Introducing Onyx - a fully open source chat UI with RAG, web search, deep research, and MCP
Posted by Weves11@reddit | LocalLLaMA | View on Reddit | 160 comments
Fuzzdump@reddit
Hi, this project looks awesome and it’s been on my radar for some time. The only thing keeping me on OpenWebUI is OIDC. Any chance you’d bring that over to the community edition?
Weves11@reddit (OP)
Yes, we are planning to do that very soon actually!
Look out for an announcement on that going fully MIT in the next couple weeks.
AccomplishedFruit689@reddit
where is the announcement?
Fuzzdump@reddit
Awesome! Looking forward to it.
Weves11@reddit (OP)
Over the past year, I’ve tried many other open-source chat tools, and I found that none of them had the mix of things I wanted: a beautiful chat UI + great RAG + deep research. This inspired me to build Onyx.
All other chats we’ve seen are missing at least one of these key features:
It’s just one line to self-host and the code is permissively licensed (MIT)
We’re constantly adding new stuff, and we’d love to hear from you in our Discord if there’s anything we’re obviously missing! I hope you find it useful 🙏
txgsync@reddit
Thanks for sharing Onyx. I appreciate the hustle, and the feature set looks interesting. However, I noticed you describe it as 'permissively licensed (MIT)' which isn't quite accurate. While the base codebase is MIT licensed, several of the features you're highlighting (multi-tenancy, advanced permissions syncing, analytics, SAML auth) are in the backend/ee directory under a proprietary Enterprise License that requires a paid subscription for production use.
While it is a legitimate business model (similar to GitLab, Sentry, etc.), claiming it's "permissively licensed" in a broad statement is unclear about what's actually MIT vs. what requires a commercial license. Developers evaluating the project need to understand which features they can freely use in production.
I was interested in this as the basis for a chat UI exploring data privacy, but the licensing ambiguity makes it hard to evaluate whether the MIT-licensed portions would be sufficient for my use case.
MDSExpro@reddit
Aaaand I lost all interest.
iMakeSense@reddit
For production use. If you're using this for a profit, I can totally understand why someone would put something like this in given that people spend 6 - 12k easily on GPUs on this sub.
Xamanthas@reddit
Oh nooes, you cant unfairly profit off someone elses work?
bhupesh-g@reddit
I agree with you, for personal use its all free...paid items are only for commercial purpose which is absolutely fine. People wants someone should build solution for them which they can pick up and sell, thats ridiculous.
Innomen@reddit
This should be top comment. /sigh
9acca9@reddit
lol, the same for me. How i hate this posts.
russianguy@reddit
https://sso.tax/
evilbarron2@reddit
Note that it only supports paid search providers (eg: no searxng support). I think the providers offer a free tier, but I lost interest after seeing lack of searxng.
Key-Singer-2193@reddit
I mean isnt this something you can get claude code to do for you in like 10 minutes?
evilbarron2@reddit
Almost certainly not
Weves11@reddit (OP)
That's a great suggestion. Will add it to my list of TODOs
nullnuller@reddit
Also duckduckgo I think it's free. In general have an endpoint and an optional API key input box.
evilbarron2@reddit
Yes, just be aware that there’s some issues with DDG, not sure if it’s rate limiting or unreliability or access limits. You may need to write more code to deal with its fussiness.
But lmk if I can help with searxng in any way. I have a local reliable install, I’m not a complete idiot with Linux sysadmin, and I have some experience with testing
Free-Internet1981@reddit
Okay i was excited there for a second thanks for letting us know
griffinsklow@reddit
Eww. Another one for https://ssotax.org/
Elvarien2@reddit
thank you for clearing this up, bait and switched !
Weves11@reddit (OP)
sorry, definitely didn't mean to bait and switch. To clarify, all listed features (and all future chat features, like memory, code interpreter, etc.) are MIT licensed and completely free to use
Weves11@reddit (OP)
Very fair point.
The way I think about it is that any/every feature needed to have a great chat experience (all of the things listed in my comment above, although this is just a subset) should be completely free to use.
The project actually started as a pure RAG/enterprise search system called Danswer, and the enterprise features were built for that world. I’m moving all features from ee to MIT that fit into the bucket above (e.g. advanced SSO).
If there’s anything missing you feel like you need to feel confident to use it, let me know.
Xamanthas@reddit
Unsure how how you would structure this but I would strongly reccomend seperating commercial and 'selfhost' repos somehow. Also reccomend AGPLv3 if your concerns are corpos taking your work and profiting without contributing back.
