hosted developer control center for coding agents.
• aider (Apache-2.0): an AI pair programmer in your terminal that works with your git repository.
• cline (Apache-2.0): a VS Code agent that plans, edits and runs commands.
• gemini-cli (Apache-2.0): a free terminal agent built on Google's Gemini.

Chat with documents (RAG)

RAG stands for retrieval-augmented generation: the system finds the relevant parts of your documents and gives them to the model, so its answers come from your material.

• Docling (MIT): turns PDFs, Word files and presentations into clean, AI-ready text.
• LlamaIndex (MIT): connects your notes and PDFs to a model for question answering.
• RAGFlow (Apache-2.0): a ready-made application for chatting with large document sets.

Agents

• CrewAI (MIT): teams of role-playing agents that work together.
• LangGraph (MIT): reliable, stateful, multi-step agent workflows.
• browser-use (MIT): lets an AI agent operate a web browser.

How to use this list without getting overwhelmed

1. Start with Ollama and one small model. Ask it five questions about a text file of your own.
2. Add Docling and LlamaIndex to build "chat with my notes" over your course PDFs.
3. Only then try an agent framework, and give it a small, safe task.

Small local models make mistakes, so check what they say, just as you would with any AI tool. Next in this series: images, video, voice and no-code apps.

Keep reading

• World-Class AI Education for Rs. 0: A Free Four-Step Learning Roadmap
• Free AI Toolkit (2/3): Images, Video, Voice and No-Code Apps With No Paid Subscription
• Free AI Toolkit (3/3): Test It, Secure It, Learn It From the Source

See the complete AI Era series

---