OPEN AGENT STACK
Open Agent Stack is an independent, umbrella toolkit for building safe, structured ai agent ecosystems
Think of Open Agent Stack as "HTTP + middleware" for agents!
Open Agent Stack is a software solution provided by Prime Vector
Open Agent Spec (OAS)
A YAML-based specification to declaratively define AI agents, including prompt structure, memory format, and task config. Used as the foundation for scaffolding, deployment, and orchestration.
Behavioral Contracts (BCE)
Runtime enforcement for agent behavior. Define what agents must, should, and must not do, and enforce these at inference time. Adds guardrails, governance, and observability.
DACP
Declarative Agent Communication Protocol, a message-passing and collaboration layer built around structured task exchange, enabling multi-agent systems to interoperate predictably.
Cortex (Intelligence Engine)
Cortex acts as the full reasoning brain in a layered agent architecture, where sensors provide perception (eyes and ears), a reactive base layer handles reflexes, and Cortex performs structured, contract-bound cognition using LLMs.
Agent Tool Ecosystem
Configure your agents to use tools at runtime - or use standalone!
File Writer
Write to and modify files on disk
Terminal
Command-line interface for agents to run scripts and code
API Generator
Automatically generate REST APIs from external Web Service Specifications
Monitor
Real-time monitoring and observability for agent systems
See OAS in Action
open_agent_spec: 1.0.4
agent:
name: hello-world-agent
description: A simple agent that responds with a greeting
role: chat
intelligence:
type: llm
engine: openai
model: gpt-4
endpoint: https://api.openai.com/v1
config:
temperature: 0.7
max_tokens: 150
tasks:
greet:
description: Say hello to a person by name
timeout: 30
input:
type: object
properties:
name:
type: string
description: The name of the person to greet
minLength: 1
maxLength: 100
required: [name]
output:
type: object
properties:
response:
type: string
description: The greeting response
minLength: 1
required: [response]
prompts:
system: >
You are a friendly agent that greets people by name.
Respond with: "Hello <name>!"
user: "{{name}}"
behavioural_contract:
version: "0.1.2"
description: "Simple contract requiring a greeting response"
role: "Friendly agent"
behavioural_flags:
conservatism: "moderate"
verbosity: "compact"
response_contract:
output_format:
required_fields: [response]Mission & Vision
The Problem
Today's AI agents are fragmented, unpredictable, and lack standardised governance. Developers face inconsistent APIs, no behavioural guarantees, and limited observability. Multi-agent systems are brittle, with poor communication protocols and no unified supervision. This creates security risks, deployment complexity, and prevents scalable AI applications.
Our Solution
The Open Agent Stack provides a comprehensive, interoperable framework for building, deploying, and governing AI agents. We combine declarative specifications, runtime behavioral enforcement, structured communication protocols, and intelligent supervision to create reliable, scalable multi-agent systems.
Beyond MCP
While Model Context Protocol (MCP) focuses on tool integration and context management, the Open Agent Stack addresses the broader ecosystem needs: behavioral governance, structured and declarative agent-to-agent communication, lifecycle management, and system-wide supervision. We complement MCP by providing the missing layers for production-ready, multi-agent deployments.
Get Involved
We're building this in the open and welcome contributions from the community. Whether you're interested in specifications, behavioral contracts, communication protocols, or supervision systems, there are opportunities to contribute.