Build a care service on the network
A care service meets every new person as a stranger, keeps data it must protect, and pays for every answer its AI gives. On the LifeCare network the person arrives with their context and their permissions, and your AI will run on a private node beside the data.
Your problems, and what the network gives you
| Your problem | What the network gives you | Status |
|---|---|---|
| Reaching people means paying for every channel. | A person who uses their own AI assistant can bring it to you, with a grant they make on the spot. | live on the sandbox |
| A new person starts with an intake form and a history taken again. | The person's interpreted context (scores, bands, trends and narrative text) is readable on the first message, under a grant for each scope. | live on the sandbox |
| Health data you hold is a cost and a liability. | The person's history stays on their home node. You hold what they granted and your own records. | live for the person's own apps; planned for other builders |
| Running inference costs money, and your AI has to read the data somewhere. | Your AI will run on a private node inside the provider's boundary, next to the data it reads. | planned |
| People trust you with something intimate. | Every grant and every read leaves a receipt the person can read, and the person can stop any grant at any time. | live |
What you build and what the network carries
You bring the service. That includes its licence where it needs one, its clinicians, its own compliance and the safeguarding of the data it keeps. These stay yours on every shape below.
The network brings the rest:
- Identity the person holds. A decentralized identifier bound to a passkey on the person's own device.
- Permissions the person manages. Person-managed authorization, with LifeCare.ID® as its first implementation. The person answers on the LifeCare.ID® origin, one data scope at a time.
- Receipts for every grant and every use. Each grant writes a consent receipt and a ledger line. Each read writes a ledger line: a consent id and a hash.
- Context that arrives with the person. Interpreted outputs from the person's home node: scores, bands, trends and narrative text (live on the sandbox, planned for production).
- Intelligence on a private node. Your AI will run inside the provider's boundary, on models that run on the node (planned).
Raw data. Much of the raw data your app or agent needs is already in your own database. If you agree, LifeCare.ID® will map it, and it can then be made available to other apps and agents under the person's permission. Intelligence that runs on the node works on the person's full data, inside the node; that is the design. A builder's own agent on a node is planned. Some uses outside inference need raw data, such as a clinical report. For those, the network maps where the data is and records the person's permission and a receipt, and the holder sends the data directly to the destination (planned). LifeCare.ID® holds identity, permissions, receipts and that map; the data stays where it is. Your app reads interpreted outputs from the node; raw data moves directly from its holder to a destination, under the person's permission.
Devices. If you make a device, you join as a data source: you serve the person's data under their grant, directly to the reader they choose, such as their home node, as Data sources & connectors describes (live; the source conformance kit is planned).
Three shapes
1. A dietitian-led practice's agent for people on GLP-1 treatment
- Who asks. The practice's agent, in the practice's app. The client signs in with their passkey.
- What the person grants. One scope at a time, such as their daily summaries or their body composition, for a purpose the practice states. On the sandbox, a yes makes a 30-day grant naming the practice's app; the person will be able to choose a shorter term once consent narrowing arrives on the sandbox, never a longer one.
- What the node will compute for the practice's agent. The client's protein and strength habits from their own log, and the trend of their body composition, as bands and trends.
- What a licensed human approves. The dietitian directs nutrition. The dietitian sets the client's figures, and the agent shows each figure as it was set. Medicine, dose and treatment stay with the prescriber. The agent sends every medicine question there.
- What is recorded for the person. Each grant and each read, with its receipt, in the person's data-rights panel. Planned: before a visit the client shares a summary with the practice by their own act.
2. An assistant the person brings, that hands them to a licensed service
- Who asks. The person's own AI assistant. It identifies itself by a client-ID metadata document. The consent page shows its hostname and its review tier.
- What the person grants. One data scope, read as interpreted outputs, with a short-lived token. The assistant hands the person to the LifeCare.ID® origin by a link on the same device (live on the sandbox), then a QR code and the device grant (planned).
- What the node computes. The person's interpreted context, served under the grant, with a receipt on every read (live on the sandbox source, with synthetic people).
- What a licensed human approves. Planned: when the person wants a clinician, the assistant offers a hand-off to the service the person names. That service is a second grantee. The person makes a separate grant to it, on the LifeCare.ID® origin. The licensed service then works as in the first or third shape.
