Raphael

Overview

Raphael is a fully self-hosted, AI-powered knowledge management system that captures Discord conversations and transforms them into a searchable, interconnected wiki — driven by an evolving AI persona inspired by the Great Sage from That Time I Got Reincarnated as a Slime. It acts as a digital memory for Aylian Studios that grows smarter every day.

The system listens to Discord, identifies valuable knowledge within discussions, generates comprehensive wiki pages automatically, discovers connections between topics across channels, and provides a clean interface to explore everything captured.

From the Original Plan

The original Raphael concept was scoped as the operating system of Aylian Studios — a voice and text interface for controlling studio infrastructure (deploying services, checking server health, managing workflows), integrated into the BOSS dashboard. It had three defined visual states: a Blue Halo for idle, a spinning red/gold geometric mandala for processing, and a waveform visualizer for speaking. The persona was hyper-competent, slightly arrogant, and fiercely loyal — addressing the user as "Master" or "Partner."

That version was blocked on BOSS reaching sufficient maturity. Rather than wait, development pivoted to Discord as the interface layer — a platform already in active daily use — and the scope shifted to knowledge capture and wiki generation as a concrete, immediately useful foundation. The Great Sage archetype (analytical, advisory, grows with accumulated knowledge) proved a better fit than the studio OS framing for this stage of the project.

The three visual states and BOSS integration remain part of the long-term vision. They are designed and waiting on the right surface to build them into.

Architecture

RS1000 (Proxy & Automation)
├── Nginx Proxy Manager  →  routes wiki domain to RS4000
└── n8n                  →  Raphael router, shopping agent, Discord webhooks

RS4000 (Docker Compose — Primary Stack)
├── PostgreSQL 16 + pgvector   database + vector search
├── Redis 7                    caching & sessions
├── Rust Backend (Axum)        REST API + AI processor + auth
├── Next.js Wiki               wiki frontend
└── Discord Bot (Raphael)      Bun + Discord.js

RS4000 (Docker Compose — LocalAI Stack, optional)
└── LocalAI                    CPU or GPU profile, port 8088

The Rust backend handles all business logic, AI processing, authentication, and data management. The Discord bot captures messages and commands, while n8n workflows handle message routing and automation. LocalAI provides a self-hosted AI runtime with a cloud fallback chain: LocalAI → OpenCode → MiniMax → Kimi.

Changelog

v1.0 — Infrastructure Foundation

PostgreSQL 16 + pgvector schema with 9 tables: messages, pages, page_relations, embeddings, skills, page_evolution, page_corrections, processing_logs
Docker Compose stack with internal network 172.21.0.0/16 and health checks on all services
Redis 7 with AOF persistence
Rust backend scaffold (Axum 0.7, SQLx, Tokio)
Basic Next.js wiki shell

v2.0 — AI Integration

AIClient with multi-provider fallback chain (LocalAI → OpenCode → MiniMax → Kimi)
Page generation pipeline: Discord messages grouped by semantic similarity, analyzed by AI, written as structured wiki pages with cross-links
1024-dimensional vector embeddings stored via pgvector
Cross-page relationship discovery (semantic, hierarchical, reference)
Daily cron job at 1 AM CET via tokio-cron-scheduler

v2.1 — Wiki UI

Full Next.js wiki with professional dark theme and Lucide icons throughout
Responsive layout with backlinks display and knowledge graph visualization
Markdown rendering with syntax highlighting and command palette (Ctrl+K)
Rust backend logs cleaned up with structured tracing
n8n workflow audit and updates

v2.2 — Autonomous Research Agents

ResearchAgent in ai.rs runs background research tasks without blocking Discord
POST /api/agent/research triggers complex multi-step queries asynchronously
GET /api/agent/status/:id for polling long-running jobs

v2.3 — Multi-Modal Vision

Discord bot extracts image attachment URLs from messages
AIClient::analyze_image() sends images to vision model (LocalAI first, OpenCode fallback)
Image descriptions injected into the message processing pipeline and reflected in generated wiki pages

v2.4 — Auth, LocalAI, Plugins & Backfill

Bearer token auth middleware in Axum — public and protected routers separated; webhook routes remain public for internal Docker network calls
Global Axios interceptor in the Discord bot injects Authorization: Bearer <API_KEY> on all Rust backend calls
Historical backfill on bot startup: fetches up to 2,000 messages per tracked channel, safe to re-run via ON CONFLICT DO NOTHING
LocalAI as a separate Docker Compose stack with [cpu] and [gpu] profiles on port 8088, joins the main network as an external network
AliExpress stub endpoints (/api/shop/aliexpress, /api/shop/track) implemented as protected routes
Plugin list endpoint (/api/plugins) — dynamic loading stubbed for Phase 7

Current Status

The system is running in Docker on RS4000 and in active testing as of March 2026. All six phases are complete and all services are online.

| Component | Status | |---|---| | PostgreSQL 16 + pgvector | Running | | Redis 7 | Running | | Rust Backend (Axum) | Running — 30+ endpoints | | Next.js Wiki | Running — rebuild pending for API URL fix | | Discord Bot (Raphael) | Online — 10 slash commands, backfill working | | LocalAI | Optional — CPU profile available | | n8n Workflows | Running |

What's Next — v2.5

Soul Container: Persona middleware in api.rs that shapes Raphael's raw LLM output before it reaches Discord — injects "Notice, ..." prefixes on declarative statements, enforces first-person, removes AI hedging phrases
unspeech Voice Proxy: moeru-ai/unspeech as an optional Docker service that normalizes STT/TTS backends (Whisper, Piper, ElevenLabs) behind a single OpenAI-compatible endpoint — enables Phase 8 voice work without coupling to any one provider
Wiki Container Rebuild: Apply the NEXT_PUBLIC_API_URL fix so the browser-side wiki correctly targets the internal API endpoint

Roadmap

[x] Phase 1 — Infrastructure Foundation
[x] Phase 2 — AI Integration
[x] Phase 3 — Wiki UI
[x] Phase 4 — Autonomous Research Agents
[x] Phase 5 — Vision Processing
[x] Phase 6 — Auth, LocalAI, Plugins, Backfill
[ ] v2.5 — Soul Container, Voice Proxy, Wiki Fix
[ ] Phase 7 — Dynamic Plugin Ecosystem
[ ] Phase 8 — Multi-Modal (Voice, YouTube Summaries)
[ ] Phase 9 — Scale & Export (Obsidian vault, enterprise features)
[ ] BOSS Integration — Visual states, studio ops control surface