The Starlight builder thesis: an open home for useful intelligence

The work should have somewhere to go.

A chat can produce a useful answer and still leave the next person with a difficult handoff. Which sources mattered? Why did we choose this approach? What should the next agent keep, and what should it question again?

Starlight starts with that practical problem. Make the context readable. Give each agent a useful responsibility. Keep the evidence near the decision. Let the person who owns the work carry it forward.

This is a builder thesis and a research direction. The current product entry is Memory Studio, where you can edit and export five context files. The public source index identifies released code and licences. The wider capabilities below need their own implementations and evidence.

An open foundation can support many independent places.

One builder may create a research tool. Another may maintain a local agent installation, teach an effective workflow, or serve a community with particular needs. A shared foundation is useful when it reduces repeated work while leaving those builders room to own their products and responsibilities.

The community we envision grows through working examples, careful repairs and knowledge people can reuse. Participation begins in the public repositories. Showcases, a contributor programme and commercial collaborations remain proposals; each agreement would define its terms separately.

Seven questions for a wider web

We sometimes use “Web7” as a name for this proposed direction. It is our thesis, not an established web standard or a claim that earlier forms of the web have been replaced. Reading, publishing, communities, knowledge, programmable ownership and agents can coexist.

1. People: who owns the project and makes the consequential decisions?
2. Knowledge: what should persist, how is it corrected, and who can read it?
3. Agents: which role is allowed to perform which bounded task?
4. Tools: how do software interfaces expose useful actions and report failures?
5. Economic networks: what needs settlement or provenance, and who controls signing and spending?
6. Machines and living environments: how does a task relate to a real sensor, device or physical consequence?
7. Evidence: what result, cost and intervention can another person inspect?

A memory export, a tool call, a transaction and a robot action each need their own tests. Answering these questions makes the boundaries clear before we connect them.

Use an interface because the task needs it.

Model Context Protocol provides a way for hosts and tools to exchange context and capabilities. The application still has to define the meaning of its records, enforce permissions and decide how a result becomes accepted knowledge.

Ethereum and Solana are research candidates when a task has a concrete need for on-chain state, settlement or provenance. A proposed adapter would need chain-state checks, explicit transaction simulation, signing authority, spending limits and a recovery method. A private knowledge record does not need to be put on-chain to become useful.

ROS 2's client library is a useful reference for interfaces to robotics. Physical actions introduce deadlines, feedback, cancellation and consequences beyond a browser session. A robotics research path must begin with bounded simulation and human review before any deployment to a device.

These references describe where experiments could lead. Starlight does not claim live Ethereum, Solana, robotics or biological-system integrations from this thesis.

Start with the smallest result someone else can reproduce.

Change one project decision, start a fresh agent session, and check whether the current decision is used correctly. Export the record. Try it in a clean environment. Keep the source revision, observed output, failure and repair effort.

That makes the next step concrete. A published design becomes more valuable when a builder can test it, challenge it, improve it and create something of their own above it.

Read the portable memory architecture, explore the research notebook, or bring a contribution.