Docuflow¶
Persistent knowledge bases for AI agents — built once, compounding forever.
The problem AI agents face¶
Every new conversation, your AI agent starts from zero. It re-reads files, re-discovers patterns, re-learns your codebase. Hours of context disappear the moment the session ends. Even RAG re-extracts knowledge on every query — there's no accumulation, no memory, no compounding value.
The solution: LLM Wiki pattern¶
Docuflow implements the LLM Wiki pattern — a three-layer architecture where knowledge compounds over time:
You add a source document
↓
Docuflow reads it once → extracts entities, concepts, relationships
↓
Updates the wiki → cross-references, integrates, tracks contradictions
↓
Next query: search the richer wiki (not raw source again)
↓
Answer gets better as wiki grows — knowledge compounds
The LLM does the bookkeeping once. That work reuses forever.
Quick install¶
What you get¶
🧠 15 MCP Tools
Code extraction, wiki management, health checks, dependency graphs — all accessible from any MCP-compatible agent (Claude, Copilot, Cursor).💻 8 CLI Commands
docuflow init, ingest, query, status, rewiki, suggest, ui, doctor — and more via docuflow advanced --help. 🌐 Web UI
Live browser interface with 6 views: Ask, Wiki, Graph, Health, Sync, Onboard. No build step — ships inside the npm package.📦 4 Packages
@doquflow/core (4 core tools), @doquflow/studio (11 advanced tools + UI), @doquflow/server (back-compat alias), @doquflow/cli (meta CLI). 🔍 Universal Extractor
Understands TypeScript, JavaScript, Python, Go, Ruby, Rust, Java, C#, PHP, Kotlin, Swift, Angular, Vue, SQL, Shell, and more.🔐 Waymark Integration
Optional security policy enforcement. AI operates only within allowed paths and commands, with full audit trail and human approval for sensitive operations.Philosophy¶
Knowledge should accumulate, not evaporate.
Docuflow is built on the belief that AI-assisted development should get better over time, not start fresh every session. Every source you ingest makes every future question more accurate. Every question answered makes future answers richer.