P.S Respectfully, dont use LLMs to answer if you can avoid it, can see it in a few of your comments
Weves11@reddit (OP)
Yea, I'll figure out a way to do that (likely separate repos). I'm also moving some of the old "ee" features like SSO into the MIT repo.
9acca9@reddit
GFY
coder543@reddit
I just think “fully” open source carries a very different connotation than open core. I am still interested to look into this project.
HollyNatal@reddit
Ao analisar o diagnóstico, perdi completamente o interesse na plataforma. Já estava com um pé atrás, pois a apresentação parecia mais uma estratégia de vendas, e quando percebi que era necessário assinar um plano, minhas suspeitas se confirmaram.
rm-rf-rm@reddit
did you try Msty? How does it compare to that?
jadbox@reddit
Lovely app, but I really wish it was a native app executable as starting a Docker app on Windows is painfully slow.
dobrych@reddit
Love your product, I'm using it for crawling a few websites of my interest to build niche private knowledge base with chat interface. Very cool and practical!
MidAirRunner@reddit
Consider linking the GitHub in the post.
Weves11@reddit (OP)
Hm, I can't edit the post any more... I've added it to my comment ¯\_(ツ)_/¯ thanks
akierum@reddit
Great project, but no native installers for Windows, Mac. Fail.
NoFudge4700@reddit
Does it work with Qwen models?
Weves11@reddit (OP)
Yes. Easiest way is through Ollama + Onyx
NoFudge4700@reddit
Why not llama.cpp?
Weves11@reddit (OP)
You can configure any provider that does openai compatible endpoints. I only suggest Ollama because it's what I use for self-hosted models and I'm building a direct integration
debackerl@reddit
If you want to support guys on Ryzen AI APUs, be sure to trst with llama-server, it works awesome with Vulkan!!
Ollama is a PITA on the APUs... it insists on using ROCm, which is heavy to install without additional speed advantage, and Ollama doesn't detect GTT memory properly.
Thx a lot in any case!
CSEliot@reddit
lm studio supports the stryx halo APUs great, and can be run as a server using openai endpoints that OP claims also uses.
debackerl@reddit
I looked into it, but I needed something without a GUI. I installed all that on a headless machine, it's a pure server distro.
CSEliot@reddit
Makes sense.
Chance_Value_Not@reddit
Lm studio is based on llama.cpp
OcelotMadness@reddit
Its not based on it. Its a seperate wrapper that downloads and swaps Llama cpp instances as needed.
Chance_Value_Not@reddit
What runs the LLM? Llama.cpp.
rm-rf-rm@reddit
if you have first class support of llama-swap or llama-server, then im sold.
Im sick of open source projects not supporting (in the emotional sense rather than the sofrware sense) the most legitimate open source project (llama.cpp).
DistanceSolar1449@reddit
Looks like they do
https://docs.onyx.app/admin/ai_models/custom_inference_provider
They use LiteLLM as their backend so it carries over support from there.
https://docs.litellm.ai/docs/providers/openai_compatible
rm-rf-rm@reddit
DistanceSolar1449@reddit
They use LiteLLM for OpenAI and Ollama too, and there's no difference in code path between them.
NoFudge4700@reddit
I am trying to get it work with llama.cpp but the docs are not very clear about it, could you please add llama-server support to it or docs instructions on how to add that? I tried a couple of things mentioned in docs but it won't let me enable it.
I keep getting this error: litellm.NotFoundError: NotFoundError: OpenAIException - Error code: 404 - {'detail': '404: Not Found'}
Chance_Value_Not@reddit
Ollama is hot garbage compared to the llama variants. LM Studio is a honorable mention, or just give vanilla llama.cpp a try
planetearth80@reddit
We have to admit that ollama makes it very easy to serve multiple models without having to worry about swapping them manually. There’s a reason why it’s so popular.
Sloppyjoeman@reddit
I do agree, llama-swap isn’t too complex and achieves the same thing. It would be nice if it were a native feature of llama-cpp though
klop2031@reddit
yeah except sometimes llama-swap wants to keep models in memory and I have to kill it
rm-rf-rm@reddit
what is your
ttlsetting in the config.yaml?Free-Internet1981@reddit
Does it have sso auth? Or is it geared towards personal use?
Weves11@reddit (OP)
Right now we only support Google SSO, but arbitrary OIDC/SAML support is coming soon!
shinkamui@reddit
Will oidc be free for lab/non commercial use or a licensed only feature?
