Update an Agent

Change an Agent: instructions, first message, name, language, time zone, settings, integration tools, Knowledge, procedures, connection procedures or status. Omitted fields stay and null clears one; settings, integrationTools, integrationProcedures, procedures and knowledgeDocumentIds replace the whole value when given (keep each existing procedure's id), so read the Agent with get_agent first and send its version as expectedVersion (VERSION_CONFLICT: read it again). status: paused stops it from answering and calling. modality changes only while it has no channels and no published voice line (AGENT_MODALITY_LOCKED). A voice Agent republishes in the background: check synced.

Scope: raiz:agents:write
Credentials: an API key, a client-credentials token or a person's access token
Effect: writes; idempotent, safe to retry
Limits: request body up to 512 KB, response up to 1 MB
CLI: arbol raiz agents update · MCP tool: raiz_agents_update

PATCH

https://api.getarbol.com/v1/raiz/agents/{agentId}

Authorization

Authorization
string
required
Bearer <token>, where the token is an API key, a client-credentials token or a person's access token with the scope raiz:agents:write. Get an API key.

Path parameters

agentId
string
required
Agent id.

pattern: ^agt_(?:[0-7][0-9a-hjkmnp-tv-z]{25}|(?:[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}|[a-f0-9]{32}))$

Request body JSON required

description
optional
string | null
maxLength: 2000
firstMessage
optional
string | null
What a voice Agent says first when a call connects, inbound or outbound. Text Agents do not use it: how they open a WhatsApp conversation is settings.greeting, and an outbound WhatsApp conversation opens with an approved template.

maxLength: 2000 · minLength: 1
instructions
optional
string | null
What the Agent does and how: its role, what it may and may not do, the facts it relies on.

maxLength: 100000 · minLength: 1
integrationProcedures
optional
array of
object
Text Agents. Procedures a connection publishes (listConnections, checked live in Account), each saved with its current text and run as a free-form procedure beside procedures. In an update, the complete new set.
- connectionId
  string
  required
  A connection's id in Account (con_…).
  
  maxLength: 255 · minLength: 1 · pattern: ^[A-Za-z0-9_-]+$
- procedure
  string
  required
  The procedure's name, as its connection lists it.
  
  maxLength: 64 · minLength: 1 · pattern: ^[a-z0-9]+(-[a-z0-9]+)*$
}
integrationTools
optional
array of
object
The integrations' tools the Agent may call, each on one of the organization's connections (listConnections, checked live in Account). Each is saved with its current definition. In an update, the complete new set.
- connectionId
  string
  required
  A connection's id in Account (con_…).
  
  maxLength: 255 · minLength: 1 · pattern: ^[A-Za-z0-9_-]+$
- tool
  string
  required
  The tool's name, as its connection lists it.
  
  maxLength: 128 · minLength: 1 · pattern: ^[A-Za-z0-9_.-]+$
}
knowledgeDocumentIds
optional
array of
string
The Knowledge documents the Agent consults. In an update, the complete new set.

pattern: ^(?:kdc_(?:[0-7][0-9a-hjkmnp-tv-z]{25}|(?:[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}|[a-f0-9]{32}))|(?:con)_(?:[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}|[a-f0-9]{32}))$
language
optional
string
BCP 47 language tag.

pattern: ^[a-z]{2,3}(-[A-Za-z0-9]{2,8})*$
procedures
optional
array of
object
The Agent's procedures, as a complete set: a procedure left out is removed. Keep each existing procedure's id; a new one may omit it.
- name
  string
  required
  A short label for people. The model never reads it.
  
  maxLength: 120 · minLength: 1
- type
  free_form
  or
  structured
  required
  free_form: instructions the Agent adapts. structured: steps that run the same way every time. Fixed once created.
- content
  optional
  string
  Free-form procedures: what to do, in Markdown, with inline references such as [tool id="buscar_citas"], [kb id="kdc_…"], [procedure id="prc_…"], [system_tool id="end_conversation"] and {{contact_name}}. Empty for structured procedures.
  
  maxLength: 50000
- id
  optional
  string
  The procedure's id (prc_…), stable for the Agent's life.
  
  pattern: ^prc_[0-7][0-9a-hjkmnp-tv-z]{25}$
- status
  optional
  active
  or
  paused
  A paused procedure runs on no channel.
- steps
  optional
  array of
  Structured procedures: the steps, in order. Empty for free-form procedures.
  
