- From: Ivo Brett <ivo.brett@gmail.com>
- Date: Fri, 14 Nov 2025 16:33:54 +0000
- To: Enrico Zimuel <enrico@zimuel.it>
- Cc: guha@guha.com, public-nlweb@w3.org, Chelsea Carter <checarter@microsoft.com>, lindsey.li@microsoft.com, Ramanathan Guha <rvguha@microsoft.com>, mojahaverihi@microsoft.com
- Message-ID: <CAJOJ2YRGNdL5RWsaw2Jgs6-i845z3mv_MuzE2HwTjoqnGBG7wQ@mail.gmail.com>
Hi Enrico, Guha, and all, Thank you again for inviting me to join this initiative. I appreciate the opportunity to contribute. I have two comments for consideration: I agree with the direction Enrico mentioned. NLWeb seems to have potential beyond websites and could support conversational interfaces to various structured data sources. If that’s the case, it may be worth revisiting how resources are identified. Currently, everything is URL-based, but how would non-web resources, such as IoT devices (e.g., a fridge), be referenced? The specification currently leaves authentication outside its scope, assuming it can be handled at the HTTP layer, which is reasonable. However, should the specification consider an optional mechanism for finer-grained permissions? As written, any NLWeb instance can access all documents in the collection. One approach could be passing a permission scope through HTTP headers (similar to how the authenticated user is currently provided). This could enable an optional permissions layer that filters documents based on user scope before they are made available to the LLM. Scopes could be handled by Oauth. I believe this is the way MCP is going. This is my first time contributing to a specification effort, so please let me know if I should format or frame feedback differently in the future. Regards, Ivo On Fri, 14 Nov 2025 at 10:02, Enrico Zimuel <enrico@zimuel.it> wrote: > Hi Guha, > > thanks for the invitation to this NLWeb initiative and hello to everyone > in the group. > I left some comments in the specification shared document > <https://docs.google.com/document/d/1ex_Txq4Y_tTqgUTBr7xE470Hzh0wG1cfyvhiLDv0sPs/edit> > (v0.5.3) but I think it would be interesting to share a more general > thought here. > > I like the idea of NLWeb and the simplicity around the concepts of "/ask" > and "/who". > After reading the specification, it seems to me that we’re extending the > protocol beyond the “simple website” scenario. In other words, the protocol > could potentially be used to interact with any system via natural language, > not just web applications. For example, a fridge could expose an /ask API > to check whether it contains milk. Am I overthinking or is that actually > the direction we’re heading? > > Thanks. > > Regards, > Enrico Zimuel > https://www.zimuel.it > > martedì 4 novembre 2025 3:11 PM, R.V.Guha <rvguha@gmail.com> ha scritto: > > Welcome everyone to the NLWeb Community group. > > Here > <https://docs.google.com/document/d/1TZBWr8sNviEjNxXeUwQfgtFkeQUfx1jirN5WIe4lT9I/edit?tab=t.0> > is the current (very preliminary) version of the spec, updated for > output_response structure corresponding to chatgpt apps. Would love to get > a discussion thread started around that. > > Please share the details of the community group widely. It is open to > everyone. > > thanks, > > guha > > >
Received on Friday, 14 November 2025 16:35:05 UTC