Flow Definition Schema

This reference describes the JSON shape accepted by O2ID flow definitions. For a usage-focused introduction, see Flows.

Flow resource

A flow resource wraps a definition with tenant metadata:

{
  "name": "Login",
  "trigger": "login",
  "enabled": true,
  "allowedClientIds": [],
  "definition": {
    "start": "identify",
    "nodes": {}
  }
}

Field	Required	Description
name	Yes	Human-readable flow name, unique within the tenant
trigger	Yes	Flow trigger name. Use reserved system trigger values such as login for O2ID-invoked journeys; use names such as step-up or age-verification for custom journeys.
definition	Yes	Flow graph definition
allowedClientIds	No	Client IDs allowed to start this flow. Empty or omitted means no client allowlist
enabled	No	Defaults to true

Definition object

The definition object has two required fields, plus one optional field that controls whether the flow may be started with no authenticated caller:

{
  "start": "identify",
  "nodes": {
    "identify": {
      "type": "authenticator",
      "id": "identifier",
      "onSuccess": "password",
      "onFailure": "identify"
    }
  }
}

Field	Required	Description
start	Yes	ID of the first node to execute
nodes	Yes	Object whose keys are node IDs and whose values are node definitions
allowUnauthenticated	No	Whether this flow may be started with no known subject. Omit to use the default: true only for the login trigger, false for every other trigger. Set explicitly to opt any trigger in or out

allowUnauthenticated is what lets /oauth2/authorize's ?prompt= param (see Flows) launch a trigger other than login for an unauthenticated user — the trigger name itself carries no special meaning; only this flag, and whether the caller already has a subject, do.

Node IDs are local to the flow definition. They are referenced by fields such as onSuccess, onFailure, properties.onLockout, condition branch then, and condition else.

Node shape

Every node has the same top-level shape: type, an optional id naming the resource this node configures (an authenticator ID for authenticator, a connection ID for connection — left unset for node types with no such resource), the control-flow fields that type uses (onSuccess/onFailure for linear nodes, branches/else for condition), properties for anything else node-type-specific, and an optional outputs map — every node type accepts outputs, regardless of what's in properties. Keeping configuration under properties (or, for the one resource reference a node type has, under the shared id field) means new node types add fields there instead of growing the shared node shape. See Node outputs for what outputs does.

{
  "type": "authenticator",
  "id": "<totp-authenticator-id>",
  "onSuccess": "allow",
  "onFailure": "totp",
  "properties": {
    "maxAttempts": 5,
    "onLockout": "deny"
  }
}

id also doubles as the selection key when this node appears inline inside a choice node's options array (see choice) — there is no separate "option key" concept, an option is just a node.

Node references

Most transitions are plain string references:

{
  "onSuccess": "allow",
  "onFailure": "password"
}

Condition branch then and condition else also support inline condition objects:

{
  "type": "condition",
  "branches": [
    {
      "if": "'totp' in user.enrolledFactors",
      "then": "totp"
    }
  ],
  "else": {
    "type": "condition",
    "branches": [
      {
        "if": "tenant.mfaRequired",
        "then": "totp"
      }
    ],
    "else": "allow"
  }
}

authenticator.onSuccess, authenticator.onFailure, and authenticator.properties.onLockout are string node IDs.

Supported node types

Type	Purpose
authenticator	Resolves the user by their login identifier, or verifies built-in password or a configured TOTP authenticator
condition	Evaluates CEL expressions and chooses the next node
outcome	Completes the flow run
flow	Jumps into another flow, resuming this one when it completes
script	Runs tenant-authored Starlark logic in-process, no network/filesystem access
webhook	Calls a tenant-supplied HTTP endpoint; its response becomes the step result
connection_call	Calls an operation on a registered connection (e.g. a remote connection type); its response becomes the step result
credential	Grants a credential (currently: an OID4VCI credential offer) to the run's subject
choice	Presents several sign-in options at once; the client picks which one runs
organization_context	Detects a nested Organization tenant and shows a "signing in to" banner naming it

Use metadata to discover the node types accepted by the running O2ID instance.

authenticator

An authenticator node resolves the user's identifier, or verifies a credential or factor.

