Notes as compressed Markdown on disk

Keep your own reference notes and let OwnLLM use them when you ask a question, without loading them all into memory.

ownllm context add handbook.md          # stored compressed as handbook.md.gz
ownllm context list
ownllm ask --context handbook "What is our refund policy?"
ownllm context find refund              # see what would be picked, without asking the model
ownllm context remove handbook

• Small on disk. Notes are stored gzip-compressed (.md.gz); typical Markdown shrinks to about a quarter.
• Small in memory. For each question the notes are streamed from disk, split into sections at headings, ranked against your question by plain word matching, and only the best few sections (about 3,000 characters) go into the prompt. That keeps the prompt short, which matters on an 8 GB computer.
• Repeatable. The same question picks the same sections.
• Traceable. The answer ends with a Sources: line naming the notes and headings that were used.
• Honest. If nothing matches, the model is told so and says the notes did not help.
• OwnLLM's own help is built in: ownllm ask --context ownllm "How do I update?".

Word matching is simple on purpose: it will miss a section that uses different words from your question.