A private, automated knowledge system
Keep what matters.
Let the system organise it.
Second Brain turns the things you save, write, read, and plan into a connected Obsidian knowledge system, with automation doing the repetitive work and you staying in control.
See how the system works ↘For people who save a lot, want to use what they learn and are tired of organising everything by hand.
Capture once.
Keep compounding.
Capture
Bring in selected notes, saved content, calendar context, files, and deliberate AI captures.
Filter
The system checks what changed and brings in only useful new material, so old content and noise stay out.
Understand
AI extracts evidence, proposes relationships, and drafts targeted updates to the right notes.
Validate
Deterministic rules verify evidence, ownership, links, expected state, and safe Markdown before writing.
Use
Search the graph, review changes, receive meaningful tasks and digests, and keep the knowledge working.
One system for notes, saved articles, tasks and useful context.
Together, these tools keep a folder of notes organised, connected and ready to use.
Apple Notes ↔ Obsidian
A signed Mac worker keeps selected Apple Notes and Markdown in sync while Apple data stays on the Mac.
Connected knowledge
Evidence becomes structured claims, relationships, decisions, tensions, and links instead of another pile of isolated notes.
Read-only Google context
Selected Gmail, Drive, and Calendar context can inform the system without granting write authority back to Google.
Turn saved articles into notes
Saved articles and bookmarks become short notes with links to the source, so the reading backlog becomes useful knowledge.
Todoist task delivery
Useful next actions are delivered to a dedicated Todoist destination without turning every note into a task.
Search and recall
A private index makes the vault searchable from approved tools while the underlying notes remain yours.
Promotion state
Every derived note keeps source lineage and moves through a visible review state before it becomes knowledge, reference, or archive.
Insight-focused digest
Telegram highlights bounded decisions, connections, tensions, and experiments instead of repeating queue maintenance.
Choose an approved AI provider
The system can use different approved AI providers without changing the safety checks or ownership of the vault.
Obsidian Control Panel
Everyday source, output, writing, and style settings live where the knowledge lives: inside Obsidian.
Whole-system health
Separate checks cover managed notes, human maps, generated captures, source collections, connectors, sync, and backups without publishing private content.
Verified backups
Encrypted snapshots and isolated restore checks protect the vault without testing recovery against the live copy.
The value is not more notes. It is less cognitive housekeeping.
Remember the useful parts.Important context survives inboxes, tabs, feeds, and fleeting moments of attention.
See connections earlier.New evidence is routed toward the people, projects, ideas, and decisions it can actually change.
Reduce maintenance.Routine collection, organisation, validation, backup, and health checks happen without a manual nightly ritual.
Stay in control.Human edits, local Markdown, explicit approvals, and reversible writes remain more important than model confidence.
Cloud automation.
Local authority.
What stays on the Mac
Apple data, the Obsidian vault, signed validation, and final writes stay on the Mac.
- Apple Notes sync
- Obsidian vault
- Checked changes
- Local rollback
What automation handles
Private workflows collect approved non-Apple context, schedule work, call AI models and prepare focused suggestions.
- Read-only connectors
- Model routing
- Retries and cursors
- Health and digests
Organised by meaning,
not by where it came from.
The system separates raw sources from durable knowledge. New evidence is reconciled against what is already known, then routed to a small set of maintained note roles.
Immutable source references preserve where each useful observation came from without copying the source again.
Evidence is typed, ranked, and held for bounded review. Generation alone never makes a note canonical.
Useful material strengthens a decision, thread, map, connection, or reusable knowledge note; everything else stays reference or leaves the queue.
The maintained graph
Roles describe what a note does in the system. Exact paths remain the identity, so similarly named notes in different areas never collapse into one.
Automation can update only its managed region. Everything outside it is verified byte-for-byte before and after a transaction.
A safe, limited way for AI tools to use your notes.
Approved AI tools can search the Second Brain, read one chosen note or save one item you explicitly ask to keep. They cannot browse the Mac or roam through the live vault.
search_second_brainFind relevant knowledge
Searches the private index at a chosen depth, returning the most relevant results and only the requested content.
READ-ONLYget_second_brain_noteRead a selected note
Retrieves one exact note after the client has identified it through search. It cannot browse arbitrary files.
READ-ONLYkeep_in_second_brainSave an explicit capture
Creates one new AI Capture only when you clearly ask to keep something. It cannot update or delete existing knowledge.
CREATE-ONLY- Search indexed Markdown
- Read one exact result
- Create one explicit AI Capture
- Confirm exactly what happened
- Discover or execute arbitrary workflows
- Read credentials or provider configuration
- Refresh, publish, update, or delete the vault
- Bypass local validation or human ownership
An AI Capture is not automatically accepted as truth. It first enters a create-only inbox. After it reaches the original local vault, the Mac validates its metadata and may extract supported durable context at lower model-interpretation authority.
Automation stops when it cannot prove a change is safe.
Your notes stay yours
Obsidian remains the source of truth. Automation proposes and validates; it does not silently seize ownership.
AI is not the final authority
AI models return suggestions. Clear code checks the allowed scope, evidence, links and current file state before anything can change.
No change means no model call
Frequent checks stop early when nothing is new, so freshness does not automatically mean more token usage.
Every part can fail safely
A connector, model or summary failure preserves the last valid state instead of replacing it with an empty or uncertain result.
Useful now.
Still getting smarter.
Available now
Local Apple Notes sync, Obsidian as the write authority, source lineage, explicit promotion queues, cross-domain maps, read-only context collection, private search, model routing, insight digests, health reporting, and verified backups.
Owner-guided next
Reviewing proposed knowledge, rating useful insights, and choosing which connections change a decision remain deliberately human actions.
Further ahead
A productised Apple Notes Sync app, broader connector packs, safer collaborative review, easier self-hosted installation, and more useful knowledge-health and reflection surfaces.
Make saved knowledge more useful without adding more maintenance.
Second Brain is a self-hosted system for people who want AI help without giving up their notes, judgement or privacy.
Explore Drew's Digest ↗ View the source on GitHub ↗