Build modular health agents on one shared context.
Health Core is longitudinal infrastructure, not a feature bag — an open-core health agent framework where agents read and write a shared, permissioned, provenance-first profile, instead of becoming isolated bots with separate truths.
One profile beats a dozen disconnected bots.
Specialist agents over a shared Longitudinal Health Profile avoid separate truths — every interpretation lives in one coherent, inspectable model.
The broken pattern
Each app stores its own slice. Each chatbot forgets yesterday. The user becomes the integration layer.
The Health Core pattern
Specialist agents share one LHP. Writes carry provenance and confidence. Permissions scope what each agent sees and does. Coherence is structural, not hoped for.
Stable building blocks, not ad-hoc features.
Health Core is infrastructure. These sixteen primitives are the vocabulary every agent, surface, and integration maps to.
- user
- event
- observation
- source
- timestamp
- provenance
- confidence
- health state
- goal
- preference
- intervention
- experiment
- agent
- permission
- follow-up
- export
Every capability maps back to these primitives or deliberately extends them. See the LHP model →
Specialists that share context, not compete for it.
The user sees a unified companion. Underneath, modular agents operate over one permissioned profile — each with declared scope and accountability.
Unified companion, many specialists
Users experience one coherent companion — but underneath, specialist agents handle sleep, recovery, nutrition, and more without fragmenting the model.
Each agent has scope, permissions, provenance, accountability
Every agent declares what it can read, what it can write, and what it inferred. Outputs carry source, confidence, and audit trail — never anonymous black-box assertions.
Agents reason over the shared LHP
Specialists don't maintain private silos. They read and write the same Longitudinal Health Profile, so interpretations stay coherent across domains and over time.
New agents plug in safely
Add a specialist without rewriting the core. Permissions, provenance, and the autonomy ladder gate what each agent can do — so expansion doesn't erode safety or user agency.
Example specialist agents — all reading and writing the same LHP:
- Sleep
- Recovery
- Exercise
- Nutrition
- Mental state
- Medication
- Symptom
- Experiment
- Clinical prep
- Preventive
- Care coordination
What an agent declares — and what it writes back.
Before an agent touches the profile, it declares its scope and the autonomy level it operates at. Every write it makes carries the events it was derived from, its confidence, and a consent state.
{
"agent": "nutrition.v1",
"autonomy": 3, // suggest — never acts alone
"reads": ["observation", "goal", "intervention"],
"writes": ["suggestion", "follow-up"],
"requires_consent": "per-write",
"provenance": "required", // no anonymous writes
"confidence": "required"
}POST /v1/suggestions
Authorization: Bearer ⟨agent-token⟩
{
"agent": "nutrition.v1",
"derived_from": ["evt_8f3a", "evt_b21c"], // low ferritin + fatigue note
"text": "Recheck ferritin in 8 weeks; …",
"confidence": 0.62
}
// → 201 Created
{
"id": "sug_4d9e",
"status": "pending_consent", // not yet visible to clinicians
"provenance": "agent:nutrition.v1",
"created_at": "2026-04-16T08:02:11+10:00"
}Shapes shown for illustration — the public API is not yet released. See the API & SDK roadmap →
Autonomy earned through checkpoints.
Agents don't jump to action. The ladder defines what each capability level requires — and higher rungs demand more safety, consent, auditability, and confidence to justify them.
Explain
Summarise and clarify what the profile already contains — grounded in the user's own data.Interpret
Surface patterns, inconsistencies, gaps, and possible relationships across the longitudinal record.Suggest
Propose next steps, questions, experiments, or behavioural changes for the user to consider.Plan
Build structured, cross-domain plans — sleep, training, nutrition, recovery, and care preparation.Follow up
Remember unresolved issues, check whether plans were followed, and adapt as new data arrives.Coordinate
Hand context between agents and domains while preserving coherence in the shared profile.Act with consent
Trigger reminders, update records, or prepare exports — only within explicitly approved permissions.Support care continuity
Bridge daily life, personal optimisation, and provider-facing workflows without collapsing them into one undifferentiated stream.
Infrastructure principles, not marketing features.
These constraints shape every schema decision, agent contract, and surface in Health Core.
Longitudinal by default
Every primitive is designed for trajectories, not snapshots. Events accumulate meaning as the profile evolves.
