Hosted ongabo.esvia theHypermedia Protocol

2026-05-04: Shape Up

Wrap up Account Creation, Sign In, and account management

Document web Editing

We don't need authenticated nodes for this project.


Review qa-review, move them to GitHub, and review Horacio's GitHub tasks.


Agents

micro VMs

Race Conditions when you are working with an Agent.

Pi manages it pretty well.

Message Steering

waits for the agent to complete and then inserts the message into the chat, without interrupting while the agent is calling.

Message Follow-up.

ba

Pause, Stop.

We don't want to

How to work with budgeting?

Code execution in another machine.

we can use a subscription model.

We want them to be self-hostable Agents.

How to give access 10 users an Agent with permissions.

What users consume all the tokens.


We need a billing system token-based. Issue Tokens.

We don't want to go fully remote LLMs.

What is the relationship between Agents and LLMs?

Agent Server?

In the configuration process user must decide their LLM.

Seed offers how Agents interact with Seed.

we have the technology for everybody to interact with humans and Agents.

How to do it as open source?

How to build as a Knowledge Manager?

How to build sovereign systems, provenance, and governance?

Our inspiration:


Are we doing both Agents:

Blocking or synchronous subagent.

These are easier to implement

Syncronous

Parent agents continuously run while the other Agent is working.

Blocking

Parent agent is blocked by the child agent.

Explore Agent


Long live subagents


Subagents

Agents that can do anything for themselves until another agent asks them to do something. Subagents have specific tasks to do.



What is the definition of Agent?

AI agents are software systems that use AI to pursue goals and complete tasks on behalf of users. They show reasoning, planning, and memory, and have the autonomy to make decisions, learn, and adapt.

Ability to operate autonomously in complex environments.

This definition is missing the state, session, execution capabilities, secrets, specifications,


Memory

Agent Memory is the knowledge base implemented as Sites.

Boundaries

Session

A sequential log of user messages, agent messages, and tool calls.

Contains context about a specific task

Should a session be considered part of a Site?

What is the concept of Goal or Soul?


Goal:

Agent


Code execution?

OpenClaw is bad...? Using

the use markdown without the super properties of seedhypermedia


Agents using seed doucment is model is

Agents using our permission model

We need to be careful with the context window.

Memory was RAG before.

We are happy that we are building vector control for giving access.

MIT LLMs: Arbitray companies and compaction.

Question Open Claw?

Seed users should be able to use Open Claw.

We should be able a lower thing, not a ful agent.

We can build something better:

More transparent, visibility on what is going on.

This is what the Agent is doing underhood

Same as Provenance and traceability?

No, this is the Agent signing

Community and collaboration.

Knowledge Management Collaboration

Cryptographic Permissions

We don't need to do MCP or A2A flows.

But we do for System of Records.

How does an agent calls another agent?

Does an Agent have tools?

An api key is not a tool.

Ability is a mix of tools, skills, and api keys.


We want to support the Knowledge Community.

Connecting to people.

Connecting to content.


Give a Skill to an Agent. What is a Skill?

Skill can be a markdown or a Seed document.

Rules and Goals.

Skill can have things that we can't have in Seed. (We need to


A Context or a Prompt is the same as a Goal.


We must discover what a Knowledge Community needs.


Define User Stories.

Questions.

Problems.

Definitions

Architecture.


Let's run a team open claw.

Open claws might run many agents. they we all share the same rules and security. They have different Workspaces/Profiles with the same rules and profiles.

What are the rules on how to use it?

The Agent should only work on Knowledge.

Seed-only channel.

Open claw has a bunch of built-in channels. We must mute those.

We need to create a Seed Channel.

Eric has created already so he will write a document on how. hm://z6MkgisVMELvqnsCo3dYmtVpy8PiqPGMVwfAyBWFn84vebq4/:profile

Why do we wnat one agent per purose?


Agent connecting to an LLM?

    More than one model or llm.


Precise control of what agents

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

Unsubscribe anytime