Weves11@reddit (OP)
It will be MIT licensed and free for use by anyone!
geek_404@reddit
Started working with it tonight and works great on Anthropic but had issues getting lm_studio configured and the error message wasn't especially helpful. I'll try some other llm apps like ollama to see if I can figure out the config details a little better. I am excited to use it for personal use but this actually fits a use case at work so I'll be reaching out to get more details and hope to collab on our use case I think you might find it a useful use case for your tool.
Weves11@reddit (OP)
Huh, if you're able to find what's going wrong, we'd love to have a contribution to make that easier. Or if you just want to tell us what went wrong I can take it from here!
Weves11@reddit (OP)
Huh, if you're able to find what's going wrong, we'd love to have a contribution to make that easier. Or if you just want to tell us what went wrong I can take it from here!
DistanceSolar1449@reddit
Is this built on https://www.assistant-ui.com/ ?
Weves11@reddit (OP)
It’s not! NextJS + Tailwindcss + a bit of shadcn
Gsfgedgfdgh@reddit
I tried it, and it looks really promising. Basically, it can work as a locally run Perplexity "killer." From a privacy viewpoint, that is great, especially in the legal field. . It would be great if there were a way to see the sources it finds, similar to https://app.getcoralai.com. Then, you could use it for research.
DistanceSolar1449@reddit
Anything would be a perplexity killer if they'd just support openai web search.
https://platform.openai.com/docs/guides/tools-web-search?api-mode=responses
Literally, it just needs to add "tools": [{"type": "web_search"}] to the OpenAI api query, and it'd be set. No need to configure exa or serper or searxng or whatever.
It's disappointing when it's not supported.
Steus_au@reddit
i’m thinking to try its RAG and compare to openWebUi but yet struggling to run it on my mac mini
nonerequired_@reddit
What is the main differences between Openwebui besides being fully open source
Weves11@reddit (OP)
One of the biggest differences is native RAG/file indexing that scales. In my experience, it's a huge pain to set up OpenWebUI with my docs. Onyx comes with rag pre-configured with indexing and retrieval pipelines.
There's other feature differences like no deep research mode (what I show in the video), and no way to create assistants (pre-configuring prompts, tools, accessing this configuration quickly, and sharing it with others)
Key-Singer-2193@reddit
Have you done load tests and stress test as to how long it takes to index 100k documents? or some large amount of documents and at what point will the system crash under load?
Weves11@reddit (OP)
Yes I have! I've seen Onyx deployments with >5 million documents actually.
The time to index is mostly dependent on how much embedding capacity you have (the actual indexing is parallelizable). With a GPU, you should be able to do 100k in a few hours (dependent on document size).
Under the hook, Vespa is used as the vector DB. It can scale well beyond 10mil+ documents (it was apparently used at Yahoo to power search), although memory requirements do scale linearly.
MasqueradeDark@reddit
"no way to create assistants" - wait , you say that in ONYX I will not been able to create "assistants" and personas? That's a HUGE, HUGE bummer.
Weves11@reddit (OP)
Sorry to clarify — in other tools, they often don't have the ability to create personas. In Onyx you absolutely can, that's one of the key things I wanted to support from the beginning.
j17c2@reddit
I think the person you replied to just implied that you CAN do that in OpenWebUI. You can create "assistants" with a dedicated base model, prompt, image, files, tools, and select their capabilities (however that works). I also believe you can export your assistants and share it with someone.
Weves11@reddit (OP)
ah yes, you're certainly right you can. I've used OpenWebUI quite and I didn't notice since it was called Models (overlapped with LLMs).
Playing around with it a bit, it's quite similar with a bit less emphasis / quality on the RAG side of things. I have to credit the customizability / sharing though, it's quite flexible on that front.
__JockY__@reddit
No, he's mistakenly claiming the Open-WebUI doesn't do assistants, when in fact it does.
ShengrenR@reddit
"Fully open source"... but also
Weves11@reddit (OP)
See my comment in https://www.reddit.com/r/LocalLLaMA/s/2gihmqhi7g.
Again it’s 100% a fair point and should have been called out more clearly. But everything needed for an amazing chat + RAG + deep research system should (and will always be) fully open source.
ShengrenR@reddit
Everybody's got to eat - just next time maybe announce as "open core" and nobody gets confused.
From a dev perspective, I wish projects like these would just break the thing into different packages. Eg onyx-standard and onyx-enterprise, then a single license for each. Don't plan on trying to get an ee license? Just don't install the other component and no accidental misuse.
