HumlaAsk your meetings anything. It stays on your Mac.
Humla's chat searches your meeting history, reads the notes that matter, and answers with citations — fully offline with local models, or with your own OpenAI key. Free either way.
Free · Open source · macOS 13+ (Apple Silicon)
It actually looks for the answer.
Works the question
Most “chat with your notes” tools run one search and hope. Humla searches, opens the notes that look promising, tries a different angle if the first search misses, and only then answers. You watch it happen — “Searched your notes · Read 2 notes” — so you know where the answer came from.
Answers you can check
Every answer comes with citation chips. Click one and you land in the source note, on the day it happened. No “trust me” answers.
Finds it by meaning
Notes are indexed in small pieces, by keyword and by meaning. So “the pricing discussion from spring” finds the right stretch of a call from May — even if nobody said the word pricing. A passage 40 minutes into an hour-long transcript is just as findable as the first line. We test it daily in Norwegian and English — and the models underneath handle most languages, so yours is likely covered.
Ask wide or narrow
Point a question at this note, its folder, or everything you've ever recorded — the picker sits right in the composer. And conversations stick around: every note keeps its threads, so you can pick one back up next week.
How it works: agentic RAG, on your Mac.
Agentic retrieval
The model gets three tools — search, read a note, browse — and loops up to six steps before it answers. If retrieval comes up empty, it says so instead of guessing.
Hybrid search
SQLite FTS5 (BM25) plus embedding similarity, fused with reciprocal rank fusion, over ~750-token paragraph-aware chunks of body, transcript, and summary.
Local models
Chat via Ollama — Gemma and Qwen run well on Apple Silicon. Embeddings via EmbeddingGemma, on-device. The cloud path is your own OpenAI key, nothing else.
Your data, your disk
Notes, search index, and chat history live in local SQLite. And it's open source — read the code.
Fully private, if you want.
Point Humla at Ollama and everything — the chat model and the search index — runs on your Mac. Nothing leaves your machine, and the local experience is the whole product, not a demo tier. Prefer cloud? Bring your own OpenAI key. Either way, personal chat is free forever — no quotas, no meters.
The same chat, for your whole team.
On Humla Cloud, chat works across your team's shared workspace — ask about any meeting, whoever took it, and get cited answers. Conversations are visible to the workspace, so teammates build on each other's threads.
FAQ
Does chat send my notes to the cloud?+
Only if you choose OpenAI as your chat provider. With Ollama, the chat model and the search index both run on your Mac — nothing leaves your machine.
What does chat cost?+
Nothing. The app is free and open source, and personal chat has no quotas or meters. If you use OpenAI you pay them for your own usage; with local models there's nothing to pay at all.
What do I need to run it offline?+
Ollama, plus two downloads: a chat model (Gemma or Qwen) and EmbeddingGemma for search by meaning. Humla's settings walk you through both.
Which languages does it work in?+
Norwegian and English are what we test daily, including notes that mix both. The models underneath are multilingual — a hundred-plus languages — so most languages work fine. If it handles Norwegian, it can probably handle yours.
Which AI models does it use?+
Your choice: local models through Ollama, or OpenAI with your own key. The local experience is the full product — same search, same citations, same everything.