danswer VS llamafile

You might want to look at https://github.com/danswer-ai/danswer as well, as it sounds like their UI might be of suited for your use case.

There are already several RAG chat open source solutions available. Two that immediately come to mind are:

Danswer

https://github.com/danswer-ai/danswer

Khoj

https://github.com/khoj-ai/khoj

We have a connector interface and build guide for contributors: https://github.com/danswer-ai/danswer/blob/main/backend/dans...

Should be not too bad to build one out! Fun fact, more than half the connectors were built entirely by community members who needed them for their own teams and we're super grateful when they contribute it back to the repo.

Danswer is an MIT licensed project that can connect to a wide range of SaaS tools and provide a search/chat (RAG) functionality to help your team discover information and to turn that information into deeper understanding and actionable insights.

Code here: https://github.com/danswer-ai/danswer

Slack questions have always been a huge time sink for me. They’re a distraction that pulls me away from what I’m doing, and often requires digging up old knowledge. If I’m in the middle of something complex, I may take a while to context switch and get around to answering, which leaves the asker blocked for hours.

Addressing this seems simple: give an LLM your organizational context and plop it in Slack to answer things for you.

So that’s why we built DanswerBot! It’s MIT licensed (https://github.com/danswer-ai/danswer) and completely free to use. The bot can automatically sync with and back answers based on documents from Slack, Google Drive, GitHub, Confluence, Jira, Notion, local files, websites, and much more.

Quick demo vid: https://www.youtube.com/watch?v=5q35NeqsMnU

A quick note on hallucinations: in order to reduce their prevalence, all answers are backed by quotes. If the LLM-provided quotes don’t match any document or no quotes are given, we’ll warn the asker that something may have gone wrong. Additionally, all used documents are linked in case the asker wants to double check the answer. Answers can be thumbs-upped or thumbs-downed and all questions / answers are recorded in Postgres for easy future inspection / analysis.

For usability, we provide an admin dashboard where you can configure connectors (we have 14 currently). Once a connector is set up, we poll data sources every 10 minutes to keep answers up to date. Which LLM to use is also up to you - DanswerBot can be configured to use a locally hosted model, Azure OpenAI, or OpenAI directly.

Finally, if you aren’t a slack user (or if you just prefer a more tailored UI), there’s also a web interface to ask questions against your knowledge base. A short demo for that can be found at: https://youtu.be/cWWtnuVCUX0

Of course there’s a bunch more that I can’t cover in one post - happy to take questions in the comments (or in our Slack / Discord, which are linked on the Github repo).

If you’re interested in testing this out yourself, you can easily run everything locally with a single command. Docs to help you can be found at https://docs.danswer.dev/quickstart!

Code: https://github.com/danswer-ai/danswer

Do you want that integrated in the browser? Because Mozilla already distributes offline LLMs: https://github.com/Mozilla-Ocho/llamafile

See https://github.com/Mozilla-Ocho/llamafile, a standalone packaging of llama.cpp that runs an LLM locally. It will use the GPU, but it also falls back on the CPU. CPU performance of small, quantized models is still pretty decent, and the page has estimated memory requirements for different models.

I think the llamafile[0] system works the best. Binary works on the command line or launches a mini webserver. Llamafile offers builds of Mixtral-8x7B-Instruct, so presumably they may package this one up as well (potentially a quantized format).

You would have to confirm with someone deeper in the ecosystem, but I think you should be able to run this new model as is against a llamafile?

[0] https://github.com/Mozilla-Ocho/llamafile

Thermostats: https://www.sinopetech.com/en/products/thermostat/

I haven't tried running a local text-to-speech engine backed by an LLM to control Home Assistant. Maybe someone is working on this already?

TTS: https://github.com/SYSTRAN/faster-whisper

LLM: https://github.com/Mozilla-Ocho/llamafile/releases

LLM: https://huggingface.co/TheBloke/Nous-Hermes-2-Mixtral-8x7B-D...

It would take some tweaking to get the voice commands working correctly.

While I did not succeed in making the matmul code from https://github.com/Mozilla-Ocho/llamafile/blob/main/llamafil... work in isolation, I compared eigen, openblas, and mkl: https://gist.github.com/Dobiasd/e664c681c4a7933ef5d2df7caa87...

In this (very primitive!) benchmark, MKL was a bit better than eigen (~10%) on my machine (i5-6600).

Since the article https://justine.lol/matmul/ compared the new kernels with MLK, we can (by transitivity) compare the new kernels with Eigen this way, at least very roughly for this one use-case.

Yes, they're just ZIP files that also happen to be actually portable executables.

https://github.com/Mozilla-Ocho/llamafile?tab=readme-ov-file...