This 100% automatic AI Agent can do anything, just watch

Learn how to make $$ with AI - www.skool.com/...
I'M HIRING! Do you want to join my team? Apply here: forms.gle/2iz4...
Agent Zero GitHub: github.com/frd... (give it a star!)
Jan's Demo: • Agent Zero 🤖 first dem...
My Docker Video: • Build AI Agents with D...
Follow me on Instagram - / thedavit
Follow me on Twitter - x.com/DavidOnd...
Please Subscribe.
Credits: ‪@AgentZeroFW‬
David Ondrej Podcast - Episode 15

[@DavidOndrej]

🔥 Learn how to make $$ with AI Agents - www.skool.com/new-society

[@flyingcheesecake3725]

david, can someone with little to no knowledge in programming get value from your course? i watched the video halfway and got interested in it (i joined, giving it a try at least)

[@eismccc]

Are you still trying to hire someone?

[@emmanuelarthur8377]

The fact that it can feed errors in the terminal back into itself and resolve them is amazing!

[@photamasan9661]

YES. open interpreter does a similar thing and while the OS mode is great the limitations are too many.

[@MyStoryCrafters-co]

I’ve been working on making almost the exact same thing. User intervention is genius and I don’t think I would have ever thought of it. Great work here.

[@photamasan9661]

“agent one has no idea that his user is actually another agent and he doesn't need to know” - bro that’s it we’re just one of the many sub agents and we’ve just made our sub agents .

[@JarkkoHautakorpi]

We are in simulation using agent-zero but we don need to know it

[@AeroMagic_Official]

bro u just broke my brainnnnnn....

[@solank7620]

@@JarkkoHautakorpiMatrixception.

[@KCM25NJL]

I like the idea of generating function call code dynamically, especially for generic purposes such as 'File Open,' 'File Save,' or specific algorithms like 'Sort Algorithm X.' To optimize efficiency and promote reusability, I strongly recommend cloudifying these functions, either in a private or public manner. Publicly available functions could be curated to ensure quality and security. This approach can potentially reduce the number of LLM calls required, saving time and money.

[@TheAlastairBrown]

Billions of agents hammering away writing functions to pull down the current time and date 🤣. I think the first thing I'll ask Agent Zero to do is come up with a list of 50 functions and build a library of general purpose capabilities.

[@KCM25NJL]

@@TheAlastairBrown Yep, I can well imagine running AZ on Groq to rapidly build a heavy duty RAG vector DB or Knowledge Graph of reusable function calls. I'd probably go with the KG just for the sake of storing clusters of relatable functions based on use case. Frankly with all the big brains out there, I can see no reason why this hasn't been done yet.

[@DaleIsWigging]

With ai search engines getting better maybe this will exist in the future with "currently working" github repositories

[@user-qg8qc5qb9r]

Introduction to Agent Zero and Interview - 00:00:00 Overview and Key Features of Agent Zero - 00:00:29 Demonstration: Current Time in Different Locations - 00:01:28 Error Handling and Agent Self-Correction - 00:02:00 Choosing Language Models and Cost Efficiency - 00:08:01 Compatibility with Various Models and Frameworks - 00:09:44 Setup and Configuration of Agent Zero - 00:11:01 The Role of Docker in Agent Zero - 00:24:22 Advanced Features: Subordinate Agents and Task Delegation - 00:20:52 Interactive Agent Management and Error Recovery - 00:46:41 Memory and Knowledge Tool Usage - 00:32:56 Combining Agent Zero with Other Frameworks - 00:38:28 Building and Improving Agent Zero - 00:58:39 Future Development and Open Source Collaboration - 01:09:44 Final Thoughts and Closing Remarks - 01:18:45

[@AeroMagic_Official]

thanks 4 this ur a real one fr

[@StuartJ]

I was waiting for something like this. Agents that you have to define yourself, seemed counterintuitive to me. AI should be able to figure it out for itself.

[@LVE-xr5or]

@@Tom_Neverwinterthat’s what I was thinking too

[@MungeParty]

Yeah I don't think this is doing anything new relative to AutoGPT.

[@roygatz]

@@MungeParty popular framework are taking a top down approach and there are many things that render the frameworks too complicated to use. I think what agent0 is trying to do is to take a bottom up approach and define a set path to at least do something productive or hopefully userful. Though nothing production worthy, a good starting point.

[@orthodox_gentleman]

I have thought the same exact thing dude

[@MrJoe1199]

Very well thought out way to "do" agents, and complete with excellent pragmatic features for controlling agents and helping agents to recover from errors. The best feature IMHO is that all the behaviors are in external files and not in the code.

[@karlbarker2912]

I’ve used open interpreter and now Claude engineer, but this looks amazing, can’t wait to try it

[@photamasan9661]

Yeah I’ve been using open interpreter for a while now and it helps a lot but troubleshooting is so annoying because it either sucks in many other modes or it won’t stop looping on OS mode . But this, this is something pretty much next level.