Built-in identifier step — resolves a user without verifying a credential. Useful when a later node needs to act on the resolved user (branch on user.enrolledFactors, tenant policy, etc.) before deciding what to verify next:

{
  "type": "authenticator",
  "id": "identifier",
  "onSuccess": "password",
  "onFailure": "identify"
}

Built-in password step:

{
  "type": "authenticator",
  "id": "password",
  "onSuccess": "allow",
  "onFailure": "password",
  "properties": {
    "maxAttempts": 5,
    "onLockout": "deny"
  }
}

TOTP step, using the ready-to-use built-in instance (see below — no setup required):

{
  "type": "authenticator",
  "id": "totp",
  "onSuccess": "allow",
  "onFailure": "totp",
  "properties": {
    "maxAttempts": 5,
    "onLockout": "deny"
  }
}

Field	Required	Description
type	Yes	Must be authenticator
id	Yes	identifier to resolve the user without a credential, password for the built-in password step, or a configured authenticator's ID
onSuccess	Yes	Node ID to execute after successful verification
onFailure	No	Node ID to execute after failed verification. Often points back to the same node
properties.maxAttempts	No	Number of failed attempts before properties.onLockout is used
properties.onLockout	No	Node ID to execute after maxAttempts failures

id is one of:

• identifier — the built-in identifier step
• password — the built-in password step
• totp — a ready-to-use, zero-config TOTP factor every tenant gets for free (default issuer "O2ID", 6 digits, 30-second period). It's virtual — never created via POST /authenticators, and can't be updated or deleted — so create a separate totp-type instance instead if you need different settings (see Managing Authenticators)
• the id of a tenant-configured Authenticator instance — never an authenticator type name directly

To discover valid values:

• Configured instances (and their IDs): GET /authenticators (o2idctl authenticators list) — always includes the built-in totp first, even before anything is explicitly created
• Types available to configure: GET /authenticators/meta (o2idctl authenticators list types) — today totp (a synchronous code-entry factor, see the "Configured TOTP step" example above) and verifiable_credential_presentation (an out-of-band factor — see Out-of-band verification below)

The built-in password step doubles as an identify step when it's reached with no user already resolved (i.e. no identifier authenticator ran before it in this run) — it accepts identifier (or username) alongside password in the same submission and resolves the user itself, instead of requiring a separate identifier step first. This is what O2ID's implicit default login flow uses, so a plain login renders as one combined form rather than two screens. nodeView.fields reflects which shape is expected: ["identifier", "password"] when combined, or just ["password"] when an earlier identifier authenticator already resolved the user. Use a separate identifier authenticator instead when you need to act on the resolved user (e.g. branch on user.enrolledFactors) before verifying the password.

Expected identifier submission — POST /flow-runs/{id}/submit:

{ "values": { "identifier": "alice.chen" } }

Submit the value of whatever field the user's type designates as its login identifier — username for the built-in user type, or whatever a custom type's usernameField names.

Expected password submission:

{ "values": { "password": "secret" } }

Or, standing alone with no prior identifier authenticator:

{ "values": { "identifier": "alice.chen", "password": "secret" } }

Expected TOTP submission:

{ "values": { "code": "123456" } }

If the referenced TOTP authenticator has progressiveEnrollment enabled (see TOTP authenticator) and the subject has no verified enrollment yet, the node enrolls them in place instead of failing the code. First submit an empty values object to get setup material:

{ "values": {} }

nodeView.data then contains otpauthUrl and secret, and nodeView.fields becomes ["enrollmentId", "code"]. After the user scans the setup URI, submit both:

{ "values": { "enrollmentId": "<enrollment-id>", "code": "123456" } }

Out-of-band verification (e.g. credential presentation)

Some authenticator types verify asynchronously instead of via a client-submitted code — today, one such type: verifiable_credential_presentation (see Verifying Credentials), which asks the subject to present a Verifiable Credential from their wallet (OID4VP). Referencing one of these from an authenticator node works exactly like referencing TOTP — same id/maxAttempts/onLockout fields — but the node pauses instead of rendering fields: nodeView.data.requestUri carries an openid4vp:// deep link to render as a QR code, and nodeView.fields is absent (there's nothing to submit). The client should never call POST /flow-runs/{id}/submit on this node — only the authenticator's own out-of-band callback (POST /oid4vp/response) resumes it, so the portal polls GET /flow-runs/{id} instead until the view changes.

