Ctrlk
For the complete documentation index, see llms.txt. This page is also available as Markdown.

guardrails

Safety and compliance checks (adaptorType: GuardrailImport) that validate data flowing through integrations — PII detection, content moderation, and custom AI-based evaluation. Unlike ai-agents, which produce content with LLMs, guardrails screen content: they flag, reject, or mask records that violate policy.

REST API: Guardrails are a specialized import adaptor (GuardrailImport). Schemas live in the imports spec — see the Guardrail and GuardrailImport components. The CLI filters and auto-tags /v1/imports so every call targets a guardrail-shaped record.

celigo guardrails <subcommand> [args] [flags]

Supports all global flags.

Subcommands

Subcommand
Purpose

list

List all guardrail imports in the account.

get <id>

Fetch one guardrail by ID.

create

Create a guardrail from JSON on stdin (adaptorType auto-set).

update <id>

Full replace from stdin (destructive PUT).

set <id> key=value …

Safe field edit: GET → modify → PUT.

delete <id>

Delete a guardrail.

invoke [id]

Invoke a saved guardrail against piped records, or preview a piped guardrail doc when no id is given.

enable-debug <id>

Enable debug logging via PATCH on debugUntil.

disable-debug <id>

Clear debugUntil.

clone <id>

Clone the guardrail (same-env or cross-env).

replace-connection <id> <newConnectionId>

Replace the underlying connection (BYOK ai_agent guardrails only).

celigo guardrails list

List every guardrail import in the account. The CLI fetches /v1/imports and filters client-side to rows where adaptorType === "GuardrailImport".

Signature

Arguments

None.

Flags

None beyond the global flags (--format, --jq, --profile, --verbose).

Example

Corresponds to: GET /v1/imports (client-side filter on adaptorType)

celigo guardrails get

Fetch a single guardrail import by ID.

Signature

Arguments

Argument
Type
Required
Description

<id>

string

Yes

Guardrail import _id.

Example

Corresponds to: GET /v1/imports/{_id}

celigo guardrails create

Create a guardrail import from a JSON body on stdin. The CLI forces adaptorType = "GuardrailImport" before posting, so you don't need to include it in the payload.

Signature

Arguments

None. Reads JSON on stdin.

Request body

See the Guardrail and GuardrailImport schemas for the full payload shape (type, confidence threshold, and the pii / moderation / aiAgent sub-configurations). Key fields:

  • guardrail.type — one of pii, moderation, ai_agent.

  • guardrail.confidenceThreshold — detection threshold (0–1, default 0.7).

  • guardrail.pii.entities[] / .mask — required for pii.

  • guardrail.moderation.categories[] — required for moderation.

  • guardrail.aiAgent — required for ai_agent; same shape as an agents config, including optional _connectionId for BYOK.

Example

Corresponds to: POST /v1/imports (with adaptorType=GuardrailImport injected)

celigo guardrails update

Replace a guardrail import with the JSON body on stdin. Destructive — any field omitted from the payload is erased. Prefer set for targeted edits.

⚠️ PUT fully replaces the guardrail. Any field omitted from the payload is erased. Prefer set for targeted edits, or get | jq | update to preserve the rest of the document.

Signature

Arguments

Argument
Type
Required
Description

<id>

string

Yes

Guardrail import _id.

Request body

The complete guardrail document. See the GuardrailImport schema.

Example

Corresponds to: PUT /v1/imports/{_id}

celigo guardrails set

Safe field edit: GET → apply one or more key=value assignments → PUT the full object back.

Signature

Arguments

Argument
Type
Required
Description

<id>

string

Yes

Guardrail import _id.

<assignments...>

string

Yes (≥1)

One or more key=value pairs. Values are JSON-parsed (disabled=false → boolean). Dot + bracket notation supported (e.g. guardrail.pii.mask=true). key=null removes the field.

Example

Corresponds to: GET /v1/imports/{_id} then PUT /v1/imports/{_id}

celigo guardrails delete

Delete a guardrail import by ID. Prompts for confirmation unless -y is passed.

⚠️ Deleting a guardrail is destructive. Any flow or tool that references it loses the guardrail step. Confirm it is unused before deleting.

Signature

Arguments

Argument
Type
Required
Description

<id>

string

Yes

Guardrail import _id.

Flags

Flag
Type
Default
Description

-y, --yes

boolean

false

Skip the interactive confirmation prompt.

