Studio Node Reference

Every node type available in the FAOS Studio palette, organized by category. Each entry lists what the node does, its key configuration fields, and its runtime support status — this last column is sourced from the same honesty manifest Studio itself uses to warn you on the canvas, so this page never claims more than the product actually does.

Keep this page in sync

This reference is seeded from the palette's own entries (NodePalette.tsx) and the server-side runtime-support manifest. If you add or change a node type in the palette, update this page in the same change.

Runtime support statuses

Status	Meaning
✅ Full	The runtime executes this node with its real, intended semantics today.
🟡 Authoring-only	You can draw and configure this node, and it exports with full semantics to LangGraph/ADK — but the FAOS runtime itself doesn't execute its distinct behavior yet. Studio tells you this on the canvas; it never fakes a pass.

(There is no "degraded" status today — every previously-degraded node type has graduated to full support.)

Flow Control

Node	Description	Key config	Status
Start	Entry point	triggerType (manual / scheduled / event / webhook)	— (flow boundary, not a step)
End	Terminal node	outputVariable	— (flow boundary, not a step)
Loop	Iterate or repeat	maxIterations, condition, loopVariable	🟡 Authoring-only — no loop construct in the executor yet
Parallel Branch	Concurrent paths	branches[], joinStrategy (all / any / first)	✅ Full — compiled away into real depends_on fan-out/fan-in; the executor already runs same-level steps concurrently
Condition (Rule)	Rule-based branch	ruleExpression, trueLabel, falseLabel	✅ Full — compiles to a dmn step
Condition (Agent)	Agent-evaluated branch	agentId, evaluationPrompt, trueLabel, falseLabel	🟡 Authoring-only — no LLM-routing branch construct yet (rule-based dmn only)

AI

Node	Description	Key config	Status
Agent	AI agent task	agentId, model, skills[], timeout, systemPrompt	✅ Full
LLM	Direct LLM call (no agent roster resolution)	promptTemplate, model, temperature, maxTokens, responseFormat	✅ Full — a thin step calling resolve_model() directly; a missing model config fails the step honestly rather than silently falling back

Handoff Patterns

Node	Description	Key config	Status
Manager Agent	Delegates to sub-agents	agentId, model, skills[], sub_agent_pool[], delegation_criteria	✅ Full — compiles to a structured-prompt agent invocation (its semantic is a governed agent step)
A2A Delegate	Delegates to an external agent over the A2A protocol	endpoint_url, agent_card_url, protocol_version, timeout_seconds	✅ Full — same structured-prompt compile as Manager Agent

Compliance

Node	Description	Key config	Status
Compliance Gate	Multi-check gate	checks[] (type, action, threshold), failStrategy	✅ Full — runs the real DefaultComplianceEvaluator; unbacked check kinds record not_evaluated, never a fabricated pass
Audit Log	Records an audit trail	retentionYears, logInputs, logOutputs, hashSensitive	✅ Full — writes a real AuditLogEntry
Policy Check	Regulatory framework check	frameworks[] (SOX / GDPR / CCPA / MiFID II / Dodd-Frank / ECOA / FINRA), action	✅ Full
PII Scanner	Detects and redacts PII	entityTypes[] (SSN, DOB, account number, name, address, phone, email), action, redactionPattern	✅ Full — real PII detection via the canonical DLP scanner
Bias Check	Bias detection	domain (lending / insurance / hiring / general), regulations[], sensitivity	✅ Full — regulatory/bias check kinds honestly record not_evaluated where no scoring model is bound, rather than fabricating a pass

Interaction

Node	Description	Key config	Status
Human Input	Requests input from a person mid-flow	inputPrompt, inputType (text / choice / file), timeout	🟡 Authoring-only — no mid-flow input capture in the executor yet (distinct from Approval Gate, which pauses for a yes/no decision and is implemented)
Approval Gate	Requires human approval before continuing	requiredApprovers[], approvalType (any / all), timeout, escalationEmail	✅ Full
Direct Reply	Sends a response to the user	responseTemplate, channel (chat / email / webhook)	✅ Full — compiles to a delivery step

Execution

Node	Description	Key config	Status
Tool Call	Executes a configured tool	toolName, parameters, retryOnFailure, maxRetries	✅ Full — invokes through the MCP layer with tenant policy enforced; an unbound tool fails honestly rather than silently running as a generic agent step

Case Management

Node	Description	Key config	Status
Create Case	Opens a new case	caseType, initialPriority (P1-P4), sla (goal/deadline)	✅ Full — dispatches onto the real case engine
Update Case	Updates case fields	fields[] (field/value pairs)	✅ Full
Route Case	Routes a case to an agent	strategy (skill-based / round-robin / manual), agentPool[], requiredSkills[]	✅ Full
Escalate	Escalates a case	reason, targetTier, urgency	✅ Full
Resolve	Resolves a case	resolutionCode, summaryTemplate	✅ Full
SLA Timer	Monitors a case against an SLA	goalHours, deadlineHours, passedDeadlineHours, autoEscalate	🟡 Authoring-only — SLA timing lives inside the case-lifecycle workflow's own stages, not as an independent playbook step, today
Satisfaction Survey	Sends a customer survey	surveyTemplate, channel (email / SMS / in-app)	🟡 Authoring-only — the survey send-only substrate exists, but response collection into the playbook does not yet

See also

• Studio Quickstart — your first Journey in 10 minutes, walking through five of these node categories on a real example.