Guided first useful note
Use the sample vault or a real capture, run AI ingest, open the linked note, ask the first question, and inspect the graph.
Cortex.md captures web pages, Gmail, Drive, documents, and YouTube into an Obsidian-compatible vault on your disk. AI reads and links the material, then leaves a ledger of the model, credits, sources, snippets, and outputs.
Try the guided sample vault first, or install free with 25 monthly AI credits.
Cortex.md is now shaped around local file ownership, auditable AI, and context that can be reused by other tools. Capture is still fast; the differentiator is what the vault records after capture.
Use the sample vault or a real capture, run AI ingest, open the linked note, ask the first question, and inspect the graph.
Every AI action records provider, model, credits, input files, snippets, source type, outputs, and a retention note.
Save research runs as Markdown folders with claim/evidence tables, citation graphs, bibliographies, confidence, and stale-source warnings.
Create .cortex context packs with curated notes, a manifest, and prompt templates for Claude, ChatGPT, Cursor, and Codex.
Refresh captured sources, compare changed claims and citations, preserve raw history, and choose when to re-ingest.
Read .cortex activity, hot cache, and provenance from inside Obsidian, then open Cortex actions against the current note.
Onboarding now starts with a useful research loop: capture a source, run AI ingest, open the linked Markdown note, ask one grounded question, then inspect the graph and activity ledger.
Start from the guided demo vault before OAuth, or hit ⌘⇧S to capture a page, email, Drive doc, upload, or YouTube source.
Quick, standard, detailed, and max ingest levels make cost and output depth explicit. Each run writes frontmatter and an activity ledger row.
AI tags, categorizes, extracts links, and writes Obsidian-compatible Markdown you can open in Cortex, Obsidian, VS Code, or git.
Ask a grounded question, review the citations, then inspect the graph and Activity trust center to see what AI used and created.
Cortex.md writes to a folder you pick. Open it in Obsidian. Grep it. Commit it to a private git repo. Move it to a different machine. The vault is not a database we hold for you - it is just files.
Two AI surfaces, both included in your plan. Ask answers questions from notes you've already captured, with wikilink citations. Research writes sourced notes and can save the run as a portable research pack with claim/evidence tables, citation graphs, source confidence, and stale-source warnings.
Three threads ran through what you read on BCIs in spring 2026:
1. Bandwidth is the bottleneck. Current invasive electrodes max out around a few thousand neurons; [[Neuralink and the Brain's Magical Future]] argues a useful "whole-brain interface" needs ~10⁶ channels — a four-order-of-magnitude gap.
2. Read is easier than write. Decoding intent from motor cortex is largely solved; closed-loop stimulation that feels like a thought rather than a buzz remains open ([[Brain-Computer Interfaces]], §3).
3. Substrate independence is the long bet. [[Whole-Brain Emulation]] reframes BCIs as the gradual half of a Ship-of-Theseus migration off biological wetware.
Every wikilink becomes an edge. The side panel renders a force-directed view of the whole vault — clusters surface naturally, orphans surface what needs linking.
Cortex.md only sees a source when you ask it to. The extension uses activeTab + scripting for pages, and every AI action is mirrored into note frontmatter and .cortex activity files.
Best-in-class models, included in your plan. Hit your monthly credit cap and Cortex pauses expensive actions - no surprise bills.
Capture a source, audit the AI work, save research packs, export context packs, and keep the same files in Obsidian, VS Code, git, or any editor.