[@stonedoubt]

Open Interpreter does something like this. It can write code, execute it, debug if there is a problem, etc.

[@justinduveen3815]

David, thank you so much for helping us learn how to harness the power of AI agents. I’ve been a big fan for a long time and have found your content to be super helpful for AI automation builders. Jan, your new tool-creating, AI agent framework is really smart! Thank you so much for sharing!

[@krisvq]

I've built a broader version of something like this 8 months ago. More evolved since then. The bots are not smart enough to be actually functionally usable yet but the principle is simple and good. My framework allows the bot to create, use and query databases and build a code base of functions that perform various tasks. The idea is that you give the bot access to read, write and execute and give it behavioral parameters. Them evolve the process. I even have it write a clone of it's own code base clone and test it, the back itself up and run a self update. I got bored writing it after a while...

[@shivachaturvedhi8840]

Interesting... another video on 1. creating custom tools 2. custom .md files 3. generating and running code in other languages (other than python) 4. ability to develop some design pattern using some tool etc could be er useful

[@CarolinaFarming]

Why don't you have a prompt that Agent 0 checks that lists the best libraries to use for certain tasks? So it will use the correct library from the beginning to download youtube videos.

[@FriscoFatseas]

Excited for when its simple enough for me to care

[@CreativeEngineering_]

The whole name thing is easy, with explaining how to use the memory you would add something like 'If a user's request or statement contradicts a previously made statement or request then ask the user to explain the reason for the contradiction. Something along those lines, I'm sure it could be refined more, but that's the gist of it.

[@BadreddineOtky]

I have been testing lots of AI frameworks, tools and platforms. This is the BEST ONE I have seen! Great project!!! Thank you so much for sharing. This will help me so much in my everyday work

[@nanolaxyz8490]

better than Aider?

[@BadreddineOtky]

@@nanolaxyz8490 yes definitely better. the auto-sanbox and auto-correct features are awesome

[@AeroMagic_Official]

@@nanolaxyz8490 what dat

[@timmcgirl5588]

Honestly, this can achieved using Open Interpreter, by simply giving it a prompt to integrate with any open source tool for free.

[@user-qp1jq3eh3e]

open interpreter will take you into more loops than GPT 3.5 - I trusted the bastard snif ... but I like the idea of challenging all ideas so the last one stands

[@timmcgirl5588]

@@user-qp1jq3eh3e i use Claude 3.5 Sonnet, it never loops when i use it do to proper prompt engineering.

[@timmcgirl5588]

@user-qp1jq3eh3e it never loops when I use it, I use Claude 3.5 Sonnet with Open Interpreter, can use Groq inference with Llama 3.1 also. Not sure we're on the same page, proper prompt engineering is also key.

[@karlbarker2912]

Open interpreter does not have persistent memory

[@timmcgirl5588]

@@karlbarker2912 By integrating Open Interpreter with Gorilla AI and n8n, you can create a system that maintains persistent memory across sessions. This approach leverages n8n’s automation capabilities to ensure that user interactions are contextually aware and persistent, enhancing the overall user experience. The Gorilla AI API platform offers various utilities and capabilities that enhance the functionality and application of AI models.

[@Derick99]

Whats the time stamp for training 3.1 llama finetuning with your data

[@reinerheiner1148]

The price will double in 5 days!!! This marketing trick instantly made me suspicious and dropped my motivation to keep watching this video down to zero. But hey, I am sure you will get enough people who fall for this.

[@orthodox_gentleman]

I thought the same thing

[@iGhostr]

Same

[@riegaldutoit5792]

thought the same.. raising the price on new information has become his thing lately.

[@remain___]

🤨 why? You think discounts are cause for suspicion?

[@Hyp3rSon1X]

The best way to make money is selling people a guide on how to make money. The guide doesn't need to work and in most cases it won't. But as the creator of the guide you can then simply blame it on the incompetence of the user. Adding the factor of "fear of missing out" only adds in sales of the guide. So yes, I absolutely agree with this!

[@LVE-xr5or]

39:00 where it gettes good

[@KOSMIKFEADRECORDS]

what the self promo? LOL sneaky.

[@CaptGhost-u3u]

that is so amazing, I'm a student and I'm thinking, with all this LLMs that can run locally if i can build a bot like this, and now its here. truly amazing

[@supernaturalself3577]

how would on go about using agent zero in a persistent way? For example if you didn't want to lose the libraries everytime docker instance shutdown. I would imagine I would want to keep the environment intact so that one doesn't have to waste time reinstalling something every time one fires up docker.

[@JarkkoHautakorpi]

You can persist your running docker instance. Stop it and resume it where u left it.

