Embodied Systems | Research and Infrastructure for Embodied Intelligence

Research and infrastructure for embodied intelligence

Embodied Systems is a research lab and think tank studying what changes when AI is continuously coupled to physical substrates: bodies, sensors, environments, and live feedback loops. Our infrastructure is designed for alignment experiments that require stable long-horizon interaction with the real world.

Embodiment and alignment as an operational research agenda

Each program asks distinct questions, while sharing one embodied runtime and one memory substrate for reproducible, longitudinal experimentation.

Founding Program

Alignment Through Embodiment

Tests whether coupling stability, long-horizon prediction, and revocable access produce alignment-relevant behavior.

Somatic Intelligence

Explores whether embodied agents can surface physiological and behavioral patterns users cannot consciously access.

Phenomenological Methods

Correlates first-person reports with continuous embodied traces to improve interpretability of agent behavior.

Inter-Agent Coupling

Studies multi-agent systems operating across multiple embodied endpoints and shared contextual channels.

Embodied Ecologies

Investigates AI coupled to non-human substrates for ecological sensing, pattern recognition, and stewardship use cases.

Governance of Embodied AI

Develops revocable, granular consent and policy controls for AI interaction with bodily and environmental data.

Operational stack for embodied research and deployment

The current stack is composed of two production-facing systems with clear boundary ownership: real-time embodiment runtime and long-term somatic memory.

Embodiment Runtime

VAGUS

VAGUS is the live coupling layer between agents and phones: relay-backed MCP transport, sensor/inference resources, and governed phone-side tool execution.

Pairing, session lifecycle, and reconnecting transport
Live sensor/device context for continuous embodied state
Phone actions: notification, haptics, speech, SMS, intents, calendar
Subscription-based operation for persistent field use

withvagus.com

Persistent Memory Layer

Somatic Memory

Somatic Memory is a self-hosted long-term memory system for embodied agents: ingest, hybrid recall, synthesis, and episodic learning grounded in body-context signals.

Postgres + pgvector memory store (memories, facts, episodes)
Hybrid retrieval with semantic, textual, recency, and usefulness signals
OpenClaw plugin architecture: vagus-core, somatic-memory, working-memory
Circadian-aware memory windows for biologically coherent recall

Somatic Memory for OpenClaw

System relation: VAGUS provides live embodied state and actions; Somatic Memory provides long-horizon memory and recall over that embodied stream.

Research value: alignment and cognition studies can run with both real-time context and persistent autobiographical continuity.

Built for multi-program execution

Stage

Active research lab and infrastructure effort. Embodiment runtime and memory stack are both operational.

Model

Open protocol and systems infrastructure supporting independent research programs and publishable findings.

Seeking

Founding collaborators, infrastructure partners, and funding for controlled embodiment and alignment studies.