  One of the following:
  object
  object
  object
  object
  object
  object
  object
- trigger
  optional
  string
  When it applies, from the contact's side: When the contact asks to book or move an appointment. Empty makes a sub-procedure, which only runs when another procedure references it.
  
  maxLength: 1000
}
settings
optional
object
How the Agent behaves. In an update, the complete new settings: keys left out return to their default. Text and voice settings apply only to Agents of that modality.
- humanEscalation
  optional
  object | null
  Per-Agent human escalation policy. Instructions apply to voice and text. Null keeps the default instructions. A WhatsApp handoff stays with the team until a person explicitly resumes AI; no timeout resumes it.
  
  instructions
  string | null
  required
  When this Agent should ask the business's team to intervene, using only its available handoff capabilities. Null keeps the default instructions; an explicit request for a person is always respected.
  
  maxLength: 4000
- recoveryManager
  optional
  object | null
  Optional WhatsApp recovery advisor. Null or enabled=false disables it. A journaled evidence check precedes any paid review. One tool-free model call per turn before autonomous escalation or after terminal provider failure; unresolved terminal recovery hands off. Completed handoffs and conservatively recognized human requests bypass generation. Model selects the existing provider route. Prices must be conservative upper bounds including provider overhead; missing or invalid prices disable recovery. The cost bound covers only the advisor. No tools or automatic provider retries.
  
  enabled
  boolean
  required
  model
  arbolito/agile
  or
  arbolito/balanced
  or
  arbolito/advanced
  or
  arbolito/top
  required
  arbolito/agile: quickest and least expensive, for direct conversations. arbolito/balanced: more deliberate answers at moderate cost. arbolito/advanced: strong reasoning for complex instructions and many tools, with an adjustable reasoningEffort. arbolito/top: the most capable, for the hardest cases, with an adjustable reasoningEffort up to xhigh. Every tier reads images and PDF documents.
  
  maxAttempts
  number
  required
  timeoutMs
  integer
  required
  maximum: 15000 · minimum: 100
  maxInputTokens
  integer
  required
  maximum: 131072 · minimum: 4096
  maxOutputTokens
  integer
  required
  maximum: 1024 · minimum: 128
  maxCostUsd
  number
  required
  maximum: 1
  inputUsdPerMillion
  number
  required
  maximum: 1000
  outputUsdPerMillion
  number
  required
  maximum: 1000
  instructions
  optional
  string | null
  Operator criteria for when the recovery advisor should help or leave the case to a person. They never override consent, tool permissions, unknown-write fences or an explicit human request. Null keeps the default criteria.
  
  maxLength: 4000
  advisorAgentId
  optional
  string | null
  Optional active text Agent in the same organization, other than this Agent, whose instructions and procedures provide specialist guidance. The configured recovery model and cost bounds stay unchanged; no specialist tools or conversation history are accessed.
  
  pattern: ^agt_(?:[0-7][0-9a-hjkmnp-tv-z]{25}|(?:[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}|[a-f0-9]{32}))$
- automaticKnowledge
  optional
  boolean | null
  Text Agents. Retrieve relevant passages from this Agent's active Knowledge before a substantive contact message, without a model tool call. Null or false: on-demand search only. Does not store contact memories or learn unreviewed answers.
- attachmentReader
  optional
  vision
  or
  ocr
  or
  null
  Text Agents on WhatsApp. Who reads an image or PDF when the Agent calls read_attachment: vision (a vision model describes and transcribes it, with a stronger one when the first reads nothing) or ocr (text recognition, which also answers the fields the Agent asks for, each null when absent or unclear; the vision model reads the file when recognition cannot). With ocr the Agent sees files as markers and reads them with read_attachment; with vision, tiers that read files themselves see the original. Null: vision.
- budgetExhaustedMessage
  optional
  string | null
  Text Agents on WhatsApp. What the contact reads when the Agent has used its monthlyTokenBudget: the conversation then goes to the team, as an escalation does, and the Agent writes nothing until the budget is raised or the month ends. Null: handoffMessage, including an empty notice when disabled.
  
  maxLength: 1000 · minLength: 1
- businessHours
  optional
  One of the following:
  RaizBusinessHours
  closedDates
  array of
  string
  required
  Local dates (YYYY-MM-DD) closed all day, such as holidays.
  
  format: date
  outsideHoursMessage
  string
  required
  What whoever writes outside the hours receives, once per number and closed period (until the hours open again), whether the conversation's AI is on, off or paused. With outsideHoursTemplate, the template goes instead, and this text only when WhatsApp refuses the template. The notice is never taken as an answer: the AI answers what the contact wrote when a person turns it on or resumes it in business hours.
  
  maxLength: 1000 · minLength: 1
  timeZone
  string | null
  required
  IANA time zone of the windows. Null uses the organization's.
  
  maxLength: 64 · pattern: ^[A-Za-z]+(/[A-Za-z0-9_+-]+)*$
  windows
  array of
  object
  required
  Several windows may share a day, for a break at midday or an afternoon shift.
  
  days
  array of
  MO
  or
  TU
  or
  WE
  or
  TH
  or
  FR
  or
  SA
  +1 more
  required
  RFC 5545 weekdays the window applies to.
  
  from
  string
  required
  Local wall-clock time HH:MM (24-hour).
  
  pattern: ^([01][0-9]|2[0-3]):[0-5][0-9]$
  to
  string
  required
  Closing time HH:MM, exclusive; 24:00 closes at midnight. Must be after from: split a window that crosses midnight in two.
  
