Stinger Industries has released Adamanteum Community Edition, a local-first knowledge management tool that serves as a bridge between human-readable note-taking and AI-accessible data structures. The MIT-licensed application creates a live, draggable map of your vault while simultaneously exposing a Model Context Protocol (MCP) server, allowing any compatible AI to search and read your notes without leaving your machine.

Dual-Interface Architecture

The platform distinguishes itself by treating human and AI interfaces as equal citizens. While tools like Obsidian prioritize the human experience and pure LLM wrappers prioritize the model, Adamanteum runs a single backend that serves both. The human interface features a "Document Lens" for reading raw source or rendered pages directly from the map, alongside a notebook system with revisions and daily notes. For AI interaction, the tool does not ship with a proprietary model; instead, it acts as a secure door. Developers can mint read-only tokens for specific AI clients, such as Claude Desktop, Cursor, or Windsurf, ensuring that the AI can query the vault but cannot write to it unless explicitly configured.

Local-First Performance and Setup

Built on Node.js 24+, Adamanteum requires no cloud account or external service. The installation process involves cloning the repository and running npm start, which launches a local server at http://127.0.0.1:4173. The tool is designed for heavy workloads, with layout calculations running in a worker thread to maintain responsiveness in large vaults. In community benchmarks conducted on a Windows laptop, indexing 10,000 notes from scratch took 10.4 seconds, while single-word and quoted phrase searches averaged around 3 milliseconds. The tool supports failover configurations for AI providers, remembering dead endpoints for 10 minutes to prevent repeated connection attempts.

Code Graphs and Language Support

Beyond standard Markdown notes, Adamanteum parses code repositories into queryable graphs. It supports 30 languages via tree-sitter grammars, mapping functions, classes, calls, and imports into a visual network. The CLI allows developers to scan projects, query specific nodes, and export the resulting graph in formats like JSON, GraphML, Mermaid, and SVG. Crucially, the tool parses code statically without executing it, maintaining security and speed. However, the developers note that cross-language call resolution is not yet fully semantic, and unresolved calls remain labeled as such rather than being inferred incorrectly.

Key Takeaways

  • Local-First Security: All data remains on the user's machine; AI access is mediated by revocable, read-only MCP tokens.
  • Performance: Indexing 10k notes takes ~10s, with search latency averaging 3ms on local hardware.
  • AI Agnostic: Supports any OpenAI-compatible endpoint (Ollama, LM Studio, OpenAI) with automatic failover.
  • Code Intelligence: Visualizes code structures in 30 languages without running the code, supporting exports to GraphML and Mermaid.
  • License: Released under the MIT license by Stinger Industries (Ohh BeeHave LLC).

The Bottom Line

Adamanteum solves the friction of exposing personal knowledge bases to AI agents by providing a standardized, secure MCP interface that doesn't sacrifice human usability for model convenience.

Limitations to Watch

The current release is not a full Obsidian replacement. Editing notes in place lacks full parity, with the notebook existing as a separate SQLite store. Tags and properties are basic, and nested note embeds are incomplete. The Canvas editor is currently import/export only. These caveats suggest the tool is best suited right now for users who primarily consume and query their data, rather than those who require a robust, all-in-one editing environment.