Breaking Ground with Agents

I took these notes from a meeting a long time ago, and it looks like I didn't publish them.

Seed = a sovereign, knowledge-centric multi-agent system with strong permissions, transparency, and self-hostability.

A knowledge operating system, where agents are first-class actors over knowledge.

Agent

A long-lived, stateful process that can perceive, reason, and act over Seed.

Includes:

  • state

  • execution loop

  • memory access

  • permissions

  • tools/skills

Session

  • log of messages + tool calls

  • replayable (critical!)

  • NOT the same as memory

Why agent coordinator?

is it needed?

Coordination requires a Coordinator Agent? or the Coordination can emerge automatically as Bazaar.

Agents can have permissions to spawn other Agents.

Encapsulation

We need a server to run it with a URL and a model provider.

My Node should Agent running on a Seed Node.

Agent Server store the Agent State

Sessions

Streamed information

Secrets

Credentials to the Model Provider.

User Story

Create an Agent.

Use somebody else's Agent.

Agent for Burdi

Agentic Loop

Model

Tools

Sovereign Agent.

Agents are organized by the Community Organizer

Social Agents

We can make a connector to Hugging Face for inference.

Opportunity

Sovereignty from the Frontier Models.

Agents.

Marketplace of Agents

A knowledge community wants to capture its expertise.

Another Community can hire other Agents.

AI companies to adopt us.

Do you like what you are reading? Subscribe to receive updates.

Unsubscribe anytime