  pattern: ^(([01][0-9]|2[0-3]):[0-5][0-9]|24:00)$
  closingGraceMinutes
  optional
  15
  or
  30
  or
  60
  or
  120
  or
  null
  Minutes after a closing during which a conversation in progress at it is still attended: the AI keeps answering, with its tools, and no notice is sent. In progress: the contact wrote before the closing, and the AI or a person answered within these minutes before it, or since (a notice is not an answer). New conversations get the notice as always. Null or absent: off.
  
  outsideHoursTemplate
  optional
  Send the notice as this approved WhatsApp template instead of outsideHoursMessage: a template of the number's account with no variables and no document, image or video header (in list_templates: sendable, with parameterCount 0). WhatsApp's approval is checked when the hours are saved. If WhatsApp refuses the template when a notice goes (deleted, paused or disabled since), outsideHoursMessage goes instead. Null or absent: the notice is outsideHoursMessage.
  
  One of the following:
  RaizOutsideHoursTemplate
  null
  null
- closingMessage
  optional
  string | null
  Text Agents on WhatsApp. After a reply that completed an action (an integration tool that changes data succeeded), explicitly finished the case with end_conversation and left nothing pending (no question, no offered choices, no handoff), a contact message that only thanks or closes («gracias», «ok», «listo», an emoji) gets this text at once, without the model; the Agent still reads both in its next turn. Write it in the Agent's language and ask nothing in it. Null: the Agent answers such messages itself.
  
  maxLength: 1000 · minLength: 1
- evalGate
  optional
  boolean | null
  Text and voice Agents. When true, an updateAgent that changes what the model reads (instructions, procedures, connection procedures, model or reasoningEffort) must send evalRunId: a completed createEvalRun of this Agent with exactly that content, in which every case passed (EVAL_GATE_UNMET otherwise). Null: no requirement.
- followUpAfterMinutes
  optional
  integer | null
  Text Agents. When the Agent spoke last and asked something, send followUpMessage once after this many minutes of silence (WhatsApp only, inside its 24-hour window). Null disables follow-ups.
  
  maximum: 1440 · minimum: 1
- followUpMessage
  optional
  string | null
  Text Agents. The follow-up text. Null uses a short default in the Agent's language.
  
  maxLength: 1000 · minLength: 1
- greeting
  optional
  One of the following:
  RaizAgentGreeting
  null
- handoffMessage
  optional
  string | null
  Text Agents on WhatsApp. The notice appended after an escalation or a requested human handoff. Null: the current localized review notice or the case-specific notice supplied under the Agent's instructions. An empty string disables the notice; the handoff and pause still happen. The model does not write this notice.
  
  maxLength: 1000
- humanReplyTimeoutMinutes
  optional
  integer | null
  Text Agents on WhatsApp. After a manual takeover, resume automatic replies when the contact's newest message waits this many minutes. An Agent handoff always stays with the team until a person explicitly resumes AI. Null: only a person resumes automatic replies.
  
  maximum: 1440 · minimum: 1
- investigationsEnabled
  optional
  boolean | null
  Opt this Agent into internal root-cause investigations of tool failures, human escalations, failed outcomes and negative feedback. Incidents are grouped only with participating Agents in the same organization. Null or false: disabled. Turning this off stops new events and pending dispatch when no participating Agent remains; an already active investigation may finish.
- model
  optional
  arbolito/agile
  or
  arbolito/balanced
  or
  arbolito/advanced
  or
  arbolito/top
  or
  null
  Text Agents. The tier that writes replies. arbolito/agile: quickest and least expensive, for direct conversations. arbolito/balanced: more deliberate answers at moderate cost. arbolito/advanced: strong reasoning for complex instructions and many tools, with an adjustable reasoningEffort. arbolito/top: the most capable, for the hardest cases, with an adjustable reasoningEffort up to xhigh. Every tier reads images and PDF documents. Null: the default, as arbolito/agile.
- modelTrial
  optional
  One of the following:
  RaizAgentModelTrial
  model
  arbolito/agile
  or
  arbolito/balanced
  or
  arbolito/advanced
  or
  arbolito/top
  required
  The tier the trial's conversations are answered with.
  
  percent
  integer
  required
  The share of new conversations in the trial, in percent.
  
  maximum: 50 · minimum: 1
  reasoningEffort
  none
  or
  low
  or
  medium
  or
  high
  or
  xhigh
  or
  null
  required
  The trial tier's reasoning, as reasoningEffort. Null: the tier's default.
  
  null
- monthlyTokenBudget
  optional
  integer | null
  Text Agents. Tokens (input, cached input included, plus output) the Agent may use per calendar month (UTC), on every channel. On WhatsApp, a reply that would start past it does not run: the contact reads budgetExhaustedMessage and the team continues; at 80 % the API logs an alert once a month. Null: no limit.
  
  maximum: 1000000000000 · minimum: 10000
- pauseOnHumanReply
  optional
  boolean | null
  Text Agents on WhatsApp: pause automatic replies when a live WhatsApp Business App message that says something (not a reaction, an edit or a deletion) proves a person took over. Null: disabled. The recipient stays paused until resumed or until that conversation closes; their next conversation starts with the AI.
- policy
  optional
  One of the following:
  RaizAgentTurnPolicy
  maxOutputTokens
  optional
  integer
  Output tokens per model call, reasoning included. Default 4096.
  