Context written:

Path	Description
step.<nodeId>.result	success or failure
step.<nodeId>.attempts	Failed attempt count
step.<nodeId>.claims	An out-of-band authenticator's disclosed claims on success (e.g. a presented credential's claims)
user.id	Resolved user ID — set by the identifier and password authenticators
user.userType	The user's assigned user type key, e.g. employee — always set once resolved
user.username	The value of whatever field the user's type designates as its login identifier
user.enrolledFactors	Verified factor types for the user, such as totp
user.attributes	User attributes available to the run — email, when present, is always at user.attributes.email

condition

A condition node evaluates Google Common Expression Language (CEL) expressions in order. The first branch whose if expression evaluates to true wins. If no branch matches, else is used.

{
  "type": "condition",
  "branches": [
    {
      "if": "'totp' in user.enrolledFactors",
      "then": "totp"
    },
    {
      "if": "tenant.mfaRequired",
      "then": "totp"
    }
  ],
  "else": "allow"
}

Field	Required	Description
type	Yes	Must be condition
branches	Yes	Ordered list of branches
branches[].if	Yes	CEL expression
branches[].then	Yes	Node reference, or inline condition object
else	Yes	Node reference, or inline condition object

Clients do not receive a nodeView for a condition node. O2ID evaluates the condition and advances the run to the selected next node.

Condition context

Conditions can read these context variables:

Path	Type	Description
user.id	string	Resolved user ID
user.userType	string	The user's assigned user type key, e.g. employee — always set
user.username	string	The value of whatever field the user's type designates via usernameField
user.enrolledFactors	list<string>	Verified factor types for the user, such as totp
user.attributes	map	User attributes available to the run — email, when present, is always at user.attributes.email
tenant.mfaRequired	bool	Tenant MFA policy flag. Defaults to false when no policy value is available
step.<nodeId>.result	string	success or failure for a completed step
step.<nodeId>.attempts	int	Failed attempt count recorded by that step
step.<nodeId>.outputs.<name>	any	A named value declared by that node's outputs — see Node outputs
risk.score	int	Risk score. Defaults to 0 when no risk signal is available

Example expressions:

'totp' in user.enrolledFactors

step.password.result == 'success' && step.password.attempts == 0

tenant.mfaRequired || risk.score > 50

outcome

An outcome node completes the run.

Allow or deny:

{
  "type": "outcome",
  "properties": {
    "result": "allow"
  }
}

Claims-producing result:

{
  "type": "outcome",
  "properties": {
    "result": "success",
    "claims": {
      "ageOver18": true
    }
  }
}

Field	Required	Description
type	Yes	Must be outcome
properties.result	Yes	Result string, such as allow, deny, or success
properties.claims	No	JSON object returned with the outcome

deny marks the flow run as failed. Other result values complete the run.

flow

A flow node jumps into another flow, identified by its trigger, and resumes this flow's onSuccess/onFailure when that flow's run reaches an outcome node — the same way onSuccess/onFailure resume after any other node type, just with an entire other flow run in between.

The target flow is resolved exactly like POST /flow-runs resolves its own trigger: the active, enabled flow for that trigger, falling back to the built-in login flow for login when no custom one is active, or a not-found error otherwise. It starts at that flow's own start node unless properties.nodeId names a different node in it.

{
  "type": "flow",
  "onSuccess": "allow",
  "onFailure": "deny",
  "properties": {
    "trigger": "step-up",
    "nodeId": "totp"
  }
}

Field	Required	Description
type	Yes	Must be flow
onSuccess	Yes	Node ID to execute when the target flow's run completes with a non-deny result
onFailure	No	Node ID to execute when it completes with deny. If omitted, a deny ends this run entirely, exactly as if this had been a top-level outcome node
properties.trigger	Yes	Trigger of the flow to jump into
properties.nodeId	No	Node ID within the target flow's definition to start at, instead of its start

Clients do not receive a nodeView for a flow node itself — the run keeps rendering whatever node the target flow lands on next, same as condition.

