🧩 AI Context, Prompts & Skills
How to see tokens, fit more into the context window and give an AI the right knowledge at the right time: tokenizers, skills, the Model Context Protocol, prompting guides and tools for searching your own documents (RAG). Programs to run models are in Local AI; the background is in my note Tokens and Context.
Open in the interactive Digital Resources
OpenAI Tokenizer
platform.openai.comPaste any text and see how it is cut into tokens, with colours. Try the same sentence in English and in Slovenian.
Tiktokenizer
tiktokenizer.vercel.appAn open-source token viewer for many tokenizers – also shows how a whole chat with system and user messages is turned into tokens.
tiktoken
github.comThe fast tokenizer library from OpenAI for Python. Count tokens before you send a long text, or see where a document will be cut.
Agent Skills
agentskills.ioThe open standard for skills: a folder with a SKILL.md file of instructions, scripts and examples that an AI loads only when it needs them. Supported by Claude, ChatGPT, Gemini CLI, Copilot and others.
Anthropic Skills (examples)
github.comA collection of ready-made skills – documents, spreadsheets, presentations, design – and the best place to see how a good SKILL.md is written.
Model Context Protocol
modelcontextprotocol.ioThe open protocol that connects AI assistants to programs and data – files, calendars, databases, home automation. Since 2025 under the Linux Foundation.
Ollama – Context Length
docs.ollama.comHow big the context window in Ollama is by default, how to make it bigger and how to check with ollama ps whether the model still fits on the GPU.
Prompt Engineering (Anthropic)
docs.claude.comClear, practical advice on writing instructions for an AI: be specific, give examples, structure long documents, and put the question after the material.
Prompt Engineering Guide
promptingguide.aiA free, open guide to prompting techniques, RAG and agents, with links to the research behind them.
Chroma
trychroma.comAn open-source vector database for RAG: store your documents as embeddings and give the AI only the parts that match the question.
LlamaIndex
github.comAn open-source Python framework for connecting your own documents to a language model – loading, splitting, indexing and searching. Works with Ollama.
Lost in the Middle (paper)
arxiv.orgThe 2023 study showing that language models use information at the start and the end of a long context much better than in the middle. Short and readable.