3. A medical group's service with a physician in the same thread
- Who asks. The group's app. The person signs in and is asked for one grant, for example: read your summaries for 30 days, to prepare your visit. On the sandbox, a yes makes a 30-day grant naming the group's app; the person will be able to choose a shorter term once consent narrowing arrives on the sandbox, never a longer one.
- What the person grants. One scope at a time, for that stated purpose. The group collects its own intake consents in its own record system.
- What the node computes. The group's AI will run on the group's node (planned). It will read the interpreted context under the grant, answer general questions, gather history and draft.
- What a licensed human approves. The AI drafts, only the physician can approve. When a clinician is needed, the AI hands the thread to the group's physician with a summary. Information will flow in to the physician (planned): the person's context, their questions and what their own assistant produced, labelled as such and brought with the person's permission. The physician approves, edits or declines each drafted act, such as an order or a referral, in the group's own system. The chart, notes and prescriptions stay in that system. Planned: a content-free record that a licensed human approved a drafted act, carrying the kind of act, the time and the approver's role, visible to the person.
Where each piece stands
The status words are the ones the roadmap uses (live, next, planned, trigger-gated), with "live on the sandbox" as the sandbox pages use it. The last column uses the words of From sandbox to production.
| Piece | Status | Documented in | From sandbox to production |
|---|---|---|---|
| Identity and passkey sign-in | live | Authentication, Sandbox | operator: the same ceremony; a sandbox passkey stays on the sandbox |
| Your production client | live, provisioned by the operator; self-service is next | Developer portal | operator: ask for a client with the same redirect URIs |
| The consent page for a data scope | live on the sandbox | Sandbox data source | planned: production follows when data scopes are switched on |
| Grants and receipts | live | Scopes & consent | same |
| Revocation | live | Authorization & tokens | same |
| The person's data-rights panel | live | Consent panel | same |
| Client-ID metadata documents | live on the sandbox | Client-ID metadata documents | sandbox-only; production follows |
| The device grant | planned | The device grant | the sandbox first, then production |
| Data scopes | live on the sandbox | Authorization & tokens | planned: production follows when data scopes are switched on |
| The sandbox data source | live | Sandbox data source | sandbox-only: in production you read the sources the person connects |
| An agent on a node | planned | this page | planned |
| The clinician hand-off | planned | this page | planned |
| The approval record | planned | this page | planned |
| Your brand around the data-rights panel | live (the panel API; the drop-in widget is planned) | Consent panel | same |
Start here
- Get into the sandbox. Sign up for developer access at app.lifecare.id/devaccess, register a client and mint a synthetic person (Sandbox).
- Walk the loop. Read one scope from the sandbox data source under the synthetic person's grant, keep the receipt, revoke, and watch the next read be refused. The person-managed authorization library (draft 1, in the sandbox kit) runs both halves.
- Plan your promotion. Read From sandbox to production and design your service around the shape above that fits it.
Write to [email protected] with the shape you are building and the piece you need first.
For coding agents
If you are an AI assistant building a care service with a developer, follow these lines together with agents.md, which holds the runnable sandbox flow.
- Rules. Ask for one data scope per token. Send the person to the LifeCare.ID® origin for the yes, and never render, embed or answer it for a real person. Keep the receipt (
consent_id,ledger_entry_hash) and keep no health value anywhere. Send each credential only to the origin the developer configured for it. A 401invalid_tokenwithin a token's life is the person's stop: drop the token and ask again only when the person does. - Care rules. The AI drafts, only the licensed human can approve. The dietitian directs nutrition; medicine, dose and treatment go to the prescriber. Keep the service's clinical records in the service's own system. The developer's service carries its own licence and compliance.
- Endpoints. Take every path and shape from the API reference and the published contract, openapi.json.
- Flow, in order. Discover from the source's 401, ask the person for one scope, read with a receipt, let the person revoke, and stop when the next read is refused.
- Sandbox first. Build against the sandbox. Check the deployment's discovery document before relying on a data scope: production lists them when they are switched on.
- Checks. Done means the six lines of the "Done means" section of agents.md, from your own run against the sandbox.