Local UNLIMITED Memory Ai Agent | Ollama RAG Crash Course

[@theKEVman011]

New to Linux (switched 2 weeks ago), new to AI (started playing around last week), new to Python (1st day today), and after watching the video a few times over, EVERYTHING was able to be processed and retained. As someone living with a TBI, and a very long history with memory issues, this is a game changer in helping to make my own assistant. Kudos 10,000x over for putting this out there!

[@Ai_Austin]

really cool to hear Kev. thank you tor sharing that experience. part of why i code, is because i like how its made me think more efficient and logical. hearing that i can communicate what i know to people just starting, is my fuel for this channel 🙏

[@theKEVman011]

@@Ai_Austin Hell, It's been 4 days so far with me getting the program up and running, playing around with it, and implementing my own features. The jump in my ability to think more efficiently and logically, in terms of making a program do what I want it to do, is very welcoming. Hopefully it's not too much related to newbie gains and I can keep it going. I've been wanting to learn some type of programming for 10+ years now and this has finally been the project that sticks. So far, I've been able to import the time/date and get the assistant to only reference the time/date under the correct conversational context by making sure its not injected into the convo unless certain time/date references are hit inside the users prompt. It's different than the /recall, /forget, / memorize way but seems to be working good.

[@BeTheFeatureNotTheBug]

Thanks!

[@FightFlixTv]

I've never actually coded in python, and I was able to get this workin perfectly, (had to hitup claude 3.5 a few times in cursor) but it's a pretty cool feeling to complete this without any bugs, great pacing, great code, and great visual instructions! now to actually learn the code...

[@gsanjeevkumar]

Absolutely Awesome, the pace of presentation is perfect.

[@actepukc]

I'll try to build gui and streamline the installation as much as possible, will share the code later - if it's successful :) If not ... will keep using the CMD. Great job and thank you for sharing

[@martinhoneves]

This is madness... In a very good way. Thank you for your time

[@zwelimdlalose1059]

All i can say is thank you for your time

[@ginisksam]

You are a master and maestro! Finally got to install and get postgresql going. Previously avoided as most tubers recommended using Docker and complicated terminal installation in Linux. Managed to type it out and works like a charm. Just one note - stream_response func() always has that store_conversations(), so I commented out scripts related '/memorize'. Kinda redunctant I feel for me. Issuing /recall as prefix to your prompt and without makes this so flexible at times want to chat without using memory (psql). Big Thanks to you once again Ai Austin.

[@j.h.oldman7708]

A saint of the arcane, you have saved me many hours of research

[@OpenAITutor]

AI Austin, Good stuff!!

[@IgorAherne]

Thank you! True masterpiece

[@Powertech1511]

Bro takes time but creates a masterpiece code.

[@BuPhoonBaba]

I absolutely love how your explanation is so perfect.

[@phillipneal8194]

Thank you for a great tutorial. Just what I was looking for.

[@QuizmasterLaw]

God I love this man's voice.

[@michaelandersen9491]

I really liked your vid. I work with this stuff a bunch, and this sparks all kinds of ideas. Thank you for sharing! p.s. Man you deliver that stuff fluff free :o awesome.👍

[@omarfargally7012]

Hi, this tutorial is amazing. I have one question though: in the video you made a function to set the chromadb database for the embeddings but you did not make a function to update that database and sync it with the postgresql server updates. That means that whatever conversations the user has before ending the session will NOT be used in recalling. As I said, the chromadb database is only set before the loop and is not updated during the conversation (even though the sql server is being updated). I made an update function just in case someone wants to do that: def update_vector_db(): conn = connect_db() vector_db = client.get_collection(name='conversations') with conn.cursor(row_factory=dict_row) as cursor: chroma_ids = vector_db.get()['ids'] max_id = max(int(id) for id in chroma_ids) if chroma_ids else 0 cursor.execute('SELECT * FROM conversations WHERE id > %s ORDER BY id', (max_id,)) new_conversations = cursor.fetchall() conn.close() for convo in new_conversations: serialized_convo = f"prompt: {convo['prompt']} response: {convo['response']}" response = ollama.embeddings(model='nomic-embed-text', prompt=serialized_convo) embedding = response['embedding'] vector_db.add( ids=[str(convo['id'])], embeddings=[embedding], documents=[serialized_convo] ) print(f"Added {len(new_conversations)} new conversations to the vector database.")

