generated: '2026-08-14'
method: searched
source: >-
  https://developer.regal.ai/docs/webcall-testing,
  https://developer.regal.ai/docs/test-logic-chat,
  https://developer.regal.ai/docs/test-cases-simulation,
  https://developer.regal.ai/docs/test-knowledge-base,
  https://developer.regal.ai/docs/saving-and-updating-agents,
  https://developer.regal.ai/reference/overview
note: >-
  Regal ships a substantial testing surface, but it is an AGENT-testing surface, not an
  API sandbox. There is no test mode, no separate sandbox host, no test/live key prefix
  pair, and no magic test values of any kind — every documented API call goes to the
  production host with the production key. What Regal does publish is a full pre-production
  loop for the AI agents themselves (web calls, chat logic tests, bulk simulation with
  auto-evaluated success criteria, knowledge-base coverage checks) plus Draft/Active agent
  versioning so changes can be staged without touching live calls. Recorded honestly:
  api_sandbox false, agent_sandbox true.
api_sandbox:
  available: false
  test_mode: false
  test_host: null
  key_prefixes: null
  test_values: null
  detail: >-
    The docs describe exactly one key ("An API key must be included in the Authorization
    header for all requests"), issued by emailing support@regal.ai, with no test/live
    distinction. Nothing in the published OpenAPI or docs marks a sandbox server. An
    integrator's only safe rehearsal path is a non-production Regal brand, which Regal does
    not document publicly.
public_collection:
  type: postman
  name: Regal Public Workspace
  run_url: https://app.getpostman.com/run-collection/17258986-81c59f40-7e22-480e-bb40-aa29250b0e35
  docs: https://developer.regal.ai/reference/overview
  note: >-
    Regal publishes a runnable Postman collection ("Run in Postman") for the Custom Events
    endpoint so an integrator can fire test events without writing code — but it still runs
    against production with a real key.
agent_sandbox:
  available: true
  tools:
    - id: test-audio
      name: Test Audio (Web calls)
      docs: https://developer.regal.ai/docs/webcall-testing
      description: >-
        Place a web call to the agent from inside the builder before any phone number or
        routing exists.
      can_test: [voice sound/speed/volume, responsiveness and interruption handling, conversational logic, RAG responses, custom actions, personalization using test values for dynamic variables]
      cannot_test: [call transfers, end-to-end telephony latency, personalization from real profile data, routing, dispositions]
    - id: test-logic
      name: Test Logic (Web chat)
      docs: https://developer.regal.ai/docs/test-logic-chat
      description: Iteratively exercise agent logic over chat while building.
    - id: simulation
      name: Simulation Testing
      docs: https://developer.regal.ai/docs/test-cases-simulation
      description: >-
        Bulk automated validation across many test cases before go-live and after live
        iterations.
      test_case_fields:
        - {field: Name, description: Brief identifier for what the case covers.}
        - {field: Contact Prompt, description: "Prompt defining the simulated contact's behaviour, persona and dialogue."}
        - {field: Success Criteria, description: Expected agent behaviour, used to drive the auto-eval outcome and per-run stats.}
        - {field: Test Variable, description: Saved variable values giving the agent context/personalization data for the run.}
      auto_generation: >-
        "Get Me Started" auto-generates 10 test cases from the existing agent prompt, but
        only when no test cases exist yet.
      can_test: [conversational logic under diverse scenarios, objection and question handling, RAG responses, custom actions triggered during conversation, many scenarios in parallel against one prompt]
      cannot_test: [voice sound/speed/volume, responsiveness and interruption handling, transfers, end-to-end telephony latency, personalization from real profile data, dispositions]
      mcp_tools: [fetch-test-cases, create-test-cases, edit-test-case, delete-test-cases, start-simulation-run, get-simulation-progress, fetch-simulation-transcript]
    - id: knowledge-base-testing
      name: Knowledge Base Testing
      docs: https://developer.regal.ai/docs/test-knowledge-base
      description: Spot-check knowledge bases for coverage gaps.
staging_model:
  mechanism: Draft and Active agent versions
  docs: https://developer.regal.ai/docs/saving-and-updating-agents
  description: >-
    Changes are saved as a Draft version and promoted to Active, so an agent can be edited
    without disrupting live calls, and rolled back to a previous version. MCP write tools
    only ever save drafts.
  variants: >-
    Agent variants with traffic allocation provide live A/B splitting as a second staged
    rollout mechanism.