Feature Configuration

Flow Steward configuration is split across several scopes. The right scope depends on whether the setting is infrastructure-wide, project-specific, or part of a reusable workflow contract.

Configuration scopes

Scope	Use it for	Examples
Deployment environment	Runtime defaults, infrastructure, safety limits, storage, logging.	.env, .env.example, .env.example.ops, .env.example.prod.
Account or project policy	Per-project security, provider, connection, notification, and data access behavior.	Project settings, security policy, project connections.
Workflow settings	Public interface, trigger channels, callable contract, workflow-level behavior.	Workflow detail page, import/export, shared authoring services.
Phase settings	Inputs, outputs, gates, retry, timeout, SLA, and failure behavior.	Phase drawer/editor.
Step settings	Step type, mappings, connector binding, HTTP config, batch policy, and error handling.	Step drawer/editor.
Runtime evidence	Logs, events, runtime journal, security findings, notification deliveries.	Job detail, monitoring, MCP, CLI.

Uploads and file types

General attachment policy is deployment-level:

Variable	Purpose
FS_ATTACHMENT_MAX_SIZE_BYTES	Maximum size per general attachment.
FS_ATTACHMENT_MAX_FILES	Maximum files per request.
FS_ALLOWED_ATTACHMENT_MIMES	Comma-separated MIME allowlist.
FS_ALLOWED_ATTACHMENT_EXTENSIONS	Comma-separated extension allowlist.

Incoming workflow webhooks can also define per-entrypoint limits:

Webhook setting	Purpose
allowed_content_types	Request content types allowed for this webhook.
max_body_bytes	Maximum request body.
max_file_bytes	Maximum uploaded file size.
max_files	Maximum uploaded file count.
source_restrictions.exact_ips	Exact IP allowlist.
source_restrictions.cidrs	CIDR allowlist.

Zip uploads are inspected before being stored.

PII and security policy

Security policy can be global or overridden per project.

Setting	Values	Meaning
enabled	true / false	Turns policy enforcement on or off.
level	minimal, standard, strict	Applies preset severity and scanner behavior.
surfaces.*.action	scan, skip	Controls scanning on inbound, outbound, tool, and index surfaces.

PII modes:

Mode	Behavior
pseudonymize	Replaces detected PII with stable pseudonym tokens.
hard_redact	Replaces detected PII with redacted placeholders.
allow	Detects but leaves values visible.

PII detector toggles include person_name, email, phone, ssn, credit_card, ipv4, and ipv6. Projects can add custom regex rules with a stable rule_id, scanner type, pattern, severity, and description.

Use custom regex rules for identifiers that matter to your organization, such as employee IDs, customer account numbers, internal ticket IDs, tenant IDs, or provider-specific secrets. Keep the rule name stable so findings remain auditable across jobs.

Choose the mode by outcome:

Goal	Mode
Preserve correlation without exposing the raw value	pseudonymize
Remove the value from persisted messages, logs, or traces	hard_redact
Observe findings without changing payloads	allow

Security scanning can apply to inbound data, outbound data, tool calls, and indexing surfaces. Treat externally supplied task text, webhook payloads, extension payloads, and dataset rows as untrusted.

Deployment-level security variables include:

• SECURITY_SCAN_MODE;
• SECURITY_FAIL_CLOSED_ON_POLICY_ERROR;
• SECURITY_PII_SALT;
• FS_TRACE_REDACT_MAX_LENGTH;
• FS_EXTENSION_DEBUG_PAYLOADS;
• FS_EXTENSION_PAYLOAD_REDACT_KEYS;
• FS_EXTENSION_PAYLOAD_REDACT_EXACT_KEYS;
• FS_EXTENSION_PAYLOAD_REDACT_KEY_FRAGMENTS;
• FS_EXTENSION_PAYLOAD_REDACT_MAX_DEPTH;
• FS_EXTENSION_PAYLOAD_STRING_TRUNCATE_LENGTH.

Artifacts and storage

Artifact storage is configured at deployment level.

Variable	Purpose
ARTIFACT_STORAGE_BACKEND	memory, s3, or local_fs.
ARTIFACT_KEY_PREFIX / ARTIFACT_S3_KEY_PREFIX	Optional object key prefix.
ARTIFACT_S3_ENDPOINT_URL	S3-compatible endpoint.
ARTIFACT_S3_BUCKET	Object bucket.
ARTIFACT_S3_ACCESS_KEY	Access key.
ARTIFACT_S3_SECRET_KEY	Secret key.
ARTIFACT_S3_REGION	Region.
ARTIFACT_S3_USE_SSL	SSL toggle.
ARTIFACT_S3_PRESIGN_TTL_SECONDS	Presigned URL TTL.
ARTIFACT_LOCAL_ROOT_DIR	Root path for local filesystem backend.

Artifact retention classes include ephemeral, run_window, audit, and persistent.

Workflows

Workflow configuration can live on the workflow, phase, or step.

