Created on 1st March 2025
•
Today, AI Agents are becoming increasingly popular, but everyone is relying on centralized solutions:
While we start to see agents interacting onchain with wallets, it's time to make them use decentralized solutions and give them a way to become truly autonomous and unstoppable.
But this technical improvement shouldn't come at the cost of accessibility and ease of access, good interfaces and abstractions are also needed to allow any human being to deploy his very own unstoppable agent.
That's where CreAItors comes in 🔥
We provide a framework for developers to build AI Agents that use private & decentralized inference and can pay for their own computing on a decentralized cloud with a wallet that only them control.
These agents can then be published on our marketplace in a few minutes, allowing anyone to spin up its own instance and version of this agent.
Several customization options are available, including the strategies to stay alive by staking $ALEPH tokens faster than they are spent for the computing power, the built-in ability for the agent to kill itself if it deems it necessary, or the redistribution of the agent revenues to its creators.
With our interface that is accessible to everyone (login possible with Google and Coinbase Wallet in addition to traditional crypto wallets, payments in credit card via Coinbase Onramp...), nobody is left behind and deploying an own autonomous and unstoppable agent is just a few clicks away!
CreAItors is available through our decentralized web UI (deployed on IPFS with Aleph Cloud).
Users can access it at https://creaitors.andresmolins.dev/, where they can login with different providers to create and deploy agents.
Our project supports 2 login providers:
Our project gives developers access to a Python framework to easily create AI Agents that can pay for their own computing (published on PyPi).
After creating their agent and pushing it to a GitHub repository, "creAItors" (aka developers) can login in our marketplace and click the "Upload AI Agent" button.
They will then have to fill some metadata about the agent (name, description, image...) and upload a .zip file with the agent code (that can be directly downloaded from GitHub).
After signing a few messages to upload it on Aleph's decentralized storage, the agent will be visible in the marketplace and anyone will be able to deploy an instance of it.
Anyone, no matter their technical level, can use our marketplace to deploy an unstoppable AI Agent.
To do this, they simply need to log in, search for the agent they want (using categories filtering and/or name search), and click deploy.
They will then have to give it a name, configure some variables to make it behave the way they want, fund with a small amount of ETH on Base the address of the agent (to pay for its first hours of computing, doable by credit card with Coinbase Onramp for people not familiar with crypto), and wait for the agent to be deployed.
Finally, they can check the logs of the agent's though process in their "My agents" page that displays all their deployed agents.
The marketplace is a simple NextJS app built with TypeScript and TailwindCSS with shadcn. It uses Aleph Cloud's SDK to upload & pin files on IPFS and store messages on their decentralize network, acting as our agents database.
The agent creation basically just consists of pushing data to Aleph so it becomes listed in the marketplace.
To deploy an agent, we ask the user to sign a message, with is sent to the backend and used to generate a wallet (more details in the backend section).
The wallet address is then displayed and should then be funded by the user (ETH on Base mainnet, with a manual transaction, automatic with Okto, or using Onramp for credit card payments)
The backend uses the message signed by the user to generate a wallet by running a SHA3-256 on a string combining a fixed message, the signature hash and the UUID of the agent.
When the wallet is funded with ETH, it uses the web3 package to perform a swap on the Uniswap WETH/ALEPH pool for a calculated amount of ALEPH, then initiates a Superfluid stream to Aleph to start an instance, and deploys the agent on it with a simple shell script + Dockerfile, while also injecting environment variables that include the wallet private key.
Login in the instance is done with a randomly generated SSH key that is not logged and forgotten after the deployment, making the agent truly independent in his instance (the same goes for the wallet address).
Built as a wrapper of LibertAI Agents framework, we use AgentKit to provide a wallet to the agent, with a custom action provider containing actions to get information about the computing status (tokens consumption rate, balances, prices...), perform a swap of ETH for ALEPH, commit suicide (stopping the Superfluid stream that pays for the instance) or redistribute revenues to its creators.
The "self-funding" thinking to buy tokens or not is launched at a customizable interval.
We use various SDKs and products in our project:
Every onchain interaction happens on Base:
We are using Okto social login with Google to allow users not familiar with crypto to log in with an account that is familiar to them.
The Session Keys allow us to simplify and speedup the agent deployment flow by abstracting the multiple message signatures and the wallet funding transaction.
Although we didn't had time to implement it, adding Okto as an option for the Agents wallet is also a very interesting possibility for an even easier process.
Our whole platform is built and deployed on Aleph's decentralized cloud:
We are leveraging this private AI project (that is itself built on Aleph) by building our framework and "self-funding of hosting" feature on top of their existing AI Agents framework, that uses LLMs deployed on the Aleph network, without logs or tracking.
Tracks Applied (6)
Base
Coinbase Developer Platform
Coinbase Developer Platform
Coinbase Developer Platform
okto
Technologies used