[@Ai_Austin]

absolutely an option and worth it if you are having conversations extending the context limit. another thing you could add to improve even further: every time you do a retrieval for embedded context, make sure the context is not already in the convo. it will add computational overhead but improve output quality by not giving it duplicated context in one convo

[@EzyRider]

I have no idea why, but my agent answers twice on each promot. I've gone through the code, but can't find any mistakes.. that I can tell (I am NOT good with Python). Still, it works, and it's something for me to practise my Python skills with. Thanks Austin, this was a good mental exercise for me, I appreciate the hard work you put in.

[@theKEVman011]

Same thing happened to me. You may have duplicate "stream_response(prompt=prompt)" at the bottom of the code. One in the nested "if...else" statement, and one after that.

[@michaels.9632]

A tutorial on converting this Agent to have a Voice similar to your livekit video would be amazing if thats even possible

[@QuizmasterLaw]

The speaker's voice is excellent I do think it real not generated because of the occasional "I am so pissed of these people are not getting it"

[@snaxsammy6472]

I'm making an AI system named E.C.H.O. using a ton of ideas from your videos keep it up!!

[@sparkingthebliss]

Saame haha 😂 so much fun

[@SzxyoEditz]

Same here 😂 mine named gemmi

[@tech_informer77]

​@@SzxyoEditz lol u r using Gemini api 💀

[@SzxyoEditz]

@@tech_informer77 ya but still making tools adding in it. Not as the llm used in this video.

[@tech_informer77]

@@SzxyoEditz same actually I am also using Gemini

[@FredyGonzales]

Excelente trabajo maestro, demasiado bueno como para ser verdad. Muchas gracias.

[@OnigoroshiZero]

I just found your channel, great content.

[@Michael-no4oe]

Wow I’ve done some similar stuff to what you are showing here but with different technologies. I must say this looks way more advanced.

[@JoanApita]

/memorize thanks man! It's working perfectly on my side

[@Sulayman.786]

Nice. Thanks!

[@JR-kw6sd]

Bruh. Subscribed. Incredible tutorial. Thank you.

[@luminographix]

loved the video, this was just awesome, can you please make a follow up video providing user interface like OpenWebUI or something custom for UX ?

[@aendnouseforalastname8318]

That would be awesome!

[@yannisvolos5877]

@@aendnouseforalastname8318 I third that!

[@lofigamervibes]

Now, remake this whole video only using emojis. (seriously, thank you for this video, it's a masterpiece)

[@robfielding8566]

local LLMs would totally transform Accessibility. Instead of beating down web developers to make their sites accessible; you can have an agent that can see the screen.

[@codeman99-dev]

Please explain to me how "see the screen" is going to work? That's exactly what screen readers have been trying to solve for the last three decades. The web giving hints to the screen reader allows devs to hide context or show even more context for the situation.

[@robfielding8566]

@@codeman99-dev not just structural stuff that was already done by a person, but ask the computer to describe it via neural network image labelling. this is exactly why Microsoft AI is going to be taking periodic screenshots btw. but yes, that's the whole point. you can ask for a less verbose, or more verbose description; because you didn't try to conscript developers to do all this labelling ahead of time.

[@robfielding8566]

self-operating-computer already does this btw. it uses screenshots, and the screenreader, and an LLM to perform agent tasks for you. it takes over your mouse and keyboard. it needs to take screenshots to use the mouse though.

[@CM-zl2jw]

I’m pretty sure you are reading my mind. 🧠

[@sigil8784]

This is fantastic and is something I've been looking into doing for the past few days. Do you think it'd be possible to do an add-on/followup to this video on how the code could be integrated into Open WebUI?

