Brainiac v3.0 released
#brainiac #emacs #productivity #system
Dear All,
It's time to release a new version of Brainiac.
These are the highlights of the release:
ollama-pal, which is based on the mini version ofollama-buddy, implements everything I deem necessary for “chatting with your documents”. You can find detaills on the implementation here, but the major additions are:- Multilingual RAG pipeline (including OCR garbage detection) backed by a reasoning model of your choice. This performs very well even with small reasoning models that can run on CPU only.
- Deep Research Mode that expands queries and context before reasoning, increasing the relevant context that the model gets. But this takes longer time to process, although I still find it acceptable with the default qwen models.
- Deep Search through all your notes, documents, and images. This works wonders, when you have a lot of screenshots and other types of documents, that are not pure text. As the information is extracted and then stored as chunks in the
chromadb, we can do a sort of semantic search, when the classical keyword search produces no results. - Other additions are multi-session support, switch system prompts etc.
I also added
br-abbrev-pickerfor AI prompts, templates, etc., integrated withabbrev-mode. This makes it very easy to pick reusable prompts you defined by yourself.
Most of the improvements have been implemented using AI coding agents. It was fun playing with them, experimenting with this new way of working and basically implementing things, which I would never do due to my low-level coding skills.
Get the release here.
Keep on hacking!