[@cbau0809]

This agentic approach is amazing. My only recommendation is to use pydantic to instruct the llm into a structure response rather than using a parse json function.

[@jeffsteyn7174]

Json is structured

[@MungeParty]

They address this in the video, it doesn't use structured output specifically to retain compatibility with models that don't support it.

[@janalgos]

now just need to incorporate mixture of models to improve quality from the small models

[@TheAlastairBrown]

Ask it to improve itself ?

[@kniiz]

no wonder strawberry got so hype, its because of this. im in love. nice agent! Thank you both for an nice interview.

[@DataJuggler]

Congrats on the 100K subscribers. Ask Jan in his spare time to write a C# version of Agent Zero.

[@evynt9512]

Agent technology will define nature.

[@lifeonearth9261]

Super cool project. Had an issue with the database but found a workaround in Issues that fixed it. Thanks for this great project!

40:04 - Instalation / setup

[@tier1recon836]

Any tutorials using it with ollama?

[@whizadree]

Think a web gui using flask would be nice addition , that can setup the llms add api keys , and enter your request and get either the final output or see the entire output

[@stanTrX]

looks good. do i have to let this framework to "find a way to tell the time" each time i reopen this program? or it remembers and saves required stuff like this to it's knowledge base?

[@baTonkaTruck]

The problem with tools like this: Most people don’t know enough to use them, and the ones that do don’t need them.

[@Derick99]

This worked the first time now anytime you use it, it tells me its just a language model and cant do anything but it originally worked and would do research etc

[@emmanuelkolawole6720]

When will agent zero support automatic docker container restart in cases where the container created on startup is corrupted by llm mistake. I don't want to restart the whole application again and again. The agent should just launch a new docker container

[@resresres1]

This is a nice proof of concept, but LLM's at the moment can only handle fairly basic coding. Perhaps when LLM's can do much more complicated projects, complete on their own, then this concept of having it fix it's own code will be absolutely necessary.

[@CharlesZerner]

Awesome! Can’t wait to try it.

[@joesmoo9254]

Great framework. Is there a way to feed context to the system. Fi a directive or architecture directory where you put bussiness rules and standards

[@maxxflyer]

amazed, I work on a similar project but the docker shit is amazing.

[@florianstephan5745]

wow, an AI finally telling me the time...AGI is near ;-)

[@TJ-hs1qm]

TZ=Asia/Tokyo date 😂

[@MeinDeutschkurs]

I hope that perplexicA will get an API soon.

[@kellymaxwell8468]

But can it program and script and do games bring back old closed down games, remake old games, and make more of games that have gone

[@16bitbrawler]

This is really amazing. Thank you so much for sharing!

[@Dmytro-fr9nd]

What kind of hardware it requires to run?

[@jean-philippes5826]

random pc no need graphic card like big local llm

[@r3dsign]

I watched 3 minutes of the video and Im sure that this is the idea that wil make AIG possible.

[@thenextbigthing1393]

AIG?

[@thomasnook118]

Title should’ve been “Agent can tell you the current time in Berlin VIA your terminal… just watch” LOL but all jokes aside, these ai/agents are still just toys

[@JakobN-zg1st]

Why do you believe ai/agents are toys?

[@thomasnook118]

@@JakobN-zg1st no real world applications yet (meaningful applications) . I think one day they COULD be but for now they aren’t very useful.

[@christopher-george]

Truly amazing !

[@supernaturalself3577]

can I get agent zero to change itself to use claud instead of perplexity?

[@KOSMIKFEADRECORDS]

Promising will be following closely. If i had the bucks id be in your discord.

[@idkwhy4939]

Facts

[@GrahamSmart]

Absolutely amazing work.

[@SIRENSCINEMATIC-SCOTLAND]

So much to take in here incredible that this is the free stuff 😂 I'm a paid member and recommend

[@Jason-oi8hz]

I've been waiting for something like this... I'm excited

[@AD4K8]

This is incredible brothers!

[@jamesspalding1210]

This is such an excellent video, amazing framework and can’t ait to test it out

[@RhumpleOriginal]

If this guy did this, imagine what governments have going on behind closed doors...

[@DavidOndrej]

true

[@mircorichter1375]

Follow Mike Benz Cyber on X... He put it all out

[@Jim1Dean]

how can I use my local llama3.1 model? I use textgen webui with OPENAI API

[@DreadFox_official]

Well dang, I wanted to build an agent like this for my cybersecurity framework but you already made it. I'm excited to experiment with this. Thank you so much for spending time to build this.

[@DreadFox_official]

@@Tom_Neverwinter This project is pretty big, I can see how it took multiple minds to come together.

[@HistorIAsImposibles776AC]

Wow amazing, let's get it working !!! Is it possible to Collaborate with the project ?

[@RowanSheridan]

Good overview, would be helpful to have timestamps

[@thomasthelen7874]

How f*** awesome!!! Kudos to Jan!!!