[@flaviorsnunes]

Valeu!

[@yngeneer]

@Ai Austin > in about 18:00 you are setting the 'system prompt'... Does't Ollama models have it's own? Doesn't it means there will be 2 system prompts passed in the query?

[@Ai_Austin]

no ollama, like the openai api sets the system message with the formatting we showed. the model might have default system messages for their chat app but this is how you set your own.

[@yngeneer]

@@Ai_Austin oohh...sry...now I see.. if use 'show model_name --modelfile' there IS a template builded in the ollama model, but system message is empty... so.. yeah, you were right, thx

[@ATH42069]

great intro

[@cryptorich614]

thank you so much bro. u saved my life. I wonder if we can use this on open webui OR we need to create gui ourself?

[@thelalomorales]

that was a great tutorial !

[@xlr555usa]

Let's do this!

[@Bogdan-AI]

Hi AI Austin! Bogdan AI here. I want to express my gratitude for your video. I set a goal this week to just think about a solution (Agent, GPT, system prompt) to help myself with AI. Last week I already created an Obsidian based knowledgebase that I can chat with (and Fabric patterns to fill it with content). After following along your crash course, I realized that its result can be a very good starting point for creating my own AI assistant. Do you agree, or would you recommend an even better way? In response to your question, I already extended it with persistent ChromaDB and now I am wondering about what is best: persistent or in-memory db. I don't have a lot of data but persistent seems to load faster. And of course, I see that it can be extended with additional tools (google search, add knowledge from files, real-time separate database tables for tracking weight and other parameters, etc.)

[@yngeneer]

educational! learning! subscribed!

[@Agia]

I have been playing with Ollama and Open Webui, How difficult would it be to implement this into a Open Webui accessible setup?

[@trilogen]

I believe the whole point of new Orwellian Microsoft Recall feature (No one asked for!) is to capture these local data/conversations with your own local AI. You won't be able to disable it they will repair any file you remove on the next update. Best to just switch to Linux like Zorin OS (Closest to Windows like interface)

[@theKEVman011]

Switched to Linux 2 weeks ago because of the 24H2 update. I had all the spyware deleted and it came back.

[@trilogen]

@@theKEVman011 into Linux?

[@UKnowIfUKnow]

You have no idea the hell i went thru vectorizing my data a year ago... Thank you for invalidating a week of my life. 😅

[@Ai_Austin]

those weeks we spend struggling and failing, is what is required to get better at developing. wasn't a wasted week i promise lol

[@alibahrami6810]

subscirbed for your tone about we are products. thanks!

[@nradzy24]

Austin, I was wondering if this code could somehow be adapted to use in OpenWebui, as a function or in a pipeline? I am not asking you to write that code just to get your opinion on how easy or difficult it would be for someone with a beginner level experience coding. Thanks for your time and have a great holiday.

[@i2c_jason]

This is so valuable. I'm considering your subscription based on the quality of this video! I have a dumb beginner question, but could this RAG module plug into LangChain if I wanted to architect a more complex agentic workflow? I'll be parsing a lot of engineering requirements into various parameters that will make their way to various mathematical and geometrical outputs, so I'm envisioning your example as a great way to manage the main assistant, and then different flavors of this as expert agents in various disciplines within my LangChain graph. Not looking to do any finetiningof models, as we are a small team and I anticipate the LLM capabilities to continue to grow, so I'd like to future-proof my design wiht a lot of multi shot learning near my system's final outputs. -I2C_Jason

[@dethchyl2824]

Anyone have any idea on how to implement something like this under Open WebUI?

[@Otiyyy]

could you make a video on how to do this with langchain aswell? since its more in depth and some people would want to include tool usage

[@osamashurrab3277]

very helpful video ,i have small issue i steup everything as you did in the video but when i share information to ai and than close the code it don't remember the information that i share ,can you help me with this one

[@mikebeamlight8334]

Nice work! Will it handle feeding it with 60 gigs of PDFs?