JumpyAbies@reddit
I Like this!!
Weves11@reddit (OP)
That's a great idea. I'll plan to do that shortly.
nonerequired_@reddit
Didn’t check that. Thank you for information
__JockY__@reddit
I was new to LiteLLM, which made the API endpoint configuration super painful. Took waaaaay too long to figure out all the correct incantations to make it work.
I finally also figured out that an API Base of
localhostor127.0.0.1simply does nothing. I had to change the base URL to the network-facing IP address of my server for it to actually detect the vLLM API.Question for you: does RAG require Unstructured?
Key-Singer-2193@reddit
I had a strange issue concernign this as well "the interface doesn't seem to stream, ether. I have to wait for the entire response before it's rendered. Additionally, we get the first tag literally rendered onto the window, but no . Interestingly, when I try a second time it does actually stream into the interface.... weird."
I have an issue where the answer that should be the AI response to the chat window actually is in the "Steps" container. So I never see the answer from AI in the chat because for some reason it was included as a step in the container.
__JockY__@reddit
If you haven’t tried cherry studio you might find it compelling. There’s a lot of Chinese language stuff adjacent to it, if that’s a factor in your operating environment, but it seems stable, mature, fast… I like it. The RAG feature worked well in the one test I did.
But for ease of use it’s hard to beat openwebui. I very much like that I can turn it into a “real” app on my MacBook instead a shitty tab in a browser by using the personal web app (“PWA”) feature. I can cmd-tab to it, etc.
richardanaya@reddit
No offense but that’s a terrible name to use in AI.
Obvious-Program-7385@reddit
Yeah, It also shows that OP has no background in deep learning what so ever. They would have definitely chosen a different name than Onyx,
Weves11@reddit (OP)
Are you referring to the similarity to https://onnx.ai/? It's a project that I'm quite familiar with
richardanaya@reddit
Imagine I named my companies project JPEG and you’ll see my concern ;)
218-69@reddit
No one uses or knows what onnx is, and it doesn't have a monopoly on words.
rm-rf-rm@reddit
it doesnt have a monopoly on words but it does have a huge mindshare in the community namespace. OP is doing to a disservice to everyone including his own product by adding confusion.
Key-Singer-2193@reddit
people complain about anything these days. OP you are doing great work. Love what you are doing keep it up
richardanaya@reddit
A monopoly isn't needed to be confusing.
OcelotMadness@reddit
onnx is a huge framework for doing ML. That first part is absolutely untrue.
toothpastespiders@reddit
While kind of a different naming issue, I'm still amused at how much I now know about the viability of teaching pet axolotls thanks to the axolotl fine tuning platform.
Fuzzdump@reddit
Is there a “lightweight” deployment that’s mainly just the chat UI for connecting to remote inference endpoints? I noticed the model servers are pretty large, but I’m not sure if can drop them from the compose stack without breaking everything. (I tried the full deployment but my current server is too low power to run it.)
MasqueradeDark@reddit
Can you post a YouTube video and get a summary? This is one of the reasons I'm using OpenWebUI
ZYy9oQ@reddit
How does the deep research compare to the current SOTAs? Did you base it on any in particular?
Weves11@reddit (OP)
I've done some basic benchmarking + asked some friends their preference. Compared to Exa deep research Onyx consistently outperforms (>90% head to head preference), but compared to OpenAI's deep research Onyx is a bit behind (\~30% win rate).
I didn't model it off any deep research in particular, but I did take inspiration from OpenAI, Anthropic, and some LangChain blogs.
ZYy9oQ@reddit
One thing I've been wanting to try is have multiple models "play off" against each other e.g. ask one to break another's claims into verification steps and research each then challenge the first.
Would this be easy to hack into onyx? Any thoughts/tips on doing this kind of thing?
SillyLilBear@reddit
You have a list of what features are included and which are pay walled?
Weves11@reddit (OP)
Every chat/core experience feature is completely open-source! So custom assistants, RAG, web search, MCP, image gen, etc.
Currently, the ee features are permission syncing (e.g. for RAG, pulling in permissions from enterprise tools and applying them within the tool), a few admin dashboards, and some whitelabeling (ofc the code is MIT, so you can just edit things yourself if you want).
SillyLilBear@reddit
Was just looking through the demo, it looks really nice and snappy. Started looking at the docker compose, oomph that's a lot to setup. I do want to give it a try though, as I am not a fan of OpenWebUI. I saw on the website it says Knowledge Base is not in the community edition.
