Skip to main content
The agent wizard covers the common fields, but AgentConfig exposes top-level options that the wizard does not surface: pre- and post-conversation workflows, custom recording locations, and noise cancellation. Set them in the agent config JSON or via the API. The full, authoritative shape of every field is published as a JSON Schema. Fetch it before building requests so you always match the deployed version: 
curl https://api.anyreach.ai/core/agent-config-schema \
  -H "Authorization: Bearer <token>"
GET /core/agent-config-schema returns the JSON Schema generated directly from the AgentConfig model, so it is the source of truth for field names, types, defaults, and enum values.

​
Pre-conversation workflow

pre_conversation_workflow runs a workflow before the agent starts. Use it to look up the caller, gate access, or seed the agent with context. The workflow can return three optional outputs:
OutputTypeEffect
rejectbooleanSet to true to abort the conversation (defaults to false).
initial_contextstringContext injected into the agent’s system prompt.
custom_metadataobjectCustom metadata attached to the conversation.
The field accepts a WorkflowAction. Reference a published workflow, or supply an inline definition:
FieldTypeDefaultDescription
type"workflow""workflow"Action type.
workflow_idstring—ID of a published workflow. Pair with workflow_version.
workflow_versioninteger—Version of the referenced workflow.
initial_stepstring—Step to start from. Required when using workflow_id/workflow_version; defaults to trigger for inline definitions.
definitionobject—Inline workflow definition. Mutually exclusive with workflow_id/workflow_version.
conditionstring—Optional condition controlling whether the action runs.
reject_on_errorbooleanfalseIf true, abort the conversation when the workflow fails.
You must specify either workflow_id plus workflow_version, or an inline definition — not both, and not neither. When referencing a workflow, initial_step is required.
{
  "pre_conversation_workflow": {
    "type": "workflow",
    "workflow_id": "wf_...",
    "workflow_version": 3,
    "initial_step": "trigger",
    "reject_on_error": true
  }
}
reject_on_error: true plus a reject output gives you a hard gate: if the lookup fails or the workflow decides to, the conversation never starts.

​
Post-conversation workflow

post_conversation_workflow runs a workflow after the conversation ends — for logging, CRM updates, follow-up, or downstream automation. It accepts the same WorkflowAction shape as the pre-conversation workflow. Its outputs do not affect the (already finished) conversation, so reject has no effect here. 
{
  "post_conversation_workflow": {
    "type": "workflow",
    "workflow_id": "wf_...",
    "workflow_version": 1,
    "initial_step": "trigger"
  }
}
Both hooks invoke standard workflows. For how an agent embeds and triggers a workflow, see Agent-embedded triggers.

​
Custom recording locations

By default Anyreach stores recordings for you. Set custom_recording_locations to copy recordings into your own object storage instead. The field is a list (minimum one entry); each entry is discriminated by provider. Every provider supports file_type, either ogg (default) or mp4.
FieldTypeDefaultDescription
provider"s3""s3"Provider discriminator.
bucketstring—Target bucket.
regionstring—Bucket region.
access_keystring—AWS access key ID.
secret_access_keystring—AWS secret access key.
file_type"ogg" | "mp4""ogg"Recording file format.
{
  "custom_recording_locations": [
    {
      "provider": "s3",
      "bucket": "my-recordings",
      "region": "us-east-1",
      "access_key": "AKIA...",
      "secret_access_key": "...",
      "file_type": "ogg"
    }
  ]
}
For S3, Anyreach can mint a presigned URL to play back a recording from your bucket. GCP and Azure store the recording but do not return a playback URL.

​
Noise cancellation

noise_cancellation enables LiveKit noise-cancellation models on the audio stream. It holds a models list with at least one entry; each entry is discriminated by provider and model.name.
providermodel.nameWhen to use
livekitbvcGeneral-purpose background voice cancellation for wideband audio (web, SIP, VoIP).
livekitbvc_telephonyBackground voice cancellation tuned for narrowband telephony (PSTN phone calls).
{
  "noise_cancellation": {
    "models": [
      { "provider": "livekit", "model": { "name": "bvc_telephony" } }
    ]
  }
}
Match the model to the channel: use bvc_telephony for inbound/outbound phone calls and bvc for web or higher-fidelity audio.

​
Reference

These fields live at the top level of AgentConfig (alongside version, agent, and metadata). Always validate against the live schema before sending requests.
FieldTypeDescription
pre_conversation_workflowWorkflowActionWorkflow run before the agent starts; can reject, inject initial_context, and attach custom_metadata.
post_conversation_workflowWorkflowActionWorkflow run after the conversation ends.
custom_recording_locationsarray of recording locationsStore recordings in your own S3, GCP, or Azure storage.
noise_cancellationobjectLiveKit bvc / bvc_telephony noise-cancellation models.

Agent API usage

Create and update agents, including these config fields, over the API.

Agent-embedded triggers

How agents invoke workflows for pre- and post-conversation hooks.