[@unkillablespirits8582]

can i select this in home assistant , can we use ollama create or something to make this a model for home assistant ?

[@September222036]

Bro thanks so much for this, created my first self-hosted agent in less than 48h. The only issue I have if anyone can help, is that everytime I exit or reset the convo, it starts over again and considers only the system messages defined in python, not /memorized initel, or past convos. Any tips to make it like a continuous convo or to be able to recall actual context from /memorized pool would be appreciated

[@Sulayman.786]

Would be good to see it working, and what you can do with it

[@GurlalSingh-b8c]

can we add feature in this to upload documents into it, and it will learn from it and response from documents as well ?

[@UKnowIfUKnow]

Look into libraries like pymupdf. SUPER basic and untested: import fitz def get_text(filepath: str) -> str: with fitz.open(filepath) as doc: text = "" for page in doc: text += page.getText().strip() return text Pass text to memorize. Or a text file with open('text.txt','r') as file: text = " ".join(line.rstrip() for line in file) Passing to memorize again.

[@JR-kw6sd]

This. Although, even with this implementation, you can achieve it by simply copying and pasting the document text.

[@themax2go]

integrating the bspoke llm would also be cool, though i'm sitll figuring out how to do that (AKA avoiding LLM hallucinations)

[@antonpictures]

RAG missing image embedding , text is not enough.

[@itsjoker1990]

lol i added a web frontend using flask and feww button as well as textual command for the /recall /memorize and / forget

[@memegazer]

Could you use this to fine tune an LLM on your games lore and then produce npc output?

[@VideoSafari7]

at 24 :13 you're saying we've got some working code that does most of the things. It does look really good, and every indication is that this works really well, up to about this point. I mean REALLY well. There is a bug there, however. The method 'retrieve_embeddings' has a call signature 'queries, results_per_query', and you have had us use simply 'prompt' when calling this method from the 'recall' method. This results in a type error against the unexpected argument 'queries', and so brings the effort to a halt. EDIT: It was only hard for somebody who'd been chasing a cursor through a youtube video for several hours ;) It might even have been an error I made, after chasing a cursor through... Claude 3.5 Sonnet helped me hump it over the last hurdle. It seems to be working, now to add the frills at the end. I'll tackle that after a bit of a break.

[@rakshit0_0]

did you get a solution for this? currently my code saves it into the db but is unable to recall it

[@VideoSafari7]

@@rakshit0_0 I am currently testing. It does appear, though, as if It saves and cannot retrieve contextual elements.

[@Ai_Austin]

i'm thinking something was missed. because at 24:20, you should see that queries is created on line 150. on line 151 we send the queries as the single input parameter to retrieve embeddings. results_per_query is pre defined to 2 in my code. so not sure what was missed but can say something was there. try to understand your misinterpretations of python syntax in regards to the code, most likely the source of confusion. don't just type what i type, understand what the code in every line is instructing the program to do. it's like reading a book with big words, if you do not stop to use a dictionary, you'll look back and realize you didn't understand what you read.

[@VideoSafari7]

@@Ai_Austin I do much better understanding what I've learned from working tutorial code. Right now, I'm catching this Traceback at 1Hr5Mins in: USER: It allows you to remember what I'm on about, and what you've said. Ideally, across restarts and powerdowns. Tell me what I told you about ^Kurt Godel, even if I was incorrect. Vector database queries: [''] Traceback (most recent call last): File "/home/goon/python.prj/rawdog/ragged/assistant.py", line 187, in recall(prompt=prompt) File "/home/goon/python.prj/rawdog/ragged/assistant.py", line 175, in recall embeddings = retrieve_embeddings(queries) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/goon/python.prj/rawdog/ragged/assistant.py", line 112, in retrieve_embeddings results = vector_db.query(query_embeddings=[query_embedding], n_results=results_per_query) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/goon/python.prj/rawdog/ragged/ragged-python-venv-3.11/lib/python3.11/site-packages/chromadb/api/models/Collection.py", line 300, in query validate_embeddings( File "/home/goon/python.prj/rawdog/ragged/ragged-python-venv-3.11/lib/python3.11/site-packages/chromadb/api/types.py", line 494, in validate_embeddings raise ValueError( ValueError: Expected each embedding in the embeddings to be a non-empty list, got empty embedding at pos 0 python.prj/rawdog/ragged via 🐍 v3.11.2 (ragged-python-venv-3.11) took 1h5m20s EDIT: AI Austin No doubt something was missed, I just don't have a good way to go back and check it; video is a pisspoor medium for that sort of instruction.

