Agents still execute in isolation
Most agent systems are still blind mid-task. They use tools, but they do not share live cognition while work is still in motion.
Multi Agent Cognition · Protocols + systems
Infrastructure for agents that think together.
Multi Agent Cognition is building the missing stack for multi-agent execution: real-time coordination, benchmarks that measure judgment under uncertainty, and persistent systems with durable context.
ProtocolMACP
BenchmarkFreight Agency Bench
SystemPulse
FocusCoordination + evaluation + execution
Why it exists
The stack is still missing the layer between tools and coordinated cognition.
Benchmarks reward shallow behavior
If all context is handed to the model and nothing compounds over time, you are not measuring judgment under uncertainty.
Context still degrades too fast
Long-horizon systems need durable memory, structured reconstruction, and cleaner specialization than one overloaded context window can provide.
What the stack covers
Protocol, benchmark, and execution system designed as one surface.
The work is organized around one problem: agents need better coordination, better evaluation, and better long-horizon execution than current tool-only workflows provide.
Protocol layer
MACP gives active agents a shared cognitive bus for ownership, findings, priority, and in-loop coordination.
Benchmark layer
Freight Agency Bench measures planning, adaptation, and information foraging in a live environment where mistakes compound.
System layer
Pulse extends the stack into persistent coordination, heartbeat-driven execution, and reconstructed context that does not forget.
Deployment layer
The goal is a coherent stack: protocol, evaluation, and execution working together instead of as disconnected demos.
01
MACP turns isolated agent work into shared situational awareness.
02
Freight Agency Bench measures whether an agent can reason when the environment keeps changing.
03
Pulse extends that into persistent orchestration, memory reconstruction, and job routing.
How it fits together
Three tracks aimed at the same bottleneck.
- Use MACP when multiple agents need shared situational awareness during execution.
- Use Freight Agency Bench when you need a benchmark that rewards strategy, adaptation, and information seeking.
- Use Pulse-style patterns when you need persistent coordination and structured memory over long horizons.
- Use the contact channel when you want to collaborate, evaluate a deployment, or discuss applied work.
Where it applies
Any environment where coordination, memory, and live adaptation matter.
Real-time agent coordinationLong-horizon benchmarkingPersistent execution systemsAutonomous software deliveryDecision-making under uncertaintyMulti-agent infrastructure research
Current surface
The public protocol is live. The broader stack is taking shape.
What exists now
MACP Shared cognition protocol
Freight Agency Bench Long-horizon benchmark
Pulse Persistent agent harness
Primary domain multiagentcognition.dev
Protocol site macp.devContact channel
Sender Verified domain sender
Channel multiagentcognition.dev/api/contact
Provider Resend
Routing Private inboxContact
Talk to us about coordination, benchmarks, or applied agent systems.
Use the form for research inquiries, product conversations, or implementation work. Messages are reviewed directly by the team.
Multi Agent Cognition
Building the shared layer between isolated agents and useful systems.
The aim is not one more agent demo. It is a coherent operating layer for agents that need to coordinate, remember, and adapt in real environments.