  maximum: 16384 · minimum: 256
  maxSteps
  optional
  integer
  Model calls one reply may make: tool rounds plus the answer. The last call never offers tools, so the contact always gets an answer. Default 16.
  
  maximum: 32 · minimum: 2
  maxWritesPerStep
  optional
  integer
  Integration tool calls that change data (effect: write) one model step may run. A further one is refused with a fixed reason, and the Agent runs it in its next step after reading the first result. 0 forbids writes. Default 1.
  
  maximum: 3 · minimum: 0
  repeatedReadLimit
  optional
  integer
  Identical reads (same tool, same arguments) one reply may run; the next is refused with a fixed reason. Omitted: no limit, as an Agent that polls a pending operation repeats one read on purpose.
  
  maximum: 10 · minimum: 1
  tools
  optional
  object
  Per-tool rules, by the name the model sees.
  
  null
- pollPendingOperations
  optional
  boolean | null
  Text Agents on WhatsApp: when a reply ends while an integration answered that its work is still pending (status: pending with an operationId) and the same connection offers getOperation, Raíz checks it again on its own and the Agent answers with the result, without waiting for the contact to write. Null: disabled.
- reasoningEffort
  optional
  none
  or
  low
  or
  medium
  or
  high
  or
  xhigh
  or
  null
  Text Agents whose tier reasons (arbolito/advanced, arbolito/top, or an Agent already running a model that reasons): none answers fastest, high deliberates most, and xhigh, on arbolito/advanced and arbolito/top, deliberates past it. Null: the tier's default. A tier that does not reason refuses it (AGENT_SETTINGS_INVALID).
- replyChoices
  optional
  boolean | null
  Text Agents. Let the Agent offer tap-to-answer options (WhatsApp buttons and lists, web reply options). Null: allowed.
- sessionCloseAfterMinutes
  optional
  integer | null
  Text Agents. Close a conversation after this many minutes without messages. Null: 1440 on WhatsApp, 30 on web.
  
  maximum: 10080 · minimum: 5
- temperature
  optional
  number | null
  Text Agents. Sampling temperature of replies. Null: 0.2 (steady answers). arbolito/advanced and arbolito/top set their own sampling and ignore it.
  
  maximum: 1 · minimum: 0
- toolDiscovery
  optional
  boolean | null
  Retired: changes nothing. WhatsApp replies (the WhatsApp Agent Worker) offer the Agent's tools directly. Accepted so existing clients that send it stay valid.
- voice
  optional
  One of the following:
  RaizAgentVoiceSettings
  keywords
  array of
  string
  required
  Names and terms the transcriber should favor (people, places, products).
  
  maxLength: 64 · minLength: 1
  maxDurationSeconds
  integer | null
  required
  Hard limit of one call. Null: 600 seconds.
  
  maximum: 7200 · minimum: 30
  recordCalls
  boolean
  required
  Keep the call recording (copied to Raíz after the call).
  
  silenceEndCallTimeoutSeconds
  integer | null
  required
  Hang up after this much caller silence. Null: the platform default.
  
  maximum: 600 · minimum: 10
  similarityBoost
  number | null
  required
  maximum: 1 · minimum: 0
  speed
  number | null
  required
  maximum: 1.2 · minimum: 0.7
  stability
  number | null
  required
  maximum: 1 · minimum: 0
  turnTimeoutSeconds
  integer | null
  required
  Seconds to wait for the caller before prompting again. Null: the platform default.
  
  maximum: 30 · minimum: 1
  llm
  arbolito/agile
  or
  arbolito/balanced
  or
  arbolito/advanced
  or
  arbolito/top
  or
  null
  required
  The tier the Agent thinks with on calls. arbolito/agile: quickest and least expensive, for direct conversations. arbolito/balanced: more deliberate answers at moderate cost. arbolito/advanced: strong reasoning for complex instructions and many tools, with an adjustable reasoningEffort. arbolito/top: the most capable, for the hardest cases, with an adjustable reasoningEffort up to xhigh. Every tier reads images and PDF documents. Null: the default, as arbolito/agile.
  
  voiceId
  string | null
  required
  A voice from listVoices. Null uses the default voice.
  
  pattern: ^vce_[A-Za-z0-9_-]{20,156}$
  redactKeypadInput
  optional
  boolean
  Replace numbers entered on the phone keypad with <REDACTED> in stored transcripts and analysis. False (the default) keeps them visible. Applies to future calls; the Agent still receives the digits during the call.
  
  default: false
  reasoningEffort
  optional
  none
  or
  minimal
  or
  low
  or
  medium
  or
  high
  or
  xhigh
  +2 more
  Reasoning effort on calls, when supported by the selected model. Null uses the model default.
  
  responseDelayMessages
  optional
  boolean
  Speak a short contextual message while preparing a slow response. False disables automatic delay messages. Omitted preserves the current runtime setting.
  
