Brain

A personal second brain agent written in Go. Captures thoughts from multiple sources (Discord, relay, web UI), classifies them using AI, and stores them in SQLite with semantic vector search via chromem-go.

Ecosystem

Brain is part of a three-component system:

Component	Location	Purpose
brain.exe (this repo)	scripts/brain/	Local brain — capture, classify, store, search
ibeco.me	scripts/becoming/ (in scripture-study)	Cloud hub — relay, web UI, practices, journaling. Deployed via Dokploy on VPS
brain-app	scripts/brain-app/ (separate git repo)	Flutter app — Android, Windows (iOS/Mac planned)

brain.exe is the authoritative data store. ibeco.me connects via WebSocket relay and caches entries for web access. brain-app connects to either brain.exe directly or through the ibeco.me relay.

Architecture

[Discord DM]  ──┐
[ibeco.me app] ──┤──→ [Go Brain Binary] ──→ [SQLite DB] + [chromem-go vectors]
[Web UI]       ──┘         ↑                       ↓ (optional)
                     [LM Studio / Copilot]    [Archive → private-brain git repo]

Building Blocks (from Nate B Jones)

Block	Implementation
Dropbox (capture)	Discord DM, ibeco.me relay, web UI, CLI
Sorter (classifier)	LM Studio / Copilot SDK → structured JSON
Form (schema)	SQLite columns with category-specific fields
Filing Cabinet (storage)	SQLite + chromem-go vector embeddings
Receipt (audit trail)	audit_log table in SQLite
Bouncer (confidence filter)	Configurable threshold; low-confidence → inbox
Tap on Shoulder (surfacing)	Morning digest, weekly review (Phase 2)
Fix Button (correction)	fix: <category> in Discord, reclassify via web/API
Search (retrieval)	Full-text search + semantic vector search

Categories

Category	For
people	Relationship context, follow-ups, details about someone
projects	Active work with status and next actions
ideas	Thoughts, insights, concepts to explore
actions	Tasks, errands, things with a "done" state
study	Scripture insights, spiritual impressions, gospel learning
journal	Personal reflections, observations, becoming

Setup

Prerequisites

1. Go 1.21+
2. One of:
   • LM Studio running locally (default — free, private)
   • GitHub Copilot CLI installed and authenticated
3. Optional: A Discord bot token (create one)
4. Optional for semantic search: LM Studio with an embedding model, or Ollama with nomic-embed-text

Configuration

cp .env.example .env
# Edit .env with your tokens and preferences

Key settings:

• AI_BACKEND — lmstudio (default) or copilot
• EMBEDDING_BACKEND — lmstudio, ollama, openai, or none
• WEB_ENABLED / WEB_PORT — web UI (default: true / 8445)
• BRAIN_DATA_DIR — where SQLite + vectors live (default: ~/.brain-data)
• BRAIN_ARCHIVE_DIR — optional private-brain repo for markdown export

Build & Run

go build -o brain.exe ./cmd/brain
./brain.exe

Open http://localhost:8445 in your browser, or send a DM to your Discord bot.

Commands (Discord)

Command	Action
(any text)	Capture and classify a thought
fix: <category>	Reclassify the last entry
model	List available AI models
model: <name>	Switch AI model (gpt-mini, haiku, sonnet)
status	Show brain status and entry counts
stop	Pause autonomous processing

REST API

Endpoint	Method	Description
/api/entries	GET	List entries (optional ?category=)
/api/entries	POST	Create entry
/api/entries/{id}	GET	Get single entry
/api/entries/{id}	PUT	Update entry
/api/entries/{id}	DELETE	Delete entry
/api/entries/{id}/reclassify	POST	Move to new category
/api/search?q=	GET	Full-text search
/api/search/semantic?q=	GET	Semantic vector search
/api/stats	GET	Category counts
/api/tags	GET	Tag frequencies
/api/archive	POST	Export entry to markdown archive

AI Models

Powered by local LM Studio (default) or GitHub Copilot SDK.

Preset	Model	Premium Rate
gpt-mini (default)	gpt-5-mini	0x (free)
haiku	claude-haiku-4.5	0.33x
flash	gemini-3-flash	0.33x
sonnet	claude-sonnet-4.6	1x
gpt5	gpt-5	1x

Switch at runtime via Discord: model: haiku Or set in .env: AI_MODEL=sonnet Or use any Copilot model ID directly: AI_MODEL=claude-opus-4.6

Phases

• Phase 1: Core loop — capture → classify → store → git commit
• Phase A: ibeco.me relay integration — WebSocket transport
• Phase B: Relay client — brain.exe ↔ ibeco.me hub
• Phase C: Flutter mobile app
• Phase D: Integration tests
• Phase E: CLI tool (brain-cli)
• Phase F: SQLite + chromem-go — shared memory, web UI, semantic search
• Phase 2: Proactive surfacing — morning digest, weekly review
• Phase 3: Migration — import existing private-brain markdown
• Phase 4: Self-improvement — audit analysis, improvement proposals

License

MIT