Free-Internet1981@reddit
I dont get it, are enterprise features paywalled or not?
Elvarien2@reddit
What a bait and switch.
Wanting to get paid for your work is fine but at least be honest about it.
rm-rf-rm@reddit
whats the bait and switch here?
gpt872323@reddit
So Onyx is the UI layer that can work with any openai compatible api?
Hurricane31337@reddit
The last time I checked, only OpenAI models were actually using real tool calling and custom API models were just using plain text tool calling, therefore being very unreliable. Is this actual tool calling support a model/endpoint setting now or is this still hardcoded somewhere in the backend code?
Weves11@reddit (OP)
Custom models are still using plain text tool calling, but that's actually changing in \~2 weeks! There's a major refactor of the backend "agent" flow going live (moving from LangGraph to Agent SDK + simplifying quite a bit), that should improve quality + make it a lot easier to modify going forward.
With that said, the plain text tool calling should still allow for effective arbitrary tool/MCP calling.
Zor25@reddit
Curious as to why you switched from langgraph to agent sdk
dobrych@reddit
Is that your internal Agent SDK or particular open source project? curios how open that SDK would be.
Weves11@reddit (OP)
It's https://openai.github.io/openai-agents-python/ ! Even though it's maintained by OpenAI, it supports all models well + aligns well with my belief on how "agents" should work (simply defined as 1) instructions 2) tools and 3) a LLM.
Hurricane31337@reddit
Okay thanks, I’m looking forward to it!
BumbleSlob@reddit
I know for sure Qwen 3 30B uses native tool calling at least
-Django@reddit
What do you mean by real tool calling?
random-tomato@reddit
Models have this built in tool calling that they are trained with (you can see this in the model chat_template.jinja file). Apparently this project is just prompting the LLM to reply in a certain format in order to give the "illusion" of tool calling.
SoupyOriginal@reddit
What’s different than Open WebUI?
xHanabusa@reddit
I installed it, gave it an ExaAI key for web search. It worked correctly the first time, but then failed with my second try. It generated the queries, but then stalled while generating the final response leading to an empty output. There's 100x of the following line in the logs, not sure if it's related:
Generator output exceeded limit of 1000 items, output not logged. Increase BRAINTRUST_MAX_GENERATOR_ITEMS or set to -1 to disable limit..Tried it for the 3rd time, and the same thing happens. It shows queries being generated, the search being performed, then just stops.
Also, I also found the initial set-up for the LLMs to be unwieldy. I tried adding a custom provider, and found I actually have to key in a value for max_token, the 'default' means nothing, (why have it then?) and there's no indication on why it's failing (clicking the enable button just flashes it and nothing happen, no error pop-up or anthing), I had to look at the docker logs to understand the issue.
Appreciate your work, but it seems still buggy and unstable. Maybe I'm nit-picking, but if you advertise web search in the title I expect it mostly work.
No_Comparison1589@reddit
Nice, thank you for sharing. i was getting realy unhappy with openwebui in our company and will definitely have a look at this. Maybe you can say something about the pain points we have with OWUI and if they are solved with Onyx:
-the RAG is intransparent and doesnt work most of the time
-MCP is very hesistantly integrated
-no statistics about usage
Features i dont want to miss:
-Adding custom openai connections, sharing them with a group (limiting them to groups)
Careless_Garlic1438@reddit
No KRAG? I’ve been playing with RAG and they all seem to fall trough as the lack of context will always bite you in the end. Also how do you handle tables and graph‘s in documents, understanding is a must as for example you have a table with items that are supported and not supported with lets say a Red Cross and a green checkmark … the answers include a mix of both with RAG, while you maybe only want one or the other. So what I’m really hoping for is complex document ingestion + KRAG, they are really next level, just RAG is mediocre at best.
Weves11@reddit (OP)
Prototyping a KG-based RAG approach now actually! Should have something ready for beta testing hopefully next month. Will let you know when it's live
Careless_Garlic1438@reddit
Cool! If you find a solution to dissect documents that contain tables and graphs that would be the ultimate solution. I tried to get a this GitHub running, but need to put more effort into it:
https://github.com/HKUDS/RAG-Anything
maigpy@reddit
for document ingestion, the best results have come from sonnet / ChatGPT computer vision models, aptly prompted as per the document typem
projak@reddit
Add a VM sandbox to it like manus or lemonai
Weves11@reddit (OP)
Great idea! Added to my list
projak@reddit
♥️
ThinCod5022@reddit
expose chat_completions/responses API? Thanks for your work!