  toolCallSound
  optional
  typing
  or
  null
  
  typing: Play a typing sound while this Agent's tools execute. Null disables the sound; omitted preserves the current setting. Applies to future calls after synchronization.
  
  backupModels
  optional
  array | null
  Ordered backup tiers when the primary voice model fails. Null restores the runtime default. Omitted preserves the current runtime setting. An empty list is not allowed because calls need a fallback.
  
  null
}
status
optional
active
or
paused
A paused Agent answers nothing and places no calls.
timeZone
optional
string | null
IANA time zone name.

maxLength: 64 · pattern: ^[A-Za-z]+(/[A-Za-z0-9_+-]+)*$
evalRunId
optional
string
With the Agent's settings.evalGate: the completed evaluation run (createEvalRun) of exactly the content this update leaves the model reading, in which every case passed.

pattern: ^evr_(?:[0-7][0-9a-hjkmnp-tv-z]{25}|(?:[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}|[a-f0-9]{32}))$
expectedVersion
optional
integer
The version you read. When present, the change applies only if nobody changed the resource since; otherwise it fails with VERSION_CONFLICT.

minimum: 1
modality
optional
voice
or
text
voice: phone calls. text: WhatsApp and web chat.
name
optional
string
maxLength: 120 · minLength: 1

Returns

200

Successful operation.

JSON

success
boolean
required
data
object
required
- channelCount
  integer
  required
  minimum: 0
- createdAt
  string
  required
  An instant in RFC 3339 format, in UTC.
  
  format: date-time
- description
  string | null
  required
- firstMessage
  string | null
  required
- id
  string
  required
  Agent id.
  
  pattern: ^agt_(?:[0-7][0-9a-hjkmnp-tv-z]{25}|(?:[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}|[a-f0-9]{32}))$
- language
  string
  required
  BCP 47 language tag.
- modality
  voice
  or
  text
  required
  voice: phone calls. text: WhatsApp and web chat.
- name
  string
  required
- ready
  boolean
  required
  Whether the Agent can hold conversations: active, with instructions and, for voice, published for calls.
- status
  active
  or
  paused
  required
  A paused Agent answers nothing and places no calls.
- syncErrorCode
  string | null
  required
  Why calls still use an older version: the error code of the last publication attempt. Null while synced or while a publication is pending. A fix to the Agent (a valid voiceId, instructions) publishes again.
- synced
  boolean
  required
  Whether calls already use the current version (it is published on the Agent's voice channels). Always true for text Agents.
- timeZone
  string | null
  required
  IANA time zone.
- toolCount
  integer
  required
  minimum: 0
- updatedAt
  string
  required
  An instant in RFC 3339 format, in UTC.
  
  format: date-time
- version
  integer
  required
  minimum: 1
- channels
  array of
  object
  required
  Channels assigned to the Agent. Assign one with updateChannel.
  
  address
  string | null
  required
  Phone number or web address of the channel.
  
  channelType
  voice
  or
  whatsapp
  or
  web
  required
  id
  string
  required
  Channel id.
  
  pattern: ^(?:chn_(?:[0-7][0-9a-hjkmnp-tv-z]{25}|(?:[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}|[a-f0-9]{32}))|(?:agc)_(?:[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}|[a-f0-9]{32}))$
  name
  string | null
  required
  status
  pending
  or
  active
  or
  degraded
  required
  lastBusinessAppGreetingAt
  optional
  string | null
  WhatsApp only. Last observed message from the phone app matching the configured greeting. Null when none has been recognized. The phone's current greeting setting is unknown.
  
  format: date-time
- instructions
  string | null
  required
- integrationProcedures
  array of
  object
  required
  The procedures the Agent takes from its connections, each saved with the text its connection published then. They run as free-form procedures beside procedures.
  
  connectionId
  string
  required
  A connection's id in Account (con_…).
  
  maxLength: 255 · minLength: 1 · pattern: ^[A-Za-z0-9_-]+$
  digest
  string
  required
  The digest of the procedure the Agent was saved with. When the connection's procedure (listConnections) has another digest, it changed: the Agent keeps this text until it is saved again.
  
  maxLength: 64 · minLength: 1
  procedure
  string
  required
  The procedure's name, as its connection lists it.
  
  maxLength: 64 · minLength: 1 · pattern: ^[a-z0-9]+(-[a-z0-9]+)*$
- integrationTools
  array of
  object
  required
  The integrations' tools the Agent may call, each on the connection it runs on, with the definition it was saved with.
  
  connectionId
  string
  required
  A connection's id in Account (con_…).
  
  maxLength: 255 · minLength: 1 · pattern: ^[A-Za-z0-9_-]+$
  digest
  string
  required
  The digest of the definition the Agent was saved with. When the connection's tool (listConnections) has another digest, the tool changed: the Agent keeps this definition until it is saved again.
  
  maxLength: 64 · minLength: 1
  effect
  read
  or
  write
  required
  read only looks; write may change data.
  
