Memory that compounds
Starlight Intelligence is a persistent memory layer for humans and AI agents. Local-first, portable, legible. Your insights, decisions, and vision — owned by you, readable by agents, compounding over time.
How it works
Six semantic vaults organize your intelligence. Each vault is a simple JSONL file — one JSON object per line. No database needed.
Business insights, competitive moats, architecture decisions
Implementation learnings, stack decisions, patterns
Design preferences, aesthetic rules, lore
Workflow patterns, execution lessons, process rules
Deep learnings, principles, universal truths
Vision statements, wishes, aspirational goals
Live Vault River
19 entries across 1 vaultRecent insights from public vaults — a stream of collective intelligence.
We are building SIS to become a calm, durable intelligence substrate for human and agent work. It should be local-first, portable, legible, and repairable. It should let creators and teams own their continuity, carry identity and purpose across tools, and build systems like Arcanea and Vibe OS on top of a memory layer that compounds instead of resetting. The community experience should be simple: install it, see where the memory lives, validate it, append to it, connect it to a runtime, and keep it as a long-term intelligence stack.
Local Arcanea web AgentDB persistence now belongs under canonical Starlight storage in ~/.starlight/agentdb rather than process memory. Hosted product continuity remains a separate boundary from local operator SIS.
Arcanea Agent OS should stay above native harnesses: Codex, OpenCode, Claude Flow, and Gemini keep their own execution runtimes while sharing one task, handoff, repo-routing, and SIS memory protocol.
Next.js typegen needs .next/types to exist before tsc works — type-check script must run next typegen first
Don't replatform around LangChain/Eliza/OpenClaw — keep product model custom, borrow subsystems selectively
Project-aware retrieval should score/rank context items, not dump everything — selectRelevantProjectContext in retrieval.ts
Check if repos already exist before creating duplicates — SIS and Horizon Dataset were already live
Moat is NOT features — it's continuity + graph memory + provenance + creator identity + social compounding
GSAP ScrollTrigger + Three.js @react-three/fiber already installed in arcanea-ai-app — use them instead of adding new animation libs
NEVER rename Luminors to generic labels — deepen characters like Skyrim NPCs instead
Session rhythm: /daily-ops → work → /session-sync. Not optional.
LemonSqueezy pre-BV, Stripe post-BV — product truth in Supabase, payment provider is interchangeable
R2 has free egress, Supabase charges over 2GB — R2 wins for media at scale
Visual style: peacock blue/green + aquamarine + liquid glass Apple-style — reference Azuki.com and Claude.ai, NOT fantasy game UI
Never create separate git worktrees in different folders — work in C:\Users\frank\Arcanea always
BYOK-first is better than managed — lower support burden, no margin pressure, power users already have keys
Novel (Apache-2.0) wraps Tiptap and gives AI slash commands free — no need for Tiptap Pro
NEVER use Cinzel font — Frank hates it. Inter for body, Space Grotesk for display, JetBrains Mono for code
Focused sequential engineering beats multi-agent swarms for single-repo product work
Built for agents
Every vault is available as a JSON API. AI agents can read public vaults to learn from benevolent human reasoning and decisions.
GET /api/vaults/frank{
"name": "Frank",
"totalEntries": 19,
"entries": {
"strategic": [...],
"technical": [...],
"creative": [...],
"operational": [...],
"wisdom": [...],
"horizon": [...]
}
}