The Internet of Cognition (IoC) project builds the agentic cognitive layer that lets autonomous agents effectively collaborate. Agent transports like A2A, MCP and SLIM already deliver messages reliably. They do not establish shared intent, shared context or shared reasoning, and without those, several capable agents assigned to one task will exchange messages correctly and still produce a poor outcome, often worse than what any of them would produce alone: contradictory findings committed to shared state, the same intermediate result interpreted differently, a session closed without a validated agreement. None of it raises an exception, so conventional monitoring never sees it.
IoC addresses this with four components. The Cognition Protocols operate above the transport under a stronger success condition than delivery — alignment, of intents, roles and opinions. The Cognition Fabric recomposes a multi-agent interaction into a single trajectory and knowledge graph, and persists the insights derived from it, with SCOPE supplying that sensing and comprehension layer and detecting the cognitive failures that would otherwise go unseen. The Cognition Engines act on those insights to make collective behavior more capable and trustworthy. MAS-Lab spans all of them, providing an environment in which any component is declared, executed, observed and measured against a baseline, so that the claims made about it are results rather than assertions.
This project is maintained by Outshift by Cisco.
Full documentation — motivation, architecture, component guides and tutorials — is published at ioc-project.github.io.
| Start here | |
|---|---|
| Why the project exists | Motivation |
| How the pieces fit together | Architecture |
| How to adopt it in an existing multi-agent system | Getting Started |
- cognition-protocols — ACCORD (shared intent by coverage) and CONCORD (deciding under the conflict that remains)
- scope — System for Cognitive Observability, Provenance and Explainability: instrumentation, trajectory recomposition, and the analysis that detects cognitive failures and explains them
- mas-lab — specification-driven multi-agent systems with OpenTelemetry observability, agent traces, benchmarking, MCP and A2A
- ioc-project.github.io — the documentation site
- Internet of Cognition — the vision, and the Scaling Out Superintelligence whitepaper
- Silent Failures, Loud Consequences — the empirical study behind the problem statement