Element to LLM — insitu.im

Element to LLM
A new UI → LLM layer for the modern web.
Now AI actually understands interfaces.
You already know that screenshots don't tell the truth.
Raw HTML hides what really happens.
Accessibility trees miss what users see.
Element to LLM introduces a new layer between your UI and your AI tools — a precise, structured, runtime-aware representation of interface reality.
Not static.
Not theoretical.
Not partial.
Full runtime context → delivered directly to your AI model. Privately. Locally. Efficiently.

🚀 What's new in v2.8.0
Persistent capture — save to disk
Captures can now be saved directly as JSON files, not just copied to clipboard.
Configure your output path (defaults to Downloads/sifr-captures/).
Toggle between Clipboard and File mode — existing workflows are unaffected.
Saved captures unlock diff workflows, audit trails, and repeatable pipelines.

SiFR v2 — the next generation of UI representation

A fully redesigned output format optimized for LLM reasoning and minimal token usage.

Structure:
METADATA → NODES → SUMMARY → RELATIONS → DETAILS
Salience scoring (high/medium/low)
Spatial clustering with automatic role detection (nav, header, card, modal, footer…)
Relationship graph: alignment, containment, proximity, stacking
Action markers (clickable, fillable, hoverable)
CSS cleanup + shorthand compression
Smarter full-page mode with relaxed limits
Token efficiency built in
SiFR v2 is engineered for LLMs:
Removes noise
Compresses structure without losing meaning
Prioritizes the parts that matter
Cuts token consumption dramatically
This isn't just "DOM → JSON".
It's semantic compression for interface reasoning.

🔍 Optional analytics (100% opt-in, privacy-first)
Helps us improve the product — only if you choose.
Collected only when enabled:
Daily capture count
Format preference (v1/v2)
Capture type ratio
Anonymous install UUID
Never collected:
URLs, page content, DOM data, PII, browsing history, anything sensitive.
Data is flushed once per day using alarms.
Privacy policy: https://insitu.im/e2llm/privacy

⚙️ Technical hardening
Added alarms permission for scheduled telemetry
Added downloads permission for file output
Core algorithm now under BSL 1.1
CI/CD automatically injects license post-minification
Stability and performance improvements across the capture pipeline

✨ What Element to LLM actually unlocks
When AI sees what you see, everything changes.
Instead of guessing:
why a button doesn't respond
why responsive behavior collapses at specific widths
why aria attributes don't work
what styles override each other
why something works in dev but fails in prod
AI receives the actual runtime truth:
z-index conflicts, computed styles, hidden layers, bounding boxes, accessibility visibility, relations, layout flow, salience scoring — everything that affects how the UI behaves.
Finally: AI can reason about real interfaces, not abstractions.

🧰 Not just for debugging
Element to LLM is used daily for:
Debugging — understand root causes instantly
QA — capture real behavior, not theory
Design & UX — compare spec vs implementation
LLM agents — provide accurate UI state for autonomous actions
Automation / RPA — eliminate brittle selectors
Documentation — produce clear, structured UI context
Accessibility analysis — see what assistive tech actually perceives
Product reviews — communicate issues clearly across teams
It's no longer a tool for "fixing bugs".
It's a UI → LLM interface layer for any workflow that involves AI and real products.

⚡ Before vs After
Before:
"The modal seems broken."
After:
AI receives structured evidence showing stacking order, offscreen positioning, aria mismatches, and computed style chain — and gives a precise fix.
The difference feels unfair.
In a good way.

🔒 100% Local. 100% Private. 100% Yours.
All processing happens in your browser
No external servers
No remote code
No content or URL collection
No background network usage
Telemetry opt-in only
File output stays on your machine
This is one of the rare tools in the AI world that works without touching your data.

⭐ Install now — Chrome, Arc, Brave, Edge
Element to LLM becomes essential fast — usually in a few days.
Love it? Rate us ⭐