M.A.R.S

Multiple Agent Runtime System

Overview

M.A.R.S is the command-and-control operating system for your autonomous network. It allows you to orchestrate your entire agentic OS via a unified CLI. M.A.R.S acts as the central router between your left-brain procedural logic (LangChain execution flows) and your right-brain spatial memory (Graphiti storage networks), enabling you to construct deterministic, localized agent swarms that retain persistent corporate memory.

M.A.R.S Topology


                      [ Deno Shadow Brain (Hot State) ]
                                     │
         ┌───────────────────────────┴───────────────────────────┐
         ▼                                                       ▼
[ pgvector (Warm State) ]                             [ Neo4j (Cold State) ]
- Semantic Embeddings                                 - Entity Relationships
- High-speed similarity search                        - Complex path traversal
- Python / Graphiti ingestion                         - Knowledge graph mapping
         │                                                       │
         └───────────────────────────┬───────────────────────────┘
                                     ▼
                       [ Soul Journal (Immutable Log) ]
                       - Append-only event store
                       - Cryptographically hashed actions

Hardware Constraint:Running the full Tier 1 stack locally requires significant memory overhead due to the JVM (Neo4j) and Postgres operating simultaneously. Minimum 16GB RAM recommended; 32GB required for high-velocity swarm emulation.

The Execution Path

Observation: The Agent Development Kit (ADK) receives input.
Synthesis: The Python Graphiti engine generates vector embeddings (pgvector) and extracts entity relationships (Neo4j).
Retrieval: The Deno runtime queries both databases simultaneously, fetching semantic similarities and traversing graph nodes.
Commit: The final agent action is appended to the Soul Journal, making state recovery deterministic.