Hi folks, Any thoughts on integrating Palm V2 which powers Bard? It is currently free until around December which might help manage costs.

Hello, I would like to ask: I'm looking to get the MESSAGES that come from AgentGPT to outside with some sort of API. Basically I'm trying to build a custom discord bot where the user will ask to do a certain thing and it's supposed to send an "order" to AgentGPT using a template prompt as a base, and I'm looking to get the answer that it makes to outside so it can be sent as an answer by the bot in Discord. Is there any possibility to do this currently?

When generating code write the output to disk!!! this is 1 INSANE feature request, the one to rule them all? Make this a priority feature please! It could easily be integrated. Im talking for the local version. For you cloud version I suggest you write a zip file that the user can download, what a benefit that even would be, right! I would work with you on this and can even suggest other improvements that are in my head that could make this an out of your world expierence. Talking multi million dollar feature oportunity here.

We serious need a autoave/ continue where left off feature in case a agent runs crashes ( due to browser or else gets halted ) This way we can preserve $ saving us from re-reunning the whole agent from start again.

There should be an plugin architecture where you can for the local version store plugins in the docker mounted path under something like /plugins. There should be some basic template to create new plugins with documentation on how to build your own plugin. It should be possible to "dry run" plugins in testing without using real tokens, it would just respond with the intended actions but will not actually use api tokens to run them in real. You could merge/moge your current templates inside plugins each with a dedicated task, there could be a discord forum channel where you can share custom plugin code for a dedicated task, basically these would implement the templates but in such a way that you can easily modify and extend them for personal usage. One would be able to share a plugin on a github repro prefferable the #team would have an central plugin repro where users can submit to. AGENT GPT should be able to load this plugin repro or add it to the distribution of the docker image. There would be Community plugins and Custom plugins community plugins would go in a docker mounted folder /plugins_public and custom plugins would go in plugins_private For inspiration on such a plugin example repro look here: https://github.com/remix-run/examples

The pdf should put the code in blocks not just text. Preferrable also add a "Copy action" button in the pdf if thats a supported feature of the pdf file format.

Make the GOAL field larger or extend when typing upwards in size

- remove the logo on the top centered put it in the sidebar - use chatgpt free web version layout and mimic it in agent gpt. - use sidebar to show convos and settings then use the right section of the agent in full width just like on the webversion of gpt. See image, currently the ui is using to much space, it could look more professional and clean

in the right section of the screen there are subtasks generated. It would be nice if one could erase one of the tasks if its of no use like "write me a react app" first asks is to install react npm etc. this is already installed so i should be able to erase this task so it wont execute or stop executing and stop wasting tokens for usless info

The agents should be able to take the cue from the prompter/user to reorient themselves towards the tweaked goal midway. Is this achievable?

We need api connectore we can request to use in prompts. “Write me a instagram post about.. “ It will find image and write contents posting it to instagram with the ai connector. The api details would be under settings api keys socials. Same for facebook

We ould love to see a blogpost generator that writes marksown blogpost and find corrosponding image. If could support multiple blog posts You should be able to give it a markdown template it should use for the structure or srt it under settings plugins blogpost “ write 20 blog posts about agent gpt find subjects and add image” Writes to file(s)

Add a stimuli response, in order to give you an idea about something. Like add a, music stimuli where a program could identify a song snippet via a audio sample upload section. It understand chords, and emotion, just for music.(Musical Patterns) But also add other ones to understand pictures, and colours. To give an idea or covey the feeling of the image.

Would like the option for a 1-Click installation, and have the terminal just ask you for further inputs in the correct order.

I wnat to change endpoint and use API Reverse Proxy

It would be so useful if it could automatically generate code and commit it to a GitHub repo or local repo

It would be amazing if you make it possible to create an agent, in connection with the last agent we created. Building on the last step? A data analysis as a kind of "memory" to the last agent?

Currently there's google search, images, and a non implemented version of wikipedia i noticed. It would be awesome if we could add our own pinecone db for example to add a vector search capability as well.

it would be great to use SEO apis . but it seems it cannot access quickly and directly

It would nice to be able to store and organize past runs based on folders in order to look over previous runs and be able to organize said runs by topics.

Option to create files like word, pdf, code files

Option to load files like pdf, word or code files

MemGPT offers a hierarchical memory system that allows for self-directed memory management and complex control flow, effectively extending the context and memory capabilities of language models. It manages a fixed-context LLM with an operating system-like architecture, allowing it to handle extended conversations and complex document analysis tasks by managing its memory in a sophisticated and dynamic fashi​1​​2​​1​​2​​1​e】. AgentGPT, on the other hand, addresses the memory limitations of AI agents by employing Vector Databases to store execution history externally, making it retrievable when needed. This enhances the agent's ability to remember and reference past actions, mimicking human memory retrieval through text similarity sear​3​​4​​5​​6​​7​e】. By integrating these two systems, we aim to create an AI with: Extended Context Awareness: Using MemGPT's architecture to extend beyond the typical context window limitations of language models. Dynamic Memory Retrieval: Employing AgentGPT's Vector Databases to access historical data and execution history, allowing for memory retrieval over many loops. Sophisticated Prompt Design: Utilizing Reworkd’s prompt crafting techniques to guide the combined system towards accomplishing specific tasks with high accuracy. Self-Directed Memory Edits: Incorporating MemGPT's ability to autonomously update and search its memory based on current context and objectives. Hierarchical Memory Management: Combining MemGPT's hierarchical system with AgentGPT’s long-term memory storage, offering a nuanced and layered approach to memory management. This integration would create a powerful agent capable of managing a more extensive context, with improved long-term memory and sophisticated task handling.

While exporting as a PDF, the text is showing an inline citation and an url for that citation. While copying and pasting, the URL is a hindrance. It would a more viable option to make the citation an inline clickable link to the url and showing all url at the end of the task.

Would be a lot better then just a chat

Add a custom endpoint/proxy option for those who don't use openai endpoint.

Add gpt-4o support 🙂

We should work together me and the development team to make virtual.Humans that are piloted by these agents or similar agents.

You know how you can get anthropic to create something and it will actually create it over on the side Kind of like what websim does. This would be very useful for the platformerGPT

it's time to monetization AGENTGPT in ANDROID IOS PC and Macos of course to increase immediately/download immediately like Battle of warships can online and offline mod due monetization u can still earn money while their offline cause if u add all platforms in stores I definitely u ar firstly built AI chatbot easy as pie or call prompt chatbot we definitely happy if u ar add in Android IOS PC and Macos even u release vote 100% accepted I'm not hypercritics I awesome due Google AI I research popular AI agent then boom so I thx to him