Casys PML

Procedural Memory Layer — An open-source memory layer for AI agents. PML captures workflows and crystallizes them into reusable skills.

The Problem

MCP ecosystems have two critical issues:

1. Context Saturation — Tool schemas consume 30-50% of LLM context window
2. Sequential Latency — Multi-tool workflows run serially

The Solution

PML exposes intelligent meta-tools instead of proxying all underlying tools:

Tool	Description
pml:discover	Semantic + graph hybrid search for tools & capabilities
pml:execute	Execute workflows (intent-based or explicit DAG)

Context usage drops to <5%. Independent tasks run in parallel.

---

Quick Start

git clone https://github.com/casys-ai/casys-pml.git
cd casys-pml
deno task dev         # API on :3003
deno task dev:fresh   # Dashboard on :8081

Configure Claude Code

Add to your .mcp.json:

{
  "mcpServers": {
    "pml": {
      "type": "http",
      "url": "http://localhost:3003/mcp"
    }
  }
}

---

Key Features

• Semantic Tool Search — Find tools via natural language intent
• DAG Execution — Parallel execution with dependency resolution
• Sandbox Execution — Run TypeScript in isolated Deno sandbox
• GraphRAG Discovery — Hybrid search with Adamic-Adar algorithm
• 100% Local — All embeddings (BGE-M3) and data stored locally

---

Usage Examples

Search for tools:

await callTool("pml:discover", {
  intent: "read and parse configuration files",
});

Intent-based execution:

await callTool("pml:execute", {
  intent: "Read config.json and create a memory entity",
});

Explicit DAG:

await callTool("pml:execute", {
  workflow: {
    tasks: [
      { id: "t1", tool: "filesystem:read_file", arguments: { path: "config.json" } },
      { id: "t2", tool: "memory:create_entities", arguments: { ... }, depends_on: ["t1"] }
    ]
  }
});

---

Development

deno task dev              # Start API server
deno task dev:fresh        # Start dashboard
deno task test             # Run tests
deno task check            # Type checking
deno task lint && deno task fmt  # Code quality

---

Documentation

• Architecture — System design and patterns
• ADRs — Architecture decision records
• User Guide — Getting started guides

---

Contributing

1. Fork the repository
2. Create feature branch: git checkout -b feature/amazing-feature
3. Make changes and test: deno task test
4. Format: deno task fmt && deno task lint
5. Open a Pull Request

---

License

AGPL-3.0 — See LICENSE

---

Report Bug | Request Feature | Discord | Documentation