Context written:

Path	Description
step.<nodeId>.result	success or failure, from the target flow's outcome result
step.<nodeId>.claims	The target flow's outcome claims, if it set any

Nesting is capped at 10 levels deep — a flow node whose trigger (directly or transitively) leads back to itself fails with a recursion-limit error once the cap is hit, since two flows referencing each other's triggers can't be caught at authoring time (each would need the other to already be active to validate).

script

A script node runs tenant-authored Starlark source in-process. Starlark has no network or filesystem access by default, so a script node can't reach anything outside the flow run's own context — unlike condition's CEL expressions, Starlark is Turing-complete, so properties.maxSteps and properties.timeoutMs bound how much it can run.

{
  "type": "script",
  "onSuccess": "allow",
  "onFailure": "deny",
  "properties": {
    "source": "result = {'riskAdjusted': risk['score'] > 50}",
    "maxSteps": 100000,
    "timeoutMs": 2000
  }
}

Field	Required	Description
type	Yes	Must be script
onSuccess	Yes	Node ID to execute when the script sets a result
onFailure	Yes	Node ID to execute when the script raises an exception, fails to set result, or exceeds its step/time limit
properties.source	Yes	Starlark source. Must assign a top-level result global — that value becomes step.<nodeId>.result
properties.maxSteps	No	Abstract computation step ceiling. Defaults to a server-side value; always clamped to a hard maximum regardless of what's configured
properties.timeoutMs	No	Wall-clock ceiling in milliseconds. Same defaulting/clamping as maxSteps

Unlike Starlark's default BUILD-file dialect, if/for/while are allowed at the top level and reassigning a top-level name (total = 0 then later total = total + 1) is allowed too — a script node's source is a short imperative program, not a config file being loaded as a module, so it reads like an ordinary function body without needing to be wrapped in a def.

The script reads the same user/tenant/step/risk values condition nodes evaluate against (see Condition context) as predeclared globals, and nothing else — no per-node-type secret smuggling.

Clients do not receive a nodeView for a script node — like condition, it runs synchronously and the run keeps rendering whatever node comes next.

Context written:

Path	Description
step.<nodeId>.result	The script's result value, whatever shape it is
step.<nodeId>.error	Failure reason, set instead of result when the node routed to onFailure

webhook

A webhook node calls a tenant-supplied HTTP endpoint and uses its response as the step result.

{
  "type": "webhook",
  "onSuccess": "allow",
  "onFailure": "deny",
  "properties": {
    "url": "https://verify.example.com/check",
    "method": "POST",
    "timeoutMs": 5000,
    "headers": { "X-Api-Key": "..." }
  }
}

Field	Required	Description
type	Yes	Must be webhook
onSuccess	Yes	Node ID to execute after a 2xx response with a valid JSON body
onFailure	Yes	Node ID to execute on any other outcome — network error, non-2xx status, an oversized or malformed body
properties.url	Yes	Absolute http:// or https:// URL
properties.method	No	Defaults to POST
properties.timeoutMs	No	Request timeout in milliseconds. Defaults to a server-side value; always clamped to a hard maximum
properties.headers	No	Extra request headers

The request body is the same user/tenant/step/risk context CEL conditions and script nodes read, JSON-encoded. Every request carries an X-O2ID-Signature header — an HMAC-SHA256 of the request body under a key derived from the tenant, so the receiver can verify the call actually came from this tenant's flow run. The response body is size-capped (64KB) and must be JSON.

SSRF guard: by default, a webhook node cannot reach a private/loopback/link-local/unspecified/multicast address, no matter what properties.url says — this blocks a flow author from using a webhook to probe or call internal infrastructure. To call an internal endpoint deliberately, add its hostname to the tenant's webhook allowlist:

o2idctl tenant-settings update --webhook-allowlist internal-service.example.internal

--webhook-allowlist is repeatable — pass it once per host to allow more than one. --clear-webhook-allowlist removes every host. Equivalently, PATCH /settings accepts the same field directly:

PATCH /settings
{ "webhookAllowlist": ["internal-service.example.internal"] }

An allowlisted hostname is exempted from the address check entirely — only add hosts the tenant genuinely intends flows to reach.

Clients do not receive a nodeView for a webhook node — like condition, it runs synchronously and the run keeps rendering whatever node comes next.

Context written:

Path	Description
step.<nodeId>.result	The parsed JSON response body
step.<nodeId>.error	Failure reason, set instead of result when the node routed to onFailure

connection_call

A connection_call node calls an operation on a connection and uses its response as the step result — the generic counterpart to connection's redirect-based federated login. Unlike connection, it works with any connection whose type declares the operation it names, not only an identity-provider one: send an SMS, verify a document, look up a value in an external system, and so on. See Connection Types for how a connection gets an operation to call in the first place.

