Re: Moving Forward with Agentic Linked Data

The delegation is done via the manager (e.g. role/persona) at the top of the diagram.

 

An Agent in this case is not a traditional “process” like the ones that run today in your computer – they could be a GPT that is technically not “exclusive” to the scope you are requesting.

 

If a User wants to delegate access, it goes through the management flow, and the agent now can operate with the same authorisation flow without having to “hold” any information about the User.

 

Regards,

 

___________________________________

Joshua Cornejo

marketdata

smart authorisation management for the AI-era

 

From: Sindhu Raju <sindhu2@mac.com>
Date: Wednesday 19 February 2025 at 18:37
To: Melvin Carvalho <melvincarvalho@gmail.com>
Cc: Joshua Cornejo <josh@marketdata.md>, public-solid <public-solid@w3.org>
Subject: Re: Moving Forward with Agentic Linked Data
Resent-From: <public-solid@w3.org>
Resent-Date: Wed, 19 Feb 2025 18:37:07 +0000

 

hi Joshua .. can you give a real world use case for the user where they would do the delegation and how it would work.

 

 

Sent from my iPhone




On Feb 19, 2025, at 7:59 AM, Melvin Carvalho <melvincarvalho@gmail.com> wrote:



 

 

st 19. 2. 2025 v 16:25 odesílatel Joshua Cornejo <josh@marketdata.md> napsal:

Hi Melvin,

 

Following up – this is my proposed strawman flow for agents:

 
Management can create a delegation request to the PDP, this delegation has the actions that the agent can perform, and an expiry date with an optional counter (how many times the agent can interact in total, it could be “paired” to the actions, but not sure it will be really used compared to expiry times).
Nothing changes for the user making a request.
Agent has to present a property of “delegate=true” within the ID so the PDP can work out for that context if there is an active delegate-relationship between the agent and the user.
 

This way, the Agent doesn’t need to know details about the user and can behave as “delegate” where required and as a “user” where not without too much overhead.

 

 

<image001.png>

Thoughts appreciated!

 

Many thanks Joshua!

 

I raised Agentic Linked Data briefly at last week's CG call:

 

https://github.com/solid/specification/blob/main/meetings/2025-02-12.md#agentic-linked-data-radar-check

 

Thanks to the chairs and TimBL for their time and advice.

They suggested the possibility of a "special topic" session (~1hr) as a next step, potentially covering subtopics like auth. Solid World is coming up next week, so no rush. A work item or CG report is also an option.


If there are no objections, we could start a new thread to explore this further.

Best,

Melvin

 

 

Regards,

___________________________________

Joshua Cornejo

marketdata

smart authorisation management for the AI-era

 

From: Melvin Carvalho <melvincarvalho@gmail.com>
Date: Monday 10 February 2025 at 10:28
To: Joshua Cornejo <josh@marketdata.md>
Cc: public-solid <public-solid@w3.org>
Subject: Re: Moving Forward with Agentic Linked Data
Resent-From: <public-solid@w3.org>
Resent-Date: Mon, 10 Feb 2025 10:27:19 +0000

 

 

 

čt 23. 1. 2025 v 19:29 odesílatel Joshua Cornejo <josh@marketdata.md> napsal:

I am in the middle because not everything needs fixing, and maybe not all use cases will be ‘fixable’. I already put a comment on the page Melvin distributed:

 

“it would be useful to start by drawing a high-level architecture, so everyone has the same alignment. (for example "deploying agents" gives a different flavour than "accessing solid pods").”

 

The easiest approach (IMHO) is always to build the use cases and a level-0 architecture.

 

Hi Joshua

 

I think high-level architecture is a great idea, and something we can work on.  Perhaps you might want to give it a go?

 

Aside from that a protocol for agent commis is emerging, namely, Model Context Protocol (MCP)

 

This might also be something worth looking at:

 

https://modelcontextprotocol.io/introduction

 

Best

Melvin

 

 

From parallel conversations and research, there are elements of authentication/authorisation that don’t exist today for some use cases (for example, “delegating access” has always been exceptional and involves aspects from both).