  tool
  string
  required
  The tool's name, as its connection lists it.
  
  maxLength: 128 · minLength: 1 · pattern: ^[A-Za-z0-9_.-]+$
  toolsVersion
  string
  required
  The connection's tool-list version when the Agent was saved.
- knowledgeDocumentIds
  array of
  string
  required
  The Knowledge documents the Agent consults.
  
  pattern: ^(?:kdc_(?:[0-7][0-9a-hjkmnp-tv-z]{25}|(?:[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}|[a-f0-9]{32}))|(?:con)_(?:[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}|[a-f0-9]{32}))$
- procedures
  array of
  object
  required
  The Agent's procedures. The model reads only the triggers until a conversation matches one, then loads that procedure. In an update, the complete new set.
  
  content
  string
  required
  Free-form procedures: what to do, in Markdown, with inline references such as [tool id="buscar_citas"], [kb id="kdc_…"], [procedure id="prc_…"], [system_tool id="end_conversation"] and {{contact_name}}. Empty for structured procedures.
  
  maxLength: 50000
  id
  string
  required
  The procedure's id (prc_…), stable for the Agent's life.
  
  pattern: ^prc_[0-7][0-9a-hjkmnp-tv-z]{25}$
  name
  string
  required
  A short label for people. The model never reads it.
  
  maxLength: 120 · minLength: 1
  status
  active
  or
  paused
  required
  A paused procedure runs on no channel.
  
  steps
  array of
  required
  Structured procedures: the steps, in order. Empty for free-form procedures.
  
  One of the following:
  object
  object
  object
  object
  object
  object
  object
  trigger
  string
  required
  When it applies, from the contact's side: When the contact asks to book or move an appointment. Empty makes a sub-procedure, which only runs when another procedure references it.
  
  maxLength: 1000
  type
  free_form
  or
  structured
  required
  free_form: instructions the Agent adapts. structured: steps that run the same way every time. Fixed once created.
- settings
  object
  required
  humanEscalation
  object | null
  required
  Per-Agent human escalation policy. Instructions apply to voice and text. Null keeps the default instructions. A WhatsApp handoff stays with the team until a person explicitly resumes AI; no timeout resumes it.
  
  instructions
  string | null
  required
  When this Agent should ask the business's team to intervene, using only its available handoff capabilities. Null keeps the default instructions; an explicit request for a person is always respected.
  
  maxLength: 4000
  recoveryManager
  object | null
  required
  Optional WhatsApp recovery advisor. Null or enabled=false disables it. A journaled evidence check precedes any paid review. One tool-free model call per turn before autonomous escalation or after terminal provider failure; unresolved terminal recovery hands off. Completed handoffs and conservatively recognized human requests bypass generation. Model selects the existing provider route. Prices must be conservative upper bounds including provider overhead; missing or invalid prices disable recovery. The cost bound covers only the advisor. No tools or automatic provider retries.
  
  enabled
  boolean
  required
  model
  arbolito/agile
  or
  arbolito/balanced
  or
  arbolito/advanced
  or
  arbolito/top
  required
  arbolito/agile: quickest and least expensive, for direct conversations. arbolito/balanced: more deliberate answers at moderate cost. arbolito/advanced: strong reasoning for complex instructions and many tools, with an adjustable reasoningEffort. arbolito/top: the most capable, for the hardest cases, with an adjustable reasoningEffort up to xhigh. Every tier reads images and PDF documents.
  
  maxAttempts
  number
  required
  timeoutMs
  integer
  required
  maximum: 15000 · minimum: 100
  maxInputTokens
  integer
  required
  maximum: 131072 · minimum: 4096
  maxOutputTokens
  integer
  required
  maximum: 1024 · minimum: 128
  maxCostUsd
  number
  required
  maximum: 1
  inputUsdPerMillion
  number
  required
  maximum: 1000
  outputUsdPerMillion
  number
  required
  maximum: 1000
  instructions
  optional
  string | null
  Operator criteria for when the recovery advisor should help or leave the case to a person. They never override consent, tool permissions, unknown-write fences or an explicit human request. Null keeps the default criteria.
  
  maxLength: 4000
  advisorAgentId
  optional
  string | null
  Optional active text Agent in the same organization, other than this Agent, whose instructions and procedures provide specialist guidance. The configured recovery model and cost bounds stay unchanged; no specialist tools or conversation history are accessed.
  
  pattern: ^agt_(?:[0-7][0-9a-hjkmnp-tv-z]{25}|(?:[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}|[a-f0-9]{32}))$
  automaticKnowledge
  boolean | null
  required
  Text Agents. Retrieve relevant passages from this Agent's active Knowledge before a substantive contact message, without a model tool call. Null or false: on-demand search only. Does not store contact memories or learn unreviewed answers.
  
  attachmentReader
  vision
  or
  ocr
  or
  null
  required
  Text Agents on WhatsApp. Who reads an image or PDF when the Agent calls read_attachment: vision (a vision model describes and transcribes it, with a stronger one when the first reads nothing) or ocr (text recognition, which also answers the fields the Agent asks for, each null when absent or unclear; the vision model reads the file when recognition cannot). With ocr the Agent sees files as markers and reads them with read_attachment; with vision, tiers that read files themselves see the original. Null: vision.
  
