FinCore — financial platform
Separation of duties expressed as an exclusion, and approval limits as caveats that carry the payment amount as request context. Eight scenarios.
See what modern authorization looks like with SpiceDB.
Explore interactive demos, understand Google Zanzibar-inspired ReBAC, and see how to centralize permissions across applications and AI agents.
definition user {} definition team { relation member: user } definition folder { relation viewer: user | team#member permission view = viewer } definition document { relation parent: folder relation viewer: user | team#member permission view = viewer + parent->view }# written by your app as things happen team:eng#member@user:alice folder:specs#viewer@team:eng#member document:roadmap#parent@folder:specs
// start here
The idea, the failure that makes it necessary, and the whole model working in a product. Everything else on this site builds on these.
// why engineers pick spicedb
One schema, one API and one consistency model, instead of permission checks scattered through every service.
definition user {}
definition team {
relation member: user
}
definition folder {
relation viewer: user | team#member
permission view = viewer
}
definition document {
relation parent: folder
relation viewer: user | team#member
permission view = viewer + parent->view
}# write, and keep the token it returns
$ zed relationship delete document:roadmap viewer user:bob
GhUKEzE3NjM4MjAwMDAwMDAwMDAwMDA=
# this check must see that write
$ zed permission check document:roadmap view user:bob \
--consistency-at-least GhUKEzE3NjM4MjAwMDAwMDAwMDAwMDA=
false$ zed permission lookup-resources document view user:alice
roadmap
launch-plan
$ zed permission lookup-subjects document:roadmap view user
alice
mayause expiration
caveat within_limit(amount int, limit int) {
amount <= limit
}
definition payment {
relation approver: user with within_limit
relation auditor: user with expiration
permission approve = approver
permission view = approver + auditor
}// Go: follow every change to documents
stream, _ := client.Watch(ctx, &v1.WatchRequest{
OptionalObjectTypes: []string{"document"},
})
for {
resp, err := stream.Recv()
if err != nil { break }
apply(resp.Updates) // update your index
}# run SpiceDB on your own PostgreSQL
$ spicedb serve \
--grpc-preshared-key "somerandomkeyhere" \
--datastore-engine postgres \
--datastore-conn-uri "postgres://user:pass@db:5432/spicedb"
# CockroachDB, Spanner and MySQL work the same way
$ spicedb serve --datastore-engine cockroachdb ...// demos
Interactive demos you can open in a browser. Most come narrated, at normal or double speed.
The idea, and where it came from.
The difference is one more column in what gets stored. Three escalating requirements show where it starts to matter.
Eight interactive sections on Google’s 2019 paper — relation tuples, userset rewrites, check evaluation, Expand, zookies, Leopard, hot spots and the reported scale.
Every concept in the paper, and what SpiceDB calls it — relationships, schema, ZedTokens, Materialize — plus what SpiceDB adds, and what it does not claim.
The failures that are difficult to see until they bite.
Access is removed, the document changes, and a replica that has seen neither answers the check.
Starts with the picture, then proves it: one policy implemented three times, in three languages, giving three different answers — until you switch to one service.
One handler, two systems, no transaction across both. Crash it at each point and see which one leaves them disagreeing.
The mechanics, shown rather than described.
Three teams, three files, one deployed schema. Watch a team adopt a shared shape by adding a single line to their own file.
A cached allow that outlives the permission behind it — and the stream that closes the gap.
Six steps through a single save: the CRM stores its own record, then sends SpiceDB the one relationship that changed.
What it looks like inside a product, and what it costs.
Isolation is the easy half. The person in two tenants, the consultant across all of them, the auditor in none.
Five stages from ten users to AI agents, walked twice — once building it yourself, once with SpiceDB.
Pick what you'd actually build, set your team and rates, and see the three-year cost of each path.
Grant exactly what the ticket asked for, and nothing more.
// industries
Every row, balance and document is a separate permission check. Open the app, or let the narrated version walk you through it.
Separation of duties expressed as an exclusion, and approval limits as caveats that carry the payment amount as request context. Eight scenarios.
Matters, ethical walls, privilege and outside counsel. The wall is an exclusion, so it denies someone who genuinely qualifies. Six scenarios.
A multi-tenant HR product where every row, field and panel is its own permission check — including the manager who opens a profile and is denied the salary. Five scenarios.
Workspaces, teams and projects across several enterprise customers. An outside consultant gets one project and nothing else; a customer admin gets their whole account and no one else’s. Five scenarios.
The same question, asked by four people, answered from different sources. Retrieval is filtered by permission before anything reaches the model. Five scenarios.
Two doctors at the same hospital with different access to the same patient, because only one is on the care team. Administer and read are separate. Six scenarios.
Datasets, reports and groups. A partner gets one dataset and its approved report without the warehouse behind it, and the arrow only points one way. Five scenarios.
Deploy is an intersection: on the team and on the approval list. Staging on one service never implies production on another. Break-glass reads and cannot deploy. Six scenarios.
// teams
What each team cares about. The CISO page is a working demo; the rest are overviews.
Blast radius per account, computed. Least privilege, AI agents, revocation and the access review — plus five things it does not do.
Stop maintaining permission logic scattered across services. One schema, one API call, one engine.
Ship the sharing, roles and collaboration features customers keep asking for, without a rebuild each time.
Access sprawl, instant revocation, and a straight answer to who can reach what.
// articles and whitepapers
// spicedb enterprise
Audit logging, the security embargo, restricted API access, rate limiting and a support relationship, around the same open-source engine.
A cover, one slide for each of the five with what open source lacks and what it means in practice, and a closing summary. Opens as a PDF in the browser.
One switch flips the whole page between open source and Enterprise. Run the same traffic, make the same API calls, and watch what differs. Five tabs, one per capability.
Engineer-months, maintenance, horizon and the Enterprise price on a slider from $80K to $250K. Two of the five cannot be built at all, and the model says so rather than inventing a figure.
// in your ecosystem
Interactive demos and guides showing SpiceDB working with Databricks and with the identity providers your teams already use.
SpiceDB alongside Unity Catalog, Lakebase and AI on the Databricks platform.
SpiceDB decides who can see which accounts across desks, books and legal entities, and keeps Unity Catalog row filters in sync.
A facilities app on Lakebase where SpiceDB decides who can see and approve each work order, including contractor access that ends on its own.
The same AI assistant and question, side by side: which documents leak with index-level grants, and how SpiceDB filters retrieval per document.
SpiceDB next to the identity platform you already run.
How relationship-based access control with SpiceDB works alongside Ping Identity, how it differs from Ping Dynamic Authorization, and five patterns for running them together.
Using Entra object IDs, group sync, Conditional Access and ID Governance with SpiceDB, and five patterns for running them together.
// your product, modeled
Send us your company name. We’ll build a custom permission schema and a live, clickable demo of how authorization would work in your app.