{
  "type": "connection_call",
  "id": "acme-idv-connection",
  "onSuccess": "allow",
  "onFailure": "deny",
  "properties": {
    "operation": "verify",
    "input": { "documentId": "step.upload.outputs.documentId" },
    "timeoutMs": 8000
  }
}

Field	Required	Description
type	Yes	Must be connection_call
id	Yes	ID of a connection whose type declares properties.operation among its operations
onSuccess	Yes	Node ID to execute after a successful call
onFailure	Yes	Node ID to execute on any other outcome — network error, an unresolvable connection, an operation the connection's type doesn't declare, or the connector's own reported failure
properties.operation	Yes	Operation name, as declared by the connection's type (GET /connections/meta, or o2idctl connections types)
properties.input	No	Map of input key to a CEL expression, evaluated in the same user/tenant/step/risk environment as a condition node's branches
properties.timeoutMs	No	Request timeout in milliseconds. Defaults to a server-side value; always clamped to a hard maximum

Clients do not receive a nodeView for a connection_call node — like webhook, it runs synchronously and the run keeps rendering whatever node comes next.

Context written:

Path	Description
step.<nodeId>.result	The operation's JSON-object output
step.<nodeId>.error	Failure reason, set instead of result when the node routed to onFailure

credential

A credential node grants a credential to the run's current subject as a side effect of the flow — e.g. an age-verification or step-up sequence — instead of requiring a separate API call after the fact. properties.kind selects which kind of credential is granted and which other properties apply.

verifiable_credential is the only kind implemented today: it creates an OID4VCI credential offer against a verifiable credential you've already defined, instead of requiring a separate POST /verifiable-credential-offers call.

{
  "type": "credential",
  "onSuccess": "allow",
  "onFailure": "deny",
  "properties": {
    "kind": "verifiable_credential",
    "verifiableCredentialId": "verifiable-credential-id",
    "claimsFromContext": ["ageOver18"],
    "requireTxCode": false
  }
}

Field	Required	Description
type	Yes	Must be credential
onSuccess	Yes	Node ID to execute once the client acknowledges the credential offer
onFailure	Yes	Node ID to execute on any failure — unknown credential kind, unknown or disabled verifiable credential, or no credential issuance configured on this server
properties.kind	Yes	Which kind of credential to grant. Only verifiable_credential is implemented today
properties.verifiableCredentialId	Yes, for kind verifiable_credential	The verifiable credential to issue against — must exist and be enabled
properties.claimsFromContext	No, kind verifiable_credential only	Context keys to lift into the offer's claims (e.g. a value an earlier script/webhook/outcome step wrote to context). A name with no matching key in context is silently omitted, not an error — the same graph can be reused across scenarios where not every claim is always populated
properties.requireTxCode	No, kind verifiable_credential only	Requests a short numeric PIN alongside the QR code — see Requiring a tx_code PIN

Every claim name in claimsFromContext that does resolve to a value must still be one of the verifiable credential's own defined claim names — the exact same subset-of-schema validation POST /verifiable-credential-offers applies.

Unlike webhook/script, a credential node pauses on success: the run parks on this node and the client receives a nodeView of type credential whose data is the OID4VCI credential_offer object itself (credential_issuer, credential_configuration_ids, grants, credential_offer_uri), plus data.deepLinkUri (the openid-credential-offer:// URI to render as a QR code) and data.txCode (the raw PIN, only present when requireTxCode was set) for the portal to display. The client then submits an empty Advance to continue to onSuccess. A failure never pauses — it routes straight to onFailure, the same as webhook.

Context written:

Path	Description
step.<nodeId>.result	The OID4VCI credential_offer object — the same shape POST /verifiable-credential-offers returns
step.<nodeId>.error	Failure reason, set instead of result when the node routed to onFailure

choice

A choice node presents several sign-in options at once and lets the client pick which one runs, instead of O2ID deciding via a condition node's server-evaluated CEL branches. Each entry in properties.options is a full node — its own id/type/properties/onSuccess/onFailure — not a separate wrapper shape, since an option is simply a node with no name of its own in the top-level nodes map.

