Small, considered changes — shipped when they're right, not when they're loud.
Behind the scenes, SnipGraph now has a shared language for the different ways clips can relate, guarded rollout controls, and a reproducible evaluation foundation. It is groundwork for clearer connection explanations and confidence-aware discovery; today’s graph behavior stays unchanged while each step is measured.
Ask now starts from a leaner, relevance-focused reading of your graph, uses a faster path for everyday questions, and finishes the answer before preparing optional follow-ups. New privacy-safe timing and cache measurements help us keep improving responsiveness without recording what you ask or what your clips contain.
Operations can now safely manage account access, enforce per-account AI budgets, watch workers and retention, merge and file feedback, stage feature flags, and edit transactional mail. Account details also stay neatly contained across desktop and mobile screens. The same privacy line still holds: graph content stays out of the console.
SnipGraph now has a privacy-bounded operator console for measured AI use, daily cost targets, sanitized error patterns, and feedback you explicitly send us. The console works from metadata—not the titles, text, filenames, prompts, or findings inside your graph.
Ask can now explain what changed over time, List search understands meaning as well as keywords, and Rediscover surfaces cited Findings when themes rise or distant collections connect. SnipGraph also learns—transparently and deletably—from AI links you remove, while larger libraries get a bounded research mode that searches and follows the graph before answering.
Every AI-assisted action — from embedding a new clip to answering in Ask — is now tracked as one complete run with its model cost. It is behind-the-scenes work that helps us catch regressions early and keep SnipGraph dependable as your graph grows.
Graph, List, Ask, and Rediscover now sit as tabs across the top, and the sidebar holds just your collections — pick a scope once and every tab respects it. Archive became a filter on the List, so archived clips are one toggle away instead of a separate place.
The graph now notices when two topics are really one — "enterprise AI" and its variants settle into a single hub, and new clips reuse an existing topic when the concept is already there.
Drag PDFs, Word files, or Markdown straight into the graph — each becomes a node and connects like any clip. And every clip now opens in a full reader, not just its excerpt.
Both views narrow live as you type — the graph dims what does not match, and the list filters by title, source, and tag. Press Enter to hand the same words to search.
Clips from feeds and generic pages now get a descriptive title instead of the page boilerplate. And you can select many clips at once to archive or restore — one undo brings them all back.
New connections now settle in one at a time instead of all at once, so a busy graph never flickers. Every link still shows its reason on hover.
Search in plain language and get the matching clips summarized, with every source cited back to the node it came from.
Any organizing the AI does can now be reversed with a single step — merges, links, and tags all return to where they were.
A full dark theme that follows your system by default, tuned so the accent carries the same meaning in both.