[@VideoSafari7]

@@Ai_Austin I'm wondering where my last reply went: it showed the exception currently being generated by the code.

[@johnkoeck3702]

Do you have a one time payment option for pro memberships?

[@Ai_Austin]

i do not currently

[@themax2go]

pleases add at least yearly membership option that gives x% off - I'd rather pay yearly and it'd give you a much more money instantly

[@Ai_Austin]

already got yearly membership plans!

[@jamiemaidan8333]

Man this video is amazing. I'm an absolute noob and this is understandable and I'm confident this is how I'm gonna get it to work.

[@Vesuvius-f4h]

can you do that with deepseekv3?

[@Aristocle]

this could be add as a model in the vscode extension Continue?

[@potaz02]

Why im stuck in "say Jarvis followed your prompt", no message for eror, but my jarvis can't answer me

[@themax2go]

works fine. you probably have a typo, or you're using a llm not suitable for your (pc) hardware

[@deltagamma1442]

Aren't RAGs limited to best match? Are there alternatives you're aware about?

[@Ai_Austin]

not this agent. watch

[@RetiredVet]

I had issues running the code. It would complain about json not being able to handle a python set() I changed to lists and the code is working great. Is the use of the python set() do something that a list does not do?

[@Ai_Austin]

sets are faster and use less memory. they cannot have duplicates added either. when sets can be used, it's typically ideal to use them. a list is fine but not sure why it would tell you json cannot handle a python set. it is syntactically the same in python and json

[@RetiredVet]

@@Ai_Austin That is what thought. I just kept getting error messages about json not being able to handle sets. I assume it is working for you. I am going to change the branch and see about going back to sets.

[@RetiredVet]

@@Ai_Austin The set problem, I reported, was because I made an error in the create_queries() function. I program on an older Linux laptop, the screen was split between the browser and vscode and I used a dict for query_message. After finding this error, your code worked fine. I am surprised it worked at all, because the query_message became a list.

[@AustinKang-wk8cl]

is Ai Austin dead? It's been two months since his last video

[@Plash14]

Yup he dead

[@6Abdellah9]

amazing but just out of curiosity wouldnt langchain make this easier

[@Ai_Austin]

you could add langchain to this program. but idk how langchain could possibly make the program simpler. if you don't have an actual need for langchain, don't use it. you still need chromadb, ollama, postgres and every other library, you could simply add langchain if there is a feature in langchain you need to build out this program to be more complex than mine

[@6Abdellah9]

I see thank you

[@phillipneal8194]

Oops , another problem This line: serialized_convo = f''prompt: {c['prompt'] } response: {c['response']}' throws SyntaxError: f-string: unmatched '[' So I changed the line to: serialized_convo = f"prompt: {c['prompt']} response: {c['response']}" Notice the double quotes.

[@loganwilliams4958]

Can we use this and a RAG system together? I’m a little new to AI

[@loganwilliams4958]

Never mind I just realized that this is a RAG system🤦‍♂️. I guess what I’m really asking is if there is a way to give the ai access to word documents, graphs, and other text files. Or do I need to just copy and paste the documents into the prompt section and do the /memorize command?

[@robertohluna]

Perfect

[@vsubbakrishna]

Great work!!Does membership give us access to the code

[@Ai_Austin]

yep i released the source code a written tutorial over 3 weeks ago for all PRO members. same with all of my videos. even for videos that are not tutorials and simply demoes which i will make moving forward, i share my source code in there for anyone willing to help me make these videos

[@samo-zaposao8757]

Why you didnt make program that do all that for us.With purchasing license for Windows.I wish it

[@F336]

nice where can i git clone this code?