  budgetExhaustedMessage
  string | null
  required
  Text Agents on WhatsApp. What the contact reads when the Agent has used its monthlyTokenBudget: the conversation then goes to the team, as an escalation does, and the Agent writes nothing until the budget is raised or the month ends. Null: handoffMessage, including an empty notice when disabled.
  
  maxLength: 1000 · minLength: 1
  closingMessage
  string | null
  required
  Text Agents on WhatsApp. After a reply that completed an action (an integration tool that changes data succeeded), explicitly finished the case with end_conversation and left nothing pending (no question, no offered choices, no handoff), a contact message that only thanks or closes («gracias», «ok», «listo», an emoji) gets this text at once, without the model; the Agent still reads both in its next turn. Write it in the Agent's language and ask nothing in it. Null: the Agent answers such messages itself.
  
  maxLength: 1000 · minLength: 1
  evalGate
  boolean | null
  required
  Text and voice Agents. When true, an updateAgent that changes what the model reads (instructions, procedures, connection procedures, model or reasoningEffort) must send evalRunId: a completed createEvalRun of this Agent with exactly that content, in which every case passed (EVAL_GATE_UNMET otherwise). Null: no requirement.
  
  followUpAfterMinutes
  integer | null
  required
  Text Agents. When the Agent spoke last and asked something, send followUpMessage once after this many minutes of silence (WhatsApp only, inside its 24-hour window). Null disables follow-ups.
  
  maximum: 1440 · minimum: 1
  followUpMessage
  string | null
  required
  Text Agents. The follow-up text. Null uses a short default in the Agent's language.
  
  maxLength: 1000 · minLength: 1
  greeting
  required
  One of the following:
  RaizAgentGreeting
  null
  handoffMessage
  string | null
  required
  Text Agents on WhatsApp. The notice appended after an escalation or a requested human handoff. Null: the current localized review notice or the case-specific notice supplied under the Agent's instructions. An empty string disables the notice; the handoff and pause still happen. The model does not write this notice.
  
  maxLength: 1000
  humanReplyTimeoutMinutes
  integer | null
  required
  Text Agents on WhatsApp. After a manual takeover, resume automatic replies when the contact's newest message waits this many minutes. An Agent handoff always stays with the team until a person explicitly resumes AI. Null: only a person resumes automatic replies.
  
  maximum: 1440 · minimum: 1
  investigationsEnabled
  boolean | null
  required
  Opt this Agent into internal root-cause investigations of tool failures, human escalations, failed outcomes and negative feedback. Incidents are grouped only with participating Agents in the same organization. Null or false: disabled. Turning this off stops new events and pending dispatch when no participating Agent remains; an already active investigation may finish.
  
  model
  arbolito/agile
  or
  arbolito/balanced
  or
  arbolito/advanced
  or
  arbolito/top
  or
  null
  required
  Text Agents. The tier that writes replies. arbolito/agile: quickest and least expensive, for direct conversations. arbolito/balanced: more deliberate answers at moderate cost. arbolito/advanced: strong reasoning for complex instructions and many tools, with an adjustable reasoningEffort. arbolito/top: the most capable, for the hardest cases, with an adjustable reasoningEffort up to xhigh. Every tier reads images and PDF documents. Null: the default, as arbolito/agile.
  
  modelTrial
  required
  One of the following:
  RaizAgentModelTrial
  model
  arbolito/agile
  or
  arbolito/balanced
  or
  arbolito/advanced
  or
  arbolito/top
  required
  The tier the trial's conversations are answered with.
  
  percent
  integer
  required
  The share of new conversations in the trial, in percent.
  
  maximum: 50 · minimum: 1
  reasoningEffort
  none
  or
  low
  or
  medium
  or
  high
  or
  xhigh
  or
  null
  required
  The trial tier's reasoning, as reasoningEffort. Null: the tier's default.
  
  null
  monthlyTokenBudget
  integer | null
  required
  Text Agents. Tokens (input, cached input included, plus output) the Agent may use per calendar month (UTC), on every channel. On WhatsApp, a reply that would start past it does not run: the contact reads budgetExhaustedMessage and the team continues; at 80 % the API logs an alert once a month. Null: no limit.
  
  maximum: 1000000000000 · minimum: 10000
  pauseOnHumanReply
  boolean | null
  required
  Text Agents on WhatsApp: pause automatic replies when a live WhatsApp Business App message that says something (not a reaction, an edit or a deletion) proves a person took over. Null: disabled. The recipient stays paused until resumed or until that conversation closes; their next conversation starts with the AI.
  
  policy
  required
  One of the following:
  RaizAgentTurnPolicy
  maxOutputTokens
  optional
  integer
  Output tokens per model call, reasoning included. Default 4096.
  
