Відео

Thank you 🙌

When working with thousands of documents in a vector database for RAG, a few optimizations can make a big difference in managing the data efficiently. Semantic search, for example, can become computationally intensive with a large dataset. To enhance performance, consider optimized algorithms like HNSW or even hybrid search methods that combine keyword search with embeddings. These are just starting points, and we can certainly explore additional ways to optimize based on your specific use case. Feel free to join our community for a deeper discussion community.phidata.com/ |discord.gg/Ye8rQbaT

When working with thousands of documents in a vector database for RAG, a few optimizations can make a big difference in managing the data efficiently. Semantic search, for example, can become computationally intensive with a large dataset. To enhance performance, consider optimized algorithms like HNSW (Hierarchical Navigable Small World) or even hybrid search methods that combine keyword search with embeddings. These are just starting points, and we can certainly explore additional ways to optimize based on your specific use case.

We are on it !!

Thank you 😊

Hello! Yes, we do support Gemini. You can find more details in our docs here: docs.phidata.com/examples/provider/gemini

@@phidata Thanq

You're welcome!

thank you 🧡

Glad you like them!

thank you :)

By local i mean the agent sessions and memory is local. You can also use local models via ollama so in that case even your inputs arent sent anywhere -- which will be 100% local. For this demo i used gpt-4o just because it works better.

Great suggestion! Thanks a lot 🙌

Our pleasure!

Yes. They have a complete guide.

Yes, you can definitely run local LLM models like Llama 3.1! Phidata isn't limited to OpenAI-you can use smaller models locally, and we've got a video on our UA-cam channel explaining how. ua-cam.com/video/d-Kh0SvgB6k/v-deo.html You can also check out our docs on how to run Ollama: docs.phidata.com/examples/provider/ollama

hi @renesis888 - localhost:7777 will be serving the playground server, which you can view the api docs for at localhost:7777/docs To interact with the agents, please go to phidata.app/playground and select localhost:7777 as the endpoint The phidata.app/playground will reach the agents running on locahost:7777 let me know if you need help or drop by our discord :)

Hey @donFrankAlvarez! Yes, Phidata allows you to have different agents collaborate on a task. We call it "Teams". docs.phidata.com/agents/teams

Yes. Multiple agents can collaborate on a task. You can checkout agent teams. Here are few examples github.com/phidatahq/phidata/tree/main/cookbook/teams Feel free to join our community for a deeper discussion: community.phidata.com/ | discord.gg/Ye8rQbaT

You can try running phi auth to authenticate. If that doesn't work, you can manually set the PHI_API_KEY environment variable by copying your key from phidata.app Let us know if you need any more help!

Hey there! Which browser are you using, and have you had a chance to set your PHI_API_KEY? If you need more support, feel free to join our community at discord.gg/Ye8rQbaT. Our engineers will be happy to assist!

@@phidata I was on safari, I tried it from chrome on windows and it worked seamlessly, thank you for the reply. Thanks so much for local AI Support, it's awesome.

yes sir! sorry for the confusion. i have another video with llama3.1 coming up which is 8b my bad and I apologize

@@phidata Yes Llama3.1 8B is much better

Glad you enjoyed it!

thank you, document uploads is in the works :) coming soon!

We are glad you liked it!

Great to hear!

@phidata cheers guys, should be proud of your work.

its open source. its on you to implement whatever login you want

No worries, Ive tested it and the finance agent doesnt work and just says I dont have access to any realtime finacial data. You can only use one agent tool at a time?, web search for example, you cant use web search and another tool in the same agent (you should be able to click a checkbox against multiple tools and have an agent use them) The RAG system isnt functioning at all for me, im on windows sooooo.... All in all, its very buggy and to be honest if I have to deconstruct the beta in such a large way to remove the login system that NEEDS to be internet connected even though the use case is totaly localy run agents systems that might not even have internet connectivity on that specific server for security reasons. I have streamlit apps that work better than this with agent systems (I use CrewAI), Langchain support should be in theory part of the framework to begin with as an option. When using ollama, you have to hard code the ollama model? weird, ok. What you need to do is have a drop down menu under the agent thats lets you select the ollama model per agent out of a list of ALL ollama models available, you can return models list with ollama in python, i do it in streamlit and just select the model you want. If I retask this it probably would be to intergrate it with CrewAI, but crewai already has a secret prototype GUI system like this thats soon to be released. I liked the idea, implementation is iffy. Good luck, im out.

@@KingBadger3d Open Source. Change it yourself

Thank you! Cheers!

Thank you for your question! Since the agent's memory is stored in-memory, the retrieval speed remains quite fast, even for conversations from a month ago. We’ve implemented three retrieval strategies-last_n, first_n, and semantic-so you can choose the one that best suits your use case for optimized performance. For your other question, yes, it’s definitely possible to integrate a custom-trained LLM focused exclusively on company-specific information. This can be a powerful way to tailor responses precisely to your needs Feel free to join our community for a deeper discussion: community.phidata.com/ | discord.gg/Ye8rQbaT

For local models, the system requirements typically depend on the provider, and the RAM and VRAM needed can vary depending on the size of the model. If you're planning to run Llama 2 on Ollama then we recommend checking Ollama's website for specific requirements. They usually offer detailed information about the necessary hardware specifications.

Thank you for your kind words! 🙏 You can use Ollama by following the steps outlined in our documentation. Feel free to check it out here: docs.phidata.com/models/ollama Let us know if you need any more help!

Whisper turbo* as well

working on it, will have something in december (november is a bit busy)

Hello @himanshi191 👋 OpenAI's API key is not available for free. They offer a pricing structure based on usage and you'll need to create an account and provide billing details to obtain an API key.

I got a similar error, have you find a solution for this?