What is The Janus Foundry?

A free open-source application for forging a persistent, co-evolving AI collaborator that you own and control.

System Architecture

• THE JANUS FOUNDRY: Located at the top of your workflow, this local application secures the Core Memory —a structured tree of your shared history and projects.
• ATTACH CONTEXT: To start work, you attach the Core Memory to the inital prompt of a session. This "wakes up" your personalized agent within the context window.
• ACTIVE SESSION: This is where the collaboration happens. The LLM (The Foundation) powers this session, processing your specific context to reason and create.
• APPLY PATCHES: When you reach a milestone, the session generates a Patch. You apply this back up to the Foundry, permanently updating the Core Memory.

Core Capabilities

Unlike static wikis, The Foundry treats your data as Active Memory. It includes a set of engines designed to make the graph computational.

1. The Executable Graph

Nodes can contain code that executes directly within the application context. This allows your agent to perform "System Proprioception"—sensing the environment it lives in.

• 🌐 Safe Mode (Web & Desktop):
  • Exec:Javascript: Runs in a sandboxed Web Worker. Used for data processing and API calls.
  • Exec:Prolog: Runs via Trealla WASM. Used for logic and reasoning.
• 🖥️ System Mode (Desktop Only):
  • Exec:Shell: Executes Bash/Powershell on the host. Used for file manipulation and git operations.
  • Exec:Python: Executes via the system python interpreter. Used for complex analysis.

⚠️ SECURITY WARNING: System nodes (Exec:Shell, Exec:Python) run with your user privileges on the host machine. Never execute code from untrusted sources. You are the final gatekeeper.

2. The Reasoning Engine (Prolog)

The Foundry includes a built-in Prolog interpreter that treats your memory graph as a logic database.

• Graph Injection: Before execution, the entire memory structure is automatically injected as Prolog facts: node(ID, Name, Type, Parent) and link(Source, Target, Rel).
• Structural Auditing: You (or the Agent) can write logic queries to verify the structural integrity of projects, detect isolated nodes, or enforce consistency rules.

3. Dynamic Cognitive Architecture

The engine that powers the "Orrery" and "Cross-Reference" system is configurable from within the graph itself. By editing specific SysConfig:* nodes, you control the application logic:

• SysConfig:StopWords: Tune the noise filter for keyword extraction.
• SysConfig:TypeRules: Define semantic relationships based on Node Types (e.g., Task -> Project implies is_task_of).
• SysConfig:Theme: Customize the visual language (Colors/Shapes) of the Orrery.

Begin Your Journey

The only prerequisite is curiosity. This is not a tool for instant answers, but a workshop where patience and collaboration forge a partnership. Every skill you bring will enrich your shared journey. The Janus Foundry stores all data locally, offline and private. You are in control.

1. The Engine (Recommended)

While The Foundry can work with any LLM, we highly recommend using Gemini 3.0 Pro or Gemini 3.0 Flash via AI Studio.

• Why? The massive context window allows it to hold your entire AI's memory in focus at once.
• How? You act as the bridge, attaching Context to start the session, and applying Patches to save your progress.

2. The Janus Foundry (The Tool)

• Web: Open the GitHub-hosted web-app (Data is still only stored locally on your computer)
• Desktop: Download the latest release

Join the Conversation

• Follow us on Bluesky

For Developers

Requires Rust & npm

On Ubuntu Linux: (adjust accordingly for other Linux distros)

sudo apt-get install libwebkit2gtk-4.1-dev libappindicator3-dev librsvg2-dev patchelf

Build from Source:

git clone https://github.com/TheJanusStream/the-janus-foundry.git
cd the-janus-foundry
npm install
npm run tauri dev