[@Ai_Austin]

thanks. you can't. i'm not an open source dev at this point in my career. i make these free youtube videos for people that want to follow a video tutorial. if you just want to yoink my code and not watch the vid that took me 100 hours to make, that shows you how to write the code, PRO members can access the source code and written tutorials for my videos in discord.

[@manifestasisanubari]

I'm thinking of doing this project in Google Colab. How do you recommend me regarding the psycopg and setting up the database? If I connect to a local runtime, it will not use Google's hardware, right?

[@obentti]

Has anyone actually implemented the logic in this video and managed to get it to work?

[@MatrixMoney-uo6pk]

they just released llama 3.2 :)

[@Username56291]

Can you share code please?

[@VideoSafari7]

I am profoundly impressed with this body of work. I would also likely buy into your source code access at the rate you request. But your centralization on a discord server for well, anything, betrays your actual complete disdain for privacy; yours, your users, anyone's. Discord is a pox. It is bloated, opaque to the user and transparent to the operator (Discord Inc or whoever runs that crap). I'll cheerfully follow your video; but under no circumstances will I have anything to do with discord.

[@Ai_Austin]

appreciate the support. the intention isn't for people to share private information in my discord and then the concern isn't existent for the users. discord is probably less after your data than the owner of this platform we are communicating on. it's simply a place to make initial contact, discuss non private information and me to platform the links to my source code links (which the source code is on notion, not discord). it's totally possible to set up an anonymous discord account, never send messages and use the account to access information for yourself. but if discord is your enemy, and creating an account with them is against your principles, respect and thanks for supporting the youtube channel otherwise

[@lokeshart3340]

Is this really unlimited memory?

[@Ai_Austin]

SQL databases can be scaled in theory infinitely. You obviously need the hard drive space. But you could probably have a couple decades of conversation with 250gb depending on your usage. Once you have over a million rows, maybe even 100k rows would be a good number to look into horizontally scaling the database to improve database latency for a large scale SQL db

[@lokeshart3340]

@@Ai_Austin hmm like we can do like at the end of conversation our ai will delete the oldest non relevant conv as by its own Intelligence then it's literally unlimited .

[@garchafpv]

y u modulating your voice.. or is it ai?

[@Ai_Austin]

i use a voice isolator and condenser on my natural voice to give the clearest audio possible given my equipment (in this videos case). i cut silence to make the video's not waste peoples time 🤔 y, did u want to hear my fan running and my dog scratching herself?

[@themax2go]

@@Ai_Austin 😆

[@garchafpv]

@@Ai_Austin that depends...... Is your dog hot?

[@Plash14]

Hasnt been uploading for 2 months u ded?

[@Ai_Austin]

born again

[@Johndoe-176]

stop typing clear, just press ctrl+l

[@umarfarid478]

Most of us are addicted to old commands

[@angloland4539]

❤

[@codeman99-dev]

Have you heard of Docker? A simple docker-compose would have saved you a lot of pain.

[@Ai_Austin]

what pain would that have saved me as the instructor? i would have to explain a complex software tool designed to make software scalable and able to be deployed faster. it adds complexity in development, and the users will only deploy it on one local pc. i would never use docker, except for developing production software that i need to be able to deploy on 100 servers fast. since this isn't one of those situations, i don't want to confuse people by telling them to do things more complex than i would. docker reduces complexity of deployment on production servers, while increasing the complexity of development. this is 100% a development tutorial and does not require deployment, making docker practically valueless in this project. (sure you could subjectively like the docker flow, now that you have adopted it, but it was a learning curve that did add complexity to your local dev flow)

[@NoahStacy-k4g]

got this to run, but boy is llama3 INSANELY slow. like so slow it's useless.