 

Regards,

___________________________________

Joshua Cornejo

marketdata

smart authorisation management for the AI-era

 

From: "Jeff Zucker (he/him)" <dubzed@gmail.com>
Date: Thursday 23 January 2025 at 17:37
To: Sindhu Raju <sindhu2@mac.com>
Cc: Melvin Carvalho <melvincarvalho@gmail.com>, public-solid <public-solid@w3.org>, Antoine Zimmermann <antoine.zimmermann@emse.fr>, Andrei Ciortea <andrei.ciortea@inria.fr>
Subject: Re: Moving Forward with Agentic Linked Data
Resent-From: <public-solid@w3.org>
Resent-Date: Thu, 23 Jan 2025 17:37:07 +0000

 


Sindhu Raju wrote "My understanding of SOLID. it meant for individual data"

 

I strongly disagree with this statement.  There are numerous use cases and actual applications aimed at organizations and enterprises. Who do you think are Inrupt's customers?  The construction industry clients on Trinpod are not individuals.  I can name a dozen different Solid apps aimed at homeless service case workers, community health clinics, tracking sustainability information and other types of data that goes far beyond the individual.  INRIA has a major project aimed at small farm inventory.   SOLID is for both individuals and organizations, 

 

On Thu, Jan 23, 2025 at 8:51 AM Sindhu Raju <sindhu2@mac.com> wrote:

Hi Melvin,

 

I think we should have discussion on types of of agents and use cases. My understanding of SOLID. it meant for individual data and I think we should limit the scope to individuals and their use cases. I think discussing agents at large is not appropriate and not a good use of our time. In fact we should only be discussing Agents in a manner that the general public can understand.

 

Thanks,

Sindhu

 

On Jan 23, 2025, at 7:25 AM, Melvin Carvalho <melvincarvalho@gmail.com> wrote:

 

Hi everyone,

Thanks to all who’ve shown interest in the idea of Agentic Linked Data. The conversations so far have been great!

As requested, I’ve put together a short document to gather thoughts, ideas, and participants. It’s a living document, so feel free to jump in, add your name, share your expertise, or propose new topics:

 

https://hackmd.io/@jSWt69rrQAKeUTsa_s49Qg/H16RCGRv1g

For convenience, here’s the content:

Agentic Linked Data

This document is based on discussions on the Solid mailing list around Agentic Linked Data [1]. It contains a list of participants who’ve shown interest.

1. Discussion Participants So Far
Below is a list of participants with a short description of interest. Please edit the text as desired.
Melvin Carvalho: Initiated the discussion. Interested in Agents and AI.
Joshua Cornejo: Experienced in ODRL and authorization architecture.
Eric Jahn: Interested in agent roles and permissions.
Aad Versteden: Integrates AI and Linked Data in semantic.works.
Sindhu Raju: Focuses on personal data management and consumer use cases.
Matt Taylor: Consultant in ethical tech, cautious about broad agentic access.
Ram Mukund Kripa: Developed privacy agents for consent management.
Jesse Wright: Highlighted the Web Agents CG.
Jacob Beauchamp: Exploring agentic learning and entity data.
Knut-Olav Hoven: Agent Linked Data.
Leonard Rosenthol: C2PA and Trust Signals.
Harshvardhan J. Pandit: Data Privacy Vocabularies (DPV).
Antoine Zimmermann: Chair of Webagents CG, knowledge transfer.
Kurt Cagle: Interested in how this is evolving.
Davi Ottenheimer: Modelling Agents.
Mark Lizar: Agents, privacy, legal aspects, and DPV.
Add your name here!

2. Potential Topics of Interest
Below is a list of topics, in no particular order, that could be interesting to work on in the future:
A common understanding of the term "Agent"
Deploying Agents to Solid Pods
Add a topic here!

[1] https://lists.w3.org/Archives/Public/public-solid/2024Dec/0014.html

 
Next Steps:
We’ll consolidate ideas and propose a direction in the next 2 weeks.
Identify topics of interest and who’s keen to explore them further.
Liaise with the Web Agents CG.
Talk to folks in the Solid CG about structuring this (e.g., forming a Task Force).
Please feel free to edit the document with your thoughts—we’ll tidy things up and summarize in the coming weeks.

Looking forward to an Agentic 2025!

Best wishes,
Melvin