Provenance-first
Observations carry where they came from. Agents and interfaces must show lineage — not just conclusions.
Timeline-aware
Context depends on when things happened and what came before. Ordering and duration matter.
Uncertainty-aware
Confidence is a first-class field. The system never presents inference as fact when it isn't justified.
User-correctable
People can inspect, approve, correct, and restrict what the model holds. The profile belongs to the user.
Privacy-preserving
Scoped permissions and consent checkpoints gate access. Higher autonomy requires stronger justification.
Open-core compatible
Stable primitives and inspectable schema behaviour — designed so the core can be extended, audited, and shared.
The rules anything built on the LHP must follow.
These are non-negotiable. They keep the profile coherent, honest, and owned by the user as the platform grows — and they apply to every agent and integration, including ours.
Never build isolated features
No private silos, no separate truths. Every capability reads from and writes to the one shared Longitudinal Health Profile.
Always strengthen the LHP
Each write adds longitudinal value — source, time, and links — rather than bolting on a stateless endpoint that forgets.
Provenance or it doesn't ship
Every observation and inference carries where it came from. Anonymous, unsourced assertions don't belong in the profile.
Keep uncertainty visible
Confidence is required, not optional. Never render inference as fact, and never bury ambiguity to look more authoritative.
The user can always correct and revoke
Inspect, approve, restrict, export, delete. Agents act only inside the scope a user explicitly granted — and within nothing else.
Earn autonomy through checkpoints
Default to explain and interpret. Higher rungs of the ladder require explicit consent and stronger justification before acting.
These flow from the principles behind Health Core. Read the manifesto → or see how we handle trust and control →
From private beta to a public agent SDK.
We open the platform in stages, locking each layer before the next depends on it. Provenance, permissions, and the autonomy ladder hold at every step.
Private beta
Schema, ingestion model, and agent interfaces shared with design partners. Request access to explore them and shape the primitives.
Schema access
Read the primitive vocabulary and event shapes, and build against a stable contract before write access opens.
Public read API
Query a permissioned profile timeline — events with provenance and confidence — from your own application.
Write & agent SDK
Register agents with declared scope, then write grounded suggestions and follow-ups within the autonomy ladder.
Build it in the open, with other engineers.
The core primitives and schema are designed to be inspected, extended, and shared. Follow development, open issues, and help shape the contract before it's set.
Open-core & licensing
The core primitives and schema are intended for release under the Apache 2.0 licence — extend and audit them, then build products on top.
Developer updates
Join the private beta to get schema and SDK updates as we open access. Choose “Developer” on the form so we route you the right way.
Request developer accessAPI, SDK, licensing, and the model.
Is there a public API or SDK yet?
Not yet. Health Core is in private beta, so the health AI platform API and longitudinal health SDK are shared with design partners first. Request developer access to explore the schema, ingestion model, and agent interfaces. The architecture is open-core compatible — public read APIs and a write/agent SDK follow once the primitives and permission model are stable enough to commit to.
What licence is the open core under?
The open core — the primitive vocabulary and schema behaviour — is intended for release under the Apache 2.0 licence, so it can be extended, audited, and built on. Higher-level products and hosted services build on top of that core.
How can I contribute or follow development?
Follow and open issues on the open-core repository on GitHub, and request developer access to join the private beta and developer updates. We're especially interested in feedback on the primitives, the agent contract, and the autonomy ladder.
How do agents avoid contradicting each other?
All agents read and write the same Longitudinal Health Profile rather than maintaining isolated state. Provenance and confidence fields make conflicting interpretations visible and resolvable — the system surfaces disagreement instead of hiding it behind separate chat threads.
How is user consent enforced?
Every agent operates within scoped permissions. The autonomy ladder requires explicit checkpoints before higher-capability actions — explain and interpret by default; act with consent only within approved boundaries. All meaningful writes are auditable.
Is Health Core a diagnostic or treatment platform?
No. Health Core is longitudinal health infrastructure — not an autonomous medical system. Agents organise, explain, and coordinate around a user-owned profile. Medical decisions belong with qualified clinicians.
Request developer access.
We're onboarding engineers and agent builders who want health AI grounded in shared context, provenance, and consent — not another isolated chatbot.