Compare
Linda vs PageAgent
PageAgent and Linda both put an agent on the page. The difference is abstractions: Linda exposes a VFS, hooks, skills, and multi-agent; PageAgent is leaner but less primitive-rich.
Pick Linda when…
- You want a real agent loop with a VFS, not just chat-on-page.
- You need hooks, skills, multi-agent, MCP.
- You need browser-side file parsing.
- You want 5 LLM providers out of the box.
Pick PageAgent when…
- You want the absolute leanest in-page chat SDK.
- You don't need a VFS abstraction.
- Your AI feature is purely Q&A about the page.
The closest competitor
PageAgent is the project most architecturally aligned with Linda — both run in the browser, both use a tool loop, both keep state client-side. The differences come down to abstraction depth and surface.
What PageAgent does well
- Minimal. Smaller surface, easier to reason about end-to-end.
- Focused. Less to learn if your use case is purely “answer questions about this page.”
What Linda does that PageAgent doesn’t (or does differently)
- Virtual filesystem. Linda’s VFS treats all data the model touches as a uniform filesystem. PageAgent doesn’t expose this abstraction.
- Hooks. 19 lifecycle events with veto + transform pipeline.
- Skills. 8 built-in composable capability bundles +
defineSkill(). - Multi-agent. Multiple agents with behavioral triggers + handoff tool.
- MCP. Client + server.
- Browser-side parsers. PDF, DOCX, XLSX, OCR, Whisper — 7 lazy-loaded packages.
- Skills. Resume-intake, KYC, payment, signature, consent, wizard, booking, address-verify — all bundled.
- Provider count. 5 + OpenAI-compatible
baseUrl.
The architectural bet
Linda’s bet: agentic UX is rich enough that you want a real runtime with proper primitives — VFS, hooks, skills, multi-agent, MCP — instead of building those yourself.
PageAgent’s bet: most use cases are simple enough that the primitives are overhead.
Both are valid; pick by where you think your use cases will land in 12 months. If your “page Q&A” is going to grow into “page agent that does stuff,” Linda’s abstractions save you the refactor. If it’s going to stay Q&A, PageAgent’s leanness is a win.
Compare on the dimensions
| Dimension | PageAgent | Linda |
|---|---|---|
| Core size | Smaller | ~35 KB gz |
| VFS | No | Yes (8 mounts) |
| Hooks | Some | 19 events |
| Skills | None | 8 built-in + define |
| Multi-agent | No | Yes |
| MCP | No | Client + server |
| Browser parsers | No | 7 packages |
| Providers | Fewer | 5 + custom baseUrl |
| Framework | Agnostic | Agnostic |
| Chat UI | Yes | Yes (Shadow DOM) |
| License | OSS | MIT |
The honest take
If you’ve been using PageAgent and it’s been fine — great, keep using it. You’ll move to Linda when you start writing your own glue for hooks, multi-agent, or file parsing. At that point you’re building Linda; might as well use the one that’s already there.
FAQ
Why does Linda need a VFS at all?
Because the agent has to read multiple sources — page state, conversation log, dropped files, installed skills, your mounted data. A flat prompt becomes unmaintainable. A filesystem is the abstraction modern LLMs are best at.
Ship an agent-driven flow this afternoon.
Install Linda, paste a config, and your form turns into an agent that fills its own inputs.