  maximum: 16384 · minimum: 256
  maxSteps
  optional
  integer
  Model calls one reply may make: tool rounds plus the answer. The last call never offers tools, so the contact always gets an answer. Default 16.
  
  maximum: 32 · minimum: 2
  maxWritesPerStep
  optional
  integer
  Integration tool calls that change data (effect: write) one model step may run. A further one is refused with a fixed reason, and the Agent runs it in its next step after reading the first result. 0 forbids writes. Default 1.
  
  maximum: 3 · minimum: 0
  repeatedReadLimit
  optional
  integer
  Identical reads (same tool, same arguments) one reply may run; the next is refused with a fixed reason. Omitted: no limit, as an Agent that polls a pending operation repeats one read on purpose.
  
  maximum: 10 · minimum: 1
  tools
  optional
  object
  Per-tool rules, by the name the model sees.
  
  null
  pollPendingOperations
  boolean | null
  required
  Text Agents on WhatsApp: when a reply ends while an integration answered that its work is still pending (status: pending with an operationId) and the same connection offers getOperation, Raíz checks it again on its own and the Agent answers with the result, without waiting for the contact to write. Null: disabled.
  
  reasoningEffort
  none
  or
  low
  or
  medium
  or
  high
  or
  xhigh
  or
  null
  required
  Text Agents whose tier reasons (arbolito/advanced, arbolito/top, or an Agent already running a model that reasons): none answers fastest, high deliberates most, and xhigh, on arbolito/advanced and arbolito/top, deliberates past it. Null: the tier's default. A tier that does not reason refuses it (AGENT_SETTINGS_INVALID).
  
  replyChoices
  boolean | null
  required
  Text Agents. Let the Agent offer tap-to-answer options (WhatsApp buttons and lists, web reply options). Null: allowed.
  
  sessionCloseAfterMinutes
  integer | null
  required
  Text Agents. Close a conversation after this many minutes without messages. Null: 1440 on WhatsApp, 30 on web.
  
  maximum: 10080 · minimum: 5
  temperature
  number | null
  required
  Text Agents. Sampling temperature of replies. Null: 0.2 (steady answers). arbolito/advanced and arbolito/top set their own sampling and ignore it.
  
  maximum: 1 · minimum: 0
  toolDiscovery
  boolean | null
  required
  Retired: changes nothing. WhatsApp replies (the WhatsApp Agent Worker) offer the Agent's tools directly. Accepted so existing clients that send it stay valid.
  
  voice
  required
  One of the following:
  RaizAgentVoiceSettings
  keywords
  array of
  string
  required
  Names and terms the transcriber should favor (people, places, products).
  
  maxLength: 64 · minLength: 1
  maxDurationSeconds
  integer | null
  required
  Hard limit of one call. Null: 600 seconds.
  
  maximum: 7200 · minimum: 30
  recordCalls
  boolean
  required
  Keep the call recording (copied to Raíz after the call).
  
  silenceEndCallTimeoutSeconds
  integer | null
  required
  Hang up after this much caller silence. Null: the platform default.
  
  maximum: 600 · minimum: 10
  similarityBoost
  number | null
  required
  maximum: 1 · minimum: 0
  speed
  number | null
  required
  maximum: 1.2 · minimum: 0.7
  stability
  number | null
  required
  maximum: 1 · minimum: 0
  turnTimeoutSeconds
  integer | null
  required
  Seconds to wait for the caller before prompting again. Null: the platform default.
  
  maximum: 30 · minimum: 1
  llm
  arbolito/agile
  or
  arbolito/balanced
  or
  arbolito/advanced
  or
  arbolito/top
  or
  null
  required
  The tier the Agent thinks with on calls. arbolito/agile: quickest and least expensive, for direct conversations. arbolito/balanced: more deliberate answers at moderate cost. arbolito/advanced: strong reasoning for complex instructions and many tools, with an adjustable reasoningEffort. arbolito/top: the most capable, for the hardest cases, with an adjustable reasoningEffort up to xhigh. Every tier reads images and PDF documents. Null: the default, as arbolito/agile.
  
  voiceId
  string | null
  required
  A voice from listVoices. Null uses the default voice.
  
  pattern: ^vce_[A-Za-z0-9_-]{20,156}$
  redactKeypadInput
  optional
  boolean
  Replace numbers entered on the phone keypad with <REDACTED> in stored transcripts and analysis. False (the default) keeps them visible. Applies to future calls; the Agent still receives the digits during the call.
  
  default: false
  reasoningEffort
  optional
  none
  or
  minimal
  or
  low
  or
  medium
  or
  high
  or
  xhigh
  +2 more
  Reasoning effort on calls, when supported by the selected model. Null uses the model default.
  
  responseDelayMessages
  optional
  boolean
  Speak a short contextual message while preparing a slow response. False disables automatic delay messages. Omitted preserves the current runtime setting.
  
  toolCallSound
  optional
  typing
  or
  null
  
  typing: Play a typing sound while this Agent's tools execute. Null disables the sound; omitted preserves the current setting. Applies to future calls after synchronization.
  