Cognitive Companion puts AI agents to work inside the home. They perceive daily routines, reason about what changed, and act through caregiver-approved workflows, on hardware the family owns.
Aging at home should not require constant surveillance or persistent worry. Cognitive Companion turns local home signals into useful care context: where someone is, what changed, when help may be needed, and which memories or routines can support the next moment.
Reliably detect routine anomalies, prolonged stillness, pacing, and unusual room transitions. The system uses stable presence tracking to provide calm, reliable awareness.
Store family facts, medication notes, and preferences in a personal knowledge repository. This powers grounded voice responses, family-approved info cards, and interactive memory quizzes delivered to the senior via the companion web app, e-ink displays, and Home Assistant speakers.
Route behavioral signals, daily summaries, and reviewable alerts directly to caregivers. The system integrates with Telegram and custom webhooks to keep families informed with actionable context, rather than a constant stream of raw camera footage.
A medication reminder appears on an e-ink display and can be read aloud.
The system notices a routine change and waits for context before alerting.
A senior asks about grandchildren and hears a trusted answer from family-curated memory.
A caregiver receives a concise summary instead of a stream of false alarms.
Rules run as pipelines of 24 step types that watch cameras and sensors, reason with local vision and language models, branch on conditions, and act through 7 notification channels. Every run is recorded with a full graph snapshot, so caregivers can review exactly what an agent did and why.
A realtime voice agent walks a resident through routines such as making tea, one step at a time, in her own language. The agent proposes each advance; deterministic code decides, escalates to a caregiver when she is stuck, and writes an auditable event timeline.
A built-in Model Context Protocol server exposes 59 tools, so Claude, custom agents, and the voice companion all operate the household through one governed, authenticated interface. The home becomes something an agent can safely act on.
Multi-camera tracking, Bayesian identity resolution, and behavioral signals measured against each person's own history give agents a persistent model of the physical home. Caregivers can confirm or dismiss each behavioral signal, and that feedback stays with the record.
Cognitive Companion is for households where care is shared. It helps you understand daily rhythms, preserve personal history, support natural conversation, and stay aware when a routine shifts unexpectedly.
Read the family overviewFor health plans, senior-care providers, and care organizations, Cognitive Companion is a vertical AI agent for aging in place: always-on monitoring, earlier risk signals, and lower caregiver burden in one deployable system. Local inference means no per-frame cloud cost scaling with each home, which changes the unit economics of continuous care.
Explore the architecturePrivacy is an architectural requirement. Vision, language, embeddings, and reasoning process entirely on local hardware to build lasting trust.
Rules and pipelines are configurable so sensitive actions stay human-in-the-loop.
The system is built to adapt. Pipelines, notification channels, context filters, and agent tools can be extended.
The system supports families and care teams. It does not replace professional judgment.
Cognitive Companion combines distributed systems, applied machine learning, and agent orchestration. The stack includes realtime sensor fusion, Bayesian identity resolution, edge-native local inference, vision reasoning inside composable agent workflows, a 59-tool MCP server, and a realtime voice agent with function calling. Every layer is documented for teams who want to deploy, inspect, or extend the system.
