The past weeks have shown how powerful AI agents are and that they will help us in most of our businesses and daily life. But they still require a lot of external access, like good APIs. This is an open approach to solving this. We are open for contribution!
Agents should be able to interact with most of the world in the name of their human representative. Currently the web is super restrictive to agents. Every new platform needs to be manually integrated, and every action programmed. This needs to be reversed.
The core of an agents/acc is an identification method. With APIs, we currently use account-bound bearer tokens, OAuth, or similar methods. This would require an account on each platform the agent interacts with. Instead, we will use a more decentralized authentication, in which the agent identifies its operator to the service, tool, or site it's using. This allows to sign usage agreements.
With a concept of "action prompt" we can create an easy format for agents to read. Instead of developers having to code integrations for the platforms the agent should use, the creators of agents can just provide an url to the service. The agent then can parse the "action prompts" and take actions on them. These prompts follow a format like this:
Trough the signed agreement and pricing approach being handled in crypto the enforcement is handled automatically. The operator has to accept the agreement once and then gives the agent to approve or reject "transactions". A big question is if this is really handled on a blockchain or just takes inspiration from the web3 space.
If you want to add your own ideas here, feel free to add a PR. The whole site is open for everyone on Github. You can also email me, follow my twitter, or book a call with me. Happy to chat and invite more people to this project.