{
  "type": "choice",
  "properties": {
    "options": [
      { "id": "password", "type": "authenticator", "onSuccess": "allow" },
      { "id": "<connection-id>", "type": "connection", "onSuccess": "allow", "onFailure": "deny" },
      { "id": "org_sso", "type": "organization_context", "onSuccess": "allow" }
    ]
  }
}

Field	Required	Description
type	Yes	Must be choice
properties.options	Yes	At least one inline node, each with a unique id. Supported option types: authenticator, connection, organization_context

A choice node itself has no onSuccess/onFailure — control flow comes entirely from whichever option's own onSuccess/onFailure runs. An option's onSuccess/onFailure are node IDs in the outer nodes map, exactly like a top-level node's — an option is not a nested sub-graph.

The nodeView for a choice node carries only id/type per option — no display label:

{
  "type": "choice",
  "data": {
    "options": [
      { "id": "password", "type": "authenticator" },
      { "id": "org_sso", "type": "organization_context" }
    ]
  }
}

Deciding what each option looks like (button text, icon) is the client's responsibility, keyed off id/type — the flow definition never authors a label. Submit which option was picked to advance:

{ "values": { "choice": "password" } }

organization_context

An organization_context node detects whether the currently-running flow was reached via a nested Organization's own path (see Organizations) and, if so, surfaces a "signing in to" banner naming it before continuing. Normally used as one option inside a choice node rather than as a standalone entry node. On a run not reached via an Organization path, it's a pure pass-through: it succeeds immediately with no pause at all, as if the node weren't there.

{
  "type": "organization_context",
  "onSuccess": "allow",
  "properties": {
    "mode": "banner"
  }
}

Field	Required	Description
type	Yes	Must be organization_context
onSuccess	Yes	Node ID to execute once the banner is acknowledged (or immediately, on a pass-through run)
properties.mode	No	banner (default) shows the "signing in to" step naming the Organization. redirect/auto are reserved for a future release — validation rejects them today

When it pauses, nodeView.data carries the Organization's display info:

{
  "type": "organization_context",
  "data": {
    "organizationName": "Acme, Inc.",
    "organizationSlug": "acme"
  }
}

The client submits an empty Advance to continue, the same idiom credential nodes use for their own zero-input continue step.

Node outputs

Any node can declare outputs: a map of name to CEL expression, evaluated in the same context as a condition's if (see Condition context above), plus this node's own step.<nodeId> entry, once that node completes. A later node reads the computed value at step.<nodeId>.outputs.<name>.

{
  "type": "webhook",
  "onSuccess": "check_risk",
  "onFailure": "deny",
  "properties": {
    "url": "https://risk.example.com/score"
  },
  "outputs": {
    "highRisk": "step.risk_check.result.score > 70"
  }
}

step.risk_check.outputs.highRisk

Outputs are computed only on a node's success path — a failed node's step.<nodeId> entry has error instead of result/claims, so an expression written against the success shape has nothing to read. An empty or invalid expression fails flow validation at /nodes/<id>/outputs/<name>.

Validation

o2idctl flows validate <flow-id> checks:

Check	Description
Definition shape	start and nodes are present and valid JSON
Node references	Every referenced node ID exists
Authenticator IDs	Referenced configured authenticators exist
Flow references	A flow node's trigger resolves to an active flow, and its nodeId (if set) exists in that flow
CEL expressions	Condition expressions compile
Script syntax	script node source parses as valid Starlark (parsed, not executed)
Webhook URLs	webhook node properties.url is an absolute http(s) URL
Connection call operations	connection_call node properties.operation is declared by the referenced connection's type
Credential kind	credential node properties.kind is a recognized kind
Verifiable credential IDs	A verifiable_credential-kind credential node's properties.verifiableCredentialId is present
Reachability	Every node is reachable from start
Unsupported nodes	Node types not listed by metadata are rejected

Validation errors include a JSON Pointer so tooling can identify the exact field that failed:

{
  "valid": false,
  "errors": [
    {
      "pointer": "/nodes/mfa_check/branches/0/if",
      "message": "ERROR: ..."
    }
  ]
}

Metadata

Use metadata to discover the node types and context variables supported by the running O2ID instance:

o2idctl flows meta

The API also exposes a downloadable schema document:

GET /flows/meta/schema.json