Weves11@reddit (OP)
That's a great idea, thanks!
Steus_au@reddit
Hi Chris, would be Ollama API (especially the websearch) supported?
Weves11@reddit (OP)
We do support the Ollama API, but not their websearch (yet). Currently, we just support Exa, Serper, and Firecrawl.
NotLogrui@reddit
Does it support /workflows?
What about activating custom workflows via n8n?
Weves11@reddit (OP)
No workflows yet, although the RAG APIs (https://docs.onyx.app/developers/api_reference/search/document_search) are accessible via n8n. I've heard quite a few folks doing that with the project.
inmystyle@reddit
Thank you for your contribution to the project, this is exactly what I’ve been looking for all the time. Thank you
Weves11@reddit (OP)
Awesome, would love to your thoughts.
LostHisDog@reddit
It's so hard to keep up with all these tools launching. Anyone happen to know off hand if this can act as an API I can call from another machine / script and if it works with local and cloud apis? I'll dig in and read eventually but that need to read list is getting real long.
Weves11@reddit (OP)
Yes to both! Checkout https://docs.onyx.app/developers/overview for our API docs and https://docs.onyx.app/admin/ai_models/overview for connecting to local/cloud LLMs!
Money_Hand_4199@reddit
Which APIs are available in the free version and which are only working in the enterprise one?
LostHisDog@reddit
Cool, thanks for the info! I'll give it a go tonight.
Flamenverfer@reddit
My work has this one and its really annoying to use.
Automatic updates on by default and completely broke a lot of the RAG functionality and previous chats.
Websearch is non functional depending on your setup. And our infrastructure guy took to discord to get some support regarding this its been crickets.
Weves11@reddit (OP)
What issues have you been seeing with web search? We've completely re-vamped it recently + added support for additional providers.
And would love to help out with the infra side of things if y'all are still running into issues.
jazir555@reddit
Can you please add a toggle for autoupdates? Undisableable auto updates is a huge no from me dawg.
Weves11@reddit (OP)
I'm also emphasizing testing / backwards compatibility a lot more than earlier on in the project now that there are more folks depending on it. If you're willing, I encourage you to give it another shot! If you continue to see the issues you're referring to, I totally understand the frustration
SignalX_Cyber@reddit
Any benefits over something like chainlit?
WithoutReason1729@reddit
Your post is getting popular and we just featured it on our Discord! Come check it out!
You've also been given a special flair for your contribution. We appreciate your post!
I am a bot and this action was performed automatically.
badgerbadgerbadgerWI@reddit
Been looking for something exactly like this. How's the RAG implementation compared to something like Danswer? Gonna test it out this weekend.
Weves11@reddit (OP)
haha funny thing — Onyx actually is Danswer. We re-named the project \~6 months ago
badgerbadgerbadgerWI@reddit
Well, I'll give it another look! Lol
grutus@reddit
been waiting for work to update the models, need claude 4.5 and gemini 2.5 pro. dont like how openAI responds to our questions
Weves11@reddit (OP)
they should both be available now in the latest nightly!
ELPascalito@reddit
Custom mcp support? Agents?
Weves11@reddit (OP)
Yes to both! https://docs.onyx.app/overview/core_features/actions + https://docs.onyx.app/overview/core_features/agents
FantasticTraining731@reddit
onyx is an evil corporation. They poisoned our water supply, burnt our crops, and delivered a plague upon our houses!
Theio666@reddit
Interface looks really good, lowkey wanna steal this for our agent system, we wrote on streamlit and it's fucking ugly lol
Weves11@reddit (OP)
Thanks, feel free to fork and/or peek at the
web/dir in the repo. There's also an API you can plug into (just a FastAPI backend)Hurricane31337@reddit
Is the API open (MIT license) or EE only?
Weves11@reddit (OP)
API is open!
Hurricane31337@reddit
Awesome, thanks!
NickCanCode@reddit
I never use RAG with AI. Does it mean I can talk with the AI and it can drop notes to summarize what we talked?
Weves11@reddit (OP)
It’s more the other way - if you have a bunch of notes already created (e.g. everything you / anyone you know have written) you can connect up this information to the AI, and it will use it to inform its responses! You can also pull in any website, textbook, forum, and even automatically sync from other platforms like Bookstack.
“Memory”, which would involve the AI automatically pulling out key parts of conversations is actually coming soon as well!