Example

Corresponds to: DELETE /v1/imports/{_id}

celigo guardrails invoke [id]

Run a guardrail without creating a job. Two modes:

  • With <id> — invoke a saved guardrail against records piped on stdin; the response is the guardrail's evaluation for each record (POST /v1/imports/{_id}/invoke).

  • Without <id> — pipe a guardrail document on stdin to preview its evaluation without saving (POST /v1/imports/preview; the CLI sets adaptorType: GuardrailImport). Stdin is required in this mode.

Signature

Arguments

Argument
Type
Required
Description

[id]

string

No

Guardrail import _id. Omit to preview a piped guardrail document.

Examples

Corresponds to: POST /v1/imports/{_id}/invoke (with id) or POST /v1/imports/preview (without).

celigo guardrails enable-debug

Enable debug logging on a guardrail by setting debugUntil via PATCH. Debug windows are capped at 60 minutes.

Signature

Arguments

Argument
Type
Required
Description

<id>

string

Yes

Guardrail import _id.

Flags

Flag
Type
Default
Description

--duration <minutes>

number

60

Debug duration in minutes. Capped at 60.

Example

Corresponds to: PATCH /v1/imports/{_id} (JSON Patch: replace /debugUntil)

celigo guardrails disable-debug

Clear debugUntil on a guardrail import, stopping debug log collection.

Signature

Arguments

Argument
Type
Required
Description

<id>

string

Yes

Guardrail import _id.

Example

Corresponds to: PATCH /v1/imports/{_id} (JSON Patch: remove /debugUntil)

celigo guardrails clone

Clone a guardrail import. Two modes:

  • Same-env clone — no stdin. The CLI reads the source guardrail, builds a self-map from its _connectionId (empty map for non-BYOK guardrails, since pii and moderation types don't require a connection), and submits the clone.

  • Cross-env clone — pipe JSON on stdin with a connectionMap that maps source connection IDs to target connection IDs: {"connectionMap":{"sourceConnId":"targetConnId"},"name":"..."}.

Signature

Arguments

Argument
Type
Required
Description

<id>

string

Yes

Source guardrail import _id.

Flags

Flag
Type
Default
Description

--name <name>

string

"Clone - <source name>"

Name for the cloned guardrail. Defaults match the UI.

Example

Corresponds to: POST /v1/imports/{_id}/clone

celigo guardrails replace-connection

Replace the connection referenced by a guardrail — only meaningful for ai_agent guardrails with a BYOK _connectionId. pii and moderation guardrails don't carry a connection.

Signature

Arguments

Argument
Type
Required
Description

<id>

string

Yes

Guardrail import _id.

<newConnectionId>

string

Yes

The replacement connection's _id.

Example

Corresponds to: PUT /v1/imports/{_id}/replaceConnection

Gotchas

  • Three guardrail types, three required sub-configs. guardrail.type = "pii" needs guardrail.pii.entities[] with at least one entity. type = "moderation" needs guardrail.moderation.categories[] with at least one category. type = "ai_agent" needs guardrail.aiAgent (same shape as agents).

  • PII detection example entities: email_address, phone_number, us_social_security_number, credit_card_number, persons_name, ip_address, and a long list of country-specific IDs. See the Guardrail.pii.entities enum for the full list.

  • Moderation categories cover hate, harassment, self_harm, sexual, violence, illicit, and their intensified variants (*_threatening, *_graphic, *_minors, etc.). Pick only what you intend to enforce — every category adds latency.

  • confidenceThreshold defaults to 0.7. Lower values catch more potential issues but raise false positives; raise it when you're seeing noisy flags on benign records.

  • No _connectionId required for pii or moderation. The guardrail runs inside Celigo. Only ai_agent guardrails may carry a BYOK _connectionId; use replace-connection to swap it.

  • pii.mask: true rewrites the record on its way out — callers see redacted values instead of the raw input. mask: false only flags detections without modifying the record.

  • invoke doesn't create a job. Evaluations are ephemeral; use enable-debug + the imports debug-requests flow (since guardrails are imports) to capture runs inside flows.

  • API path is /v1/imports, not /v1/guardrails. The CLI command is its own namespace, but every request is on the shared imports endpoint with adaptorType=GuardrailImport as the discriminator.

Related

  • ai-agents — LLM-powered import steps that generate or transform content. Guardrails screen; agents produce.

  • tools — reusable building blocks that can embed guardrail steps alongside exports, imports, and lookups.

  • mcp-servers — expose tools (and, transitively, guardrail-protected steps) to MCP clients.

  • imports — the parent resource type. Generic import commands also work on guardrail IDs.

Previousgroups
Nexthttp-connectors

Last updated

Was this helpful?