[@Ai_Austin]

it's 100% a testament to your pc's hardware. it's like pulling out moms old laptop and saying "this new advanced graphics video game is slow" chatgpt is not fast. closedai has dozens of billions of dollars worth of GPU's running in a microsoft server to make chatgpt fast. hardware limitations are real things when we get into running Ai locally

[@themax2go]

just use a smaller llm, such as qwen2.5 7b or the even smaller one if you must

[@IAmMisterD]

could have included the code in your description 👎

[@Ai_Austin]

i could but i won't work free. mcdonalds doesn't expect people to work for free. weird that anyone expects someone who spent 8 years learning to code to give their work away for free. if you don't want to support the channel by watching the video, then i have a PRO membership where my source code is released weeks before the videos. this is youtube, where they want me to create content that you watch to the end. if your time is too valuable to support my channel by watching the video. then you should be able to afford $25 a month to get code from a professional software engineer that has companies which will pay him to write code 👍

[@xAgentVFX]

All AGi needs is memory and multimodality imo. So are you saying that the "experts" are making it sound harder than it really is?

[@orotoi1]

nah.. you just make it sound easy :P (needs more stuff.. like reasoning for example -unless you include that in multimodality..)

[@xAgentVFX]

@@orotoi1 So what else in your opinion? Because I think it already can reason...

[@orotoi1]

@@xAgentVFX Nope, we are on the first small steps towars it or maybe on the 'emerging' period of it. But not there.. Cant reason yet, at least in human levels. And same goes for other cognitive tasks too.. It must reach in performance at least the >50% of human skills to say that 'it started'.. and will have to outperform 99.99...9% of humans to say that is completed (and then goes to Super AGI or whatever we ll call it then).

[@xAgentVFX]

@@orotoi1 Huh? How can Sonnet 3.5 be able to code and yet you say it cant reason? It isnt just predicting the next word, the attention/context vector network is decoding 'meaning'.

[@orotoi1]

@@xAgentVFX It is better at coding than other models cause it trained better on that.. Having larger window or and token input/output doesnt mean "reasoning". Neither we have reached the 50% of human skill on all tasks.. And there are many other cognitive (mental) tasks yet to reach.. Anyway.. is not something I say out of my mind.. almost all researchers on AI say we re not on AGI yet..

[@bel.solutions]

Nice video I like the details you went though on describing RAG! I made a video on my channel like this and would appreciate feed back if you have time!

[@mohammedissam3651]

This AI doing is just saving my chat history ? Why , what is the application of using it The purpose of its exsiatance !! Could it solve problems I don't do chat on my system 😅 Can i teach it ?

[@jordanlane68]

Are you real?

[@Ai_Austin]

no

[@MungeParty]

So messy

[@Ai_Austin]

show us how to do it cleaner or cap

[@themax2go]

@Ai_Austin btw have you "evolved" this more? also, i'm experimenting with "sciphy/triplex"... any idea if that helps with the ai to derive context faster / more accurately?

[@phillipneal8194]

Oops. Chromadb has a problem. The tail of the error message is File "/home/pi/local_rag_agent/lib/python3.11/site-packages/chromadb/api/types.py", line 4, in np.float_ = np.flo0at64 ^^^^^^^^^^^ File "/home/pi/local_rag_agent/lib/python3.11/site-packages/numpy/__init__.py", line 414, in __getattr__ raise AttributeError("module {!r} has no attribute " AttributeError: module 'numpy' has no attribute 'flo0at64'. Did you mean: 'float64'? So I burrowed down into File "/home/pi/local_rag_agent/lib/python3.11/site-packages/chromadb/api/types.py", line 4, in And just after import numpy as np I stuck in np.float_ = np.float64 That cleared up that problem

[@Otiyyy]

for the people that are having trouble granting privileges here are teh command syou need to run austin thi si salso for you would be great if you put all the commands in your description GRANT ALL PRIVILEGES ON SCHEMA public TO example_user; GRANT ALL PRIVILEGES ON DATABASE memory_agent TO example_user;

[@philtoa334]