add a headless web browser for AI web access #51

Closed
opened 2026-07-28 08:39:16 +00:00 by hugo · 1 comment
Owner

there should be web related tools that allow the use of Obscura to visit and use the web by the llm without the need of a full browser engine integration. on github as h4ckf0r0day/obscura - this should be integrated in the app via its crates, if possible. Alternatively carry the binary inside the bundle and use it from there in a separate process. This definitely should support the full scope that Obscura gives, so that we have a proper browser environment we can use to explore the web.

Make a suggestion for the minimum set of tools that should be exposed to the LLM so that it can properly work - I envision something lile "start a web session with URL" and then query tools to get information about the current page and ways to follow navigations. Also check on the web for agents that already use Obscura to see what kinds of tools they provide to the LLM, we don't have to reinvent the wheel.

Also make a comparison with rust-headless-chrome, because that could also be an option, despite it being based on a chromium download. It could be the more mature option, depending on what you find on the web. For the same reason, also look at jonhoo/fantoccini as yet another alternative, although that needs a webdriver compatible browser. The benefit of Obscura is in the stealth mode, that could allow circumventing some scraping blockers.

The over-arching goal is to provide the LLM with a solid web client toolset to bring in external content in a reliable way. But the first step must be a plan and an overview to decide what direction to take.

there should be web related tools that allow the use of Obscura to visit and use the web by the llm without the need of a full browser engine integration. on github as h4ckf0r0day/obscura - this should be integrated in the app via its crates, if possible. Alternatively carry the binary inside the bundle and use it from there in a separate process. This definitely should support the full scope that Obscura gives, so that we have a proper browser environment we can use to explore the web. Make a suggestion for the minimum set of tools that should be exposed to the LLM so that it can properly work - I envision something lile "start a web session with URL" and then query tools to get information about the current page and ways to follow navigations. Also check on the web for agents that already use Obscura to see what kinds of tools they provide to the LLM, we don't have to reinvent the wheel. Also make a comparison with rust-headless-chrome, because that could also be an option, despite it being based on a chromium download. It could be the more mature option, depending on what you find on the web. For the same reason, also look at jonhoo/fantoccini as yet another alternative, although that needs a webdriver compatible browser. The benefit of Obscura is in the stealth mode, that could allow circumventing some scraping blockers. The over-arching goal is to provide the LLM with a solid web client toolset to bring in external content in a reliable way. But the first step must be a plan and an overview to decide what direction to take.
hugo added the enhancementcritical labels 2026-07-28 08:39:16 +00:00
hugo changed title from add Obscura as headless web browser for AI web access to add a headless web browser for AI web access 2026-07-28 08:57:32 +00:00
hugo added duplicate and removed criticalenhancement labels 2026-07-28 09:25:29 +00:00
hugo added ideawontfix and removed duplicate labels 2026-07-28 19:35:24 +00:00
Author
Owner

we stick with the chrome based thing we have right now.

we stick with the chrome based thing we have right now.
hugo closed this issue 2026-07-28 19:35:41 +00:00
Sign in to join this conversation.