Terraform Resources - Orq.ai Documentation

Each resource links to its full schema reference on the Terraform Registry, including import syntax and complete examples.

Workspace administration

Resource	Manages
`orq_project`	Projects, including team assignment
`orq_workspace_settings`	Workspace display name, model enforcement, PII redaction
`orq_api_key`	API keys: project scoping, permission presets or per-domain grants, expiry
`orq_management_key`	Management Keys for automation and delegation

Models

Resource	Manages
`orq_workspace_model`	Enabling catalog models and sharing them with projects
`orq_model`	Custom OpenAI-compatible models with a private base URL
`orq_bedrock_model`	AWS Bedrock models, via Pod Identity or a dashboard integration
`orq_routing_rule`	Model routing: fallback, weighted, latency-based, round-robin

Quality and cost controls

Resource	Manages
`orq_evaluator`	LLM-as-judge and Python Evaluators
`orq_guardrail_rule`	Guardrails: built-in PII and secret detection, or custom evaluator-backed rules
`orq_budget`	Spend, token, and rate ceilings with threshold alerts
`orq_notifier`	Email and webhook alert destinations

Data sources

Data source	Returns
`orq_projects`	Projects in the workspace, for referencing existing ones by name

Example: a guardrail backed by an evaluator

A taste of composing resources: an LLM Evaluator judged by a model, enforced as a blocking guardrail.

resource "orq_evaluator" "tone" {
  key         = "professional-tone"
  type        = "llm_eval"
  project_id  = orq_project.production.id
  mode        = "single"
  model       = "openai/gpt-5.6-luna"
  prompt      = "Return true when the response maintains a professional tone.\n\nResponse:\n{{log.output}}"
  output_type = "boolean"
}

resource "orq_guardrail_rule" "tone_guard" {
  display_name = "Professional tone only"
  enabled      = true

  guardrails {
    id           = orq_evaluator.tone.id
    execute_on   = "output" # input | output | both
    is_guardrail = true     # block on failure instead of observe-only
  }
}