WikiMind — Feature Evidence & Capabilities

WikiMind is a personal LLM-powered knowledge OS. Drop in articles, papers, PDFs, or YouTube links. The LLM compiles them into structured wiki articles. Ask questions against the wiki. The system detects contradictions, orphans, and knowledge gaps. Synthesis generates cross-cutting analyses across sources.

Backend: FastAPI + PostgreSQL + Redis + ARQ workers. Frontend: React + TypeScript + Vite. Deployed on Fly.io with Docker.

Supported formats

API Evidence — Ingest arxiv paper

$ curl -X POST http://localhost:7842/api/ingest/url \
    -H "Content-Type: application/json" \
    -d '{"url":"https://arxiv.org/abs/1706.03762"}'

{
  "source_type": "url",
  "title": "Attention Is All You Need",
  "source_url": "https://arxiv.org/abs/1706.03762",
  "id": "393d1cac-83d0-4ec9-88a3-8b43b0f607bc",
  "ingested_at": "2026-05-18T19:28:27",
  "compiled_at": null   ← compiling in background
}

Key capabilities

• LLM-compiled articles with key claims, confidence scores, and source citations
• Concept taxonomy auto-extracted from content
• Bidirectional backlinks ([[wiki links]]) resolved across articles
• Interactive draft review: approve or reject compilations before publishing
• Custom compilation schemas for different article types
• Faceted search: filter by concepts, tags, confidence, page type
• User-defined tags and saved searches

Articles rendered with full rich content support: extracted PDF figures and tables, LaTeX math, syntax-highlighted code blocks, and sanitized HTML. Images extracted via Docling are embedded inline with captions.

Preview → Refine → Confirm workflow

POST /api/wiki/synthesis/preview
{
  "article_ids": ["bert","gpt","transformer"],
  "synthesis_type": "thematic",
  "guidance": "Focus on attention"
}

Returns draft without saving

POST /api/wiki/synthesis/refine
{
  "draft_content": "<previous>",
  "article_ids": [...],
  "guidance": "More detail on
               cross-attention"
}

Iterates with feedback

POST /api/wiki/synthesis/confirm
{
  "title": "Attention Mechanisms",
  "draft_content": "...",
  "article_ids": [...]
}

Saves as real wiki article

API Evidence — Comparative synthesis across 3 arxiv papers

POST /api/wiki/synthesize
{ "query": "Compare transformer architectures", "synthesis_type": "comparative",
  "article_ids": ["bert-id", "gpt-id", "transformer-id"] }

{
  "title": "Comparative Analysis of Transformer Architectures: BERT, GPT-3, and Original Transformer",
  "themes": ["Bidirectional vs. Autoregressive Models", "Attention Mechanisms", "Task Versatility"],
  "source_count": 3,
  "page_type": "synthesis"
}

API Evidence

# Markdown export
GET /api/wiki/articles/{id}/export?format=markdown

HTTP/1.1 200 OK
content-type: text/markdown; charset=utf-8
content-disposition: attachment; filename="bert-...md"

---
title: "BERT: Pre-training of Deep Bidirectional..."
concepts: [language representation, NLP, transformers]
---

## Summary
BERT is a language representation model...

# Share link with 7-day expiry
POST /api/wiki/share-links
{ "article_id": "d65e8c6b-...", "expires_in_days": 7 }

{
  "token": "o8K1dltPkbfFT-c48...",
  "expires_at": "2026-05-25T19:29:53",
  "view_count": 0,
  "article_title": "BERT: Pre-training..."
}

# Full wiki export
POST /api/wiki/export/wiki
Formats: obsidian, markdown_json

Automated wiki quality checks run on demand or via background jobs. Detects contradictions, orphaned articles, stale content, and missing links.

POST /api/lint/run          # Trigger lint
GET  /api/lint/reports       # List reports
GET  /api/wiki/contradictions # List contradictions with status
PATCH /api/wiki/contradictions/{id} # Resolve or dismiss

Exposes WikiMind as an MCP server for Claude Desktop, Cursor, and other AI agents. Two transports: stdio (local) and HTTP (JWT-authenticated remote).

# Start MCP server
$ wikimind mcp serve
$ wikimind mcp serve --transport http

# Claude Desktop config
$ wikimind mcp config

$ wikimind status                              # Check server and auth
$ wikimind login                               # Magic link login
$ wikimind whoami                              # Show authenticated user

$ wikimind ingest url https://arxiv.org/abs/1706.03762  # Ingest URL
$ wikimind ingest file paper.pdf               # Ingest PDF
$ wikimind ingest text "Some content"          # Ingest text

$ wikimind ask "How does BERT work?"           # Ask against wiki
$ wikimind ask --conversation abc123 "And GPT?" # Continue conversation

$ wikimind wiki list                           # List articles
$ wikimind wiki list --concept transformers    # Filter by concept
$ wikimind wiki show bert-pre-training-of...   # View article
$ wikimind wiki delete some-article            # Delete (with confirmation)

$ wikimind mcp serve                           # Start MCP server (stdio)
$ wikimind mcp serve --transport http          # HTTP transport
$ wikimind mcp config                          # Print Claude Desktop config

Automated test of 138 API endpoints. Idempotent script creates test data and cleans up after itself. Endpoints that trigger LLM calls, require browser OAuth, or are destructive are skipped intentionally.

Run: ./api-test.sh  |  View script source  |  Standalone results page