Area	Settings
Public interface	trigger_channels, input_mode, strict_inputs, expected_inputs, output_mode, expected_outputs, expected_artifacts, side_effects.
Phase	expected_inputs, required_outputs, retry_policy_jsonb, timeout_policy_jsonb, sla_policy_jsonb, on_failure, gate.
Step	step_kind, input_mapping_jsonb, output_mapping_jsonb, connection_binding_jsonb, batch_policy_jsonb, execution_policy_jsonb, error_policy_jsonb, optional, override.
Agent step	agent_id, skill_id, agent and skill contracts.
Connector step	connector_id, connector_install_id, operation schema, project connection.
HTTP step	Method, URL, headers, query params, body template, auth mode/ref, timeout, expected statuses.
Dataset query step	Query template, params, row limit, timeout, result target.
Nested workflow	Child workflow id, input mapping, output mapping, artifact mapping.

Nested workflow deployment knobs:

• FS_INVOKE_CHILD_TIMEOUT_SECONDS;
• FS_INVOKE_CHILD_TIMEOUT_CAP_SECONDS;
• FS_INVOKE_CHILD_POLL_SECONDS;
• FS_INVOKE_CHILD_CANCEL_ON_TIMEOUT;
• FS_INVOKE_INLINE_EXECUTION.

Knowledge retrieval and memory

Variable	Purpose
FS_MEMORY_PROFILE	Canonical memory profile.
FS_MEMORY_REPOSITORY_BACKEND	Canonical memory repository backend.
FS_MEMORY_KNOWLEDGE_BACKEND	Knowledge backend.
FS_MEMORY_TEXT_SEARCH_BACKEND	Knowledge text-search backend.
FS_MEMORY_VECTOR_BACKEND	Knowledge vector backend.
FS_CONTEXT_GRAPH_BACKEND	Workflow/entity context graph backend, not repository code indexing.
FS_PGVECTOR_DSN	Optional pgvector DSN.

Canonical internal memory is stored in Postgres and supports runtime context and workflow knowledge. These settings do not enable a public KB editor or MCP memory tools. Core does not provide code-index refresh settings or project reindex commands. See Project Context.

Model execution

Select verified providers and models at the project or agent boundary. FS_DEFAULT_PROVIDER, FS_DEFAULT_MODEL and FS_DEFAULT_TEMPERATURE govern intentional runtime defaults; FS_LLM_REQUEST_TIMEOUT_SECONDS sets an optional request timeout. Account-scoped credentials belong in secret-backed provider profiles and are not replaced by ambient keys for an explicit account route.

Notifications

Project notification route settings:

Setting	Purpose
route_key	Stable route or channel key used by automation.
provider_key	Notification provider.
event_type	Event subscription; supports exact and wildcard matching.
severity	Minimum severity.
workflow_id	Optional workflow scope.
route_config	Provider-specific options such as channel id, suffix, timeout.
enabled	Route toggle.

Notification evidence is visible through project notification deliveries, job logs, and event dispatch records when dispatch tracking is available.

Logs, retention, trash, and watchdog

Logging variables include FS_LOG_LEVEL, FS_LOG_FILE_LEVEL, FS_LOG_STDERR_LEVEL, FS_LOG_FORMAT, FS_LOG_DIR, FS_LOG_FILE_BASENAME, FS_LOG_FILE_ENABLED, FS_LOG_STDERR_ENABLED, FS_LOG_CAPTURE_WARNINGS, FS_LOG_RETENTION_DAYS, and FS_LOG_UTC.

Retention variables include FS_RETENTION_LOGS_DAYS, FS_RETENTION_MESSAGES_DAYS, FS_RETENTION_TRASH_DAYS, FS_RETENTION_JOBS_DAYS, FS_RETENTION_EVENTS_DAYS, and FS_RETENTION_RETRY_QUEUE_DAYS. Advanced cleanup knobs include FS_RETENTION_TERMINAL_JOB_CHECKPOINTS_DAYS, FS_WORKSPACES_ROOT, and FS_WORKSPACE_MAX_AGE_DAYS.

Production observability variables include SENTRY_DSN, SENTRY_ENVIRONMENT, SENTRY_RELEASE, SENTRY_TRACES_SAMPLE_RATE, SENTRY_PROFILES_SAMPLE_RATE, SENTRY_SEND_DEFAULT_PII, FS_METRICS_TOKEN, FS_PROCESS_METRICS_HOST, FS_WORKER_METRICS_PORT, and FS_CONTROL_METRICS_PORT.

Watchdog variables include:

• FS_WATCHDOG_ENABLED;
• FS_WATCHDOG_INTERVAL_SECONDS;
• FS_WATCHDOG_AUTO_APPLY;
• FS_WATCHDOG_STUCK_SECONDS;
• FS_WATCHDOG_ORPHAN_GRACE_SECONDS;
• FS_WATCHDOG_PICKUP_GRACE_SECONDS;
• FS_WATCHDOG_MAX_JOURNAL_ENTRIES.

Use watchdog scan/recover APIs and the runtime journal when diagnosing stuck or interrupted jobs.

Retention settings can also be managed from Settings > Retention when the UI is enabled for the current role. Use deployment variables for default policy and UI settings for operator-managed retention windows.

Trash is the operator recovery layer for accidental deletion. Lifecycle plans are the safety layer for restore and hard-delete operations because they show the dependency tree and direct conflicts before an irreversible action.