BrainTree

A native SwiftUI iOS app for capturing, browsing, and querying a personal knowledge base backed by Supabase and an MCP (Model Context Protocol) server. Voice thoughts go in via the Action Button or mic tap; Claude synthesizes answers when you ask questions.

Interface

Feed

Query

Capture

Settings

Widget

What it does

Three surfaces, one backend:

Surface	What it does
Feed	Chronological stream of all captured thoughts, filterable by source (Bear, Obsidian, email, Slack, iOS)
Ask	Semantic vector search over your notes — returns matching results and a Claude-synthesized summary
Capture	Tap mic → speak → auto-submits after 1.5s silence, or type manually

Home screen widget — small/medium donut chart showing thought counts by source for the last 7 days, refreshes every 30 minutes.

Action Button integration — assign the "Capture Thought" shortcut to the iPhone Action Button. One press opens the app directly to the Capture tab. Also available as a Siri shortcut ("Capture a thought with Open Brain").

Architecture

The app is a stateless thin client — no local database, no local cache beyond widget display data. All thoughts live in Supabase.

BrainTree/
├── Models/        Thought, SearchResult, APIError
├── Services/      KeychainClient, SupabaseClient, MCPClient, AnthropicClient
├── Views/         FeedView, AskView, CaptureView, SecretsSetupView, MarkdownView
├── AppIntents/    CaptureThoughtIntent + BrainTreeShortcuts
├── ContentView.swift
└── BrainTreeApp.swift

BrainTreeWidget/   WidgetKit extension (self-contained, native Charts)

Backend calls

Operation	Endpoint
Capture thought	POST /functions/v1/open-brain-mcp → capture_thought
Load feed	GET /rest/v1/thoughts (Supabase REST)
Discover sources	POST /functions/v1/open-brain-mcp → thought_stats
Semantic search	POST /functions/v1/open-brain-mcp → search_thoughts
Filter by tag/type	POST /functions/v1/open-brain-mcp → list_thoughts
AI synthesis	POST https://api.anthropic.com/v1/messages (Claude Haiku)

All MCP calls use JSON-RPC 2.0 and receive Server-Sent Events (SSE) responses.

Ask tab routing

• Text only → search_thoughts (semantic vector search) → Claude Haiku synthesis
• Tag or type filter (with or without text) → list_thoughts; if text is also present, synthesize client-side over filtered results
• Supports combining text query + tag filter + type filter simultaneously

Deep links

Search results and feed items link back to the source app:

• Bear → bear://x-callback-url/open-note?id=<uuid>
• Obsidian → obsidian://open?vault=BrainTree&file=<folder/title>

Credentials & storage

All four API keys are stored in the iOS Keychain. The Supabase URL and key are additionally written to App Group UserDefaults (group.com.dyerlab.openbrain) so the widget extension can read them. No credentials leave the device.

Required keys (enter in Settings → API Keys):

• Supabase URL
• Supabase service role key
• MCP access key (x-brain-key header)
• Anthropic API key

Building

Open BrainTree.xcodeproj in Xcode. Requires iOS 17+.

# Build for simulator
xcodebuild -scheme BrainTree -configuration Debug -sdk iphonesimulator build

# Run unit tests
xcodebuild test -scheme BrainTree -sdk iphonesimulator \
  -destination 'platform=iOS Simulator,OS=26.4,name=iPhone 17'

# Static analysis
xcodebuild analyze -scheme BrainTree -sdk iphonesimulator

Unit tests cover SearchResult.decode — JSON array, wrapped JSON, SSE text-block parsing, deep links, and similarity label formatting.

First launch

On first launch the API Keys sheet appears automatically. Fill in all four fields and tap Save. The widget will start showing data once keys are saved (it reads Supabase credentials from the App Group).

Dependencies

• WidgetKit + Charts (system frameworks) — widget donut chart
• PresentationZen v1.0.11 (SPM, main app target only) — available for pie charts in main app views if needed
• No other third-party dependencies