Comparison with vercel-labs/agent-browser

[jellydn]

Hi there,

I've been exploring dev-browser and find its approach to persistent browser automation very interesting. I've also come across vercel-labs/agent-browser, which seems to have a similar goal of providing a long-running browser instance for AI agents.

From my understanding, both projects aim to solve the problem of maintaining browser state across multiple automation script executions.

My understanding of dev-browser is:

• It runs a persistent Chromium instance with a server that manages pages by name.
• A client connects to this server to get a Playwright Page object, allowing for stateful automation.

Could you clarify how dev-browser's architecture and approach compare to vercel-labs/agent-browser? Specifically, I'm interested in:

• Architectural differences: How does the client-server model in dev-browser compare to agent-browser's approach?
• Use-cases: Are there specific scenarios where one would be preferred over the other?
• State management: How do the two projects differ in how they handle browser state (e.g., cookies, sessions, localStorage)?

I couldn't find any existing discussion on this, so I thought it would be valuable to ask directly. This could also be helpful for other developers evaluating these tools.

Thank you!

[lzfxxx]

@claude do the comparison