The next-generation knowledge infrastructure engine.
Realizing Vannevar Bush's 1945 Memex vision with modern LLMs.
trail is a knowledge infrastructure engine that turns your sources into a persistent, compounding wiki — maintained by an LLM on your behalf.
Unlike traditional RAG, which fragments your documents and retrieves chunks at query time, trail compiles your knowledge into a structured wiki at ingest time. Every new source makes the whole system smarter. Good chat answers feed back into the wiki. Contradictions get flagged. Orphans get linked.
It's what Vannevar Bush dreamed of in 1945, finally made practical by LLMs.
- Personal knowledge bases — research, reading, learning
- Domain expertise platforms — turn an expert's writings into an AI-powered digital presence
- Internal knowledge — Slack threads, meeting notes, wiki pages into one compiled whole
- Publisher content — AI-native article platforms
- CMS modules — drop-in AI brain for any website
Phase 1 MVP in progress. First customer: Sanne Andersen (healing practice with 25 years of clinical material). See docs/PLAN.md.
trail is a pnpm + Turbo monorepo:
trail/
├── apps/
│ ├── server/ # Hono API (core engine)
│ ├── admin/ # Curator dashboard (Vite+Preact)
│ ├── mcp/ # MCP server for LLM integrations
│ └── widget/ # Embeddable <trail-chat> web component
├── packages/
│ ├── core/ # Ingest, compile, query, lint
│ ├── db/ # Drizzle + SQLite schema
│ ├── storage/ # Local FS + R2 abstraction
│ ├── llm/ # Multi-provider adapter
│ ├── pipelines/ # PDF, vision, web, video
│ └── shared/ # Types + Zod schemas
└── adapters/ # CMS adapters (Phase 2+)
FSL-1.1-Apache-2.0 — Functional Source License, converts to Apache 2.0 after 2 years.
You can use trail freely for your own knowledge bases, research, and internal tools. You cannot offer trail as a competing commercial SaaS. See LICENSE for details.
- trailcloud.com — managed SaaS (coming soon)
- trail.wiki — documentation (coming soon)
- @webhouse/cms — CMS with trail adapter (Phase 2)