[@mountshasta2002]

This is pretty cool. Currently trying to have it utilize crewai and other than using the wrong pip command we're cruising. If I want to provide context or a codebase can I just put it into the work_dir? Like if I wanted to provide docs. EDIT: adding docs to the work dir causes token limits to max and timeouts.

[@igorbetternower7800]

Sounds like the future of AI, impressing, but too high for me. I guess I'm one of the first to be substituted by AI 🙂

[@jasonsharp8621]

I'd like to see more indepth of a tasking for the agent.

[@nAme-bf9uz]

This looks like a more complicated version of open-interpreter

[@aaagaming2023]

Neni to nahodou Hynek Cermak? Uzasna prace, take jsi s tim pohraju.

[@hypertectonics7009]

So conceptually this system should be able to improve its own code and execute it...

[@leomaxwell972]

Interesting thought... I can see, and I've had, AI forget/mix-up my or other's names, but, when it comes to their own name or company, they have no problem remembering exact details... obviously this is a (un)holy grail of bias, that they would be super hard coded... but... how limited is this "hardcoded" space? surely it could fit the user's name for their model/iteration/shard/etc. of their AI conversation partner, no? or maybe it really is that limited lol. Thats semi-blackbox level stuff if you ask me.

[@rots.866]

can i use this to complex project with server and client side?

[@johnstarr2001]

I can't get the API. The site won't except my credit card? I can't go further!?

[@johnstarr2001]

Always something wrong...

[@AeroMagic_Official]

THIS I SOMETHING UNERRRREAL

[@johnmarkston4883]

any discount codes for the skool ? i was part of your skool about a year ago but didnt have enough time to dedicate to it, $197 is a weeks pay for me and i couldnt afford it while also paying for all the api fees to build an agent

[@jimg8296]

Very cool project

[@MeinDeutschkurs]

Intriguing!

[@ccdj35]

I cannot make the Nvidia drivers installed in Ubuntu otherwise I would have definitely used this kind of agent.

[@whizadree]

Can it create a structure file system and able to edit individual files…

[@atz6212]

So basically it can build a coding stack for you!? And basically any other coding stack!?

[@jessicalewisjessielew3399]

1:05:00 what is your custom GPT for coding ?

[@photamasan9661]

IT HAD TO BE TOO GOOD TO BE TRUE. its the most frustrating agent ive dealt with. if i dont type fast enough my message gets cut off. it refused to start in a docker container time and time again. and it claimed it couldnt access the internet until i told it about the knowledge tool. it then proceeded to use the knowledge tool to find information about the knowledge tool and came back saying nah never heard of it. when i asked it to scrape a website it said knowledge tool cant do that . i was raging. but im sure with time itll be great but OHLORD I WANT TO PUNCH CODE RN

[@DavidSavage-q1c]

Same issues here! It's like an AI in its bratty or edgy teen phase. Do this for me No. I can't. *Does the thing* YOU SEE! I Have no clue what you're talking bout.😒

[@thewebstylist]

Ahhhgh wish bro you had a 15 min version of this video

[@six1free]

I've been writing similar... project abandoned :D (I assume) embedding_llm = models.get_embedding_ollama() ??? can we specify the model inside the brackets like in get_ollama(X)?

[@peterbellheim720]

Thank you David ... I am big fan about your Workflow...👌👍🤝

[@FBICIA-j8j]

Wow!

[@doghashisday4612]

Llama 3.1 integration ?

[@dailyrd]

im so dumb i can't even install this without asking to llm

[@DreadFox_official]

Software Driver/ Hardware = Human Driver/ Ai Model

[@johnstarr2001]

Why not build docker a person can download.

[@BigBandoonthebeat]

This like open interpreter or soc r sum

[@AliAbassi1]

Exact same thought, very similar to open interpreter

[@BigBandoonthebeat]

@@AliAbassi1 Self operating computer tuff shouldve copied that

[@hackmedia7755]

does this integrate into IDEs like Xcode?

[@jimmyjam77]

How does this compare to Aider?

[@GmanBB]

How I add files/documents to the agent UI?

[@TJ-hs1qm]

27:52

[@felipeblin8616]

Amazing

[@d.d.z.]

Another useful tool. Thank you

[@MatichekYoutube]

would be possible to use ollama 3.1?

[@AgentZeroFW]

Yes, but it's not very good from what I've heard from users. Maybe try gemma2 instead with ollama.

[@sammcj2000]

Requires perplexity? That’s… odd

[@tareqkhammash9245]

U should have more subscribers

[@miroslavacizmarova6809]

Wow. Great job!

[@tkbwonderworld]

Has anyone run this agent on Windows? Please help me with using it.

[@TisforTech]

It's crap, doesn't work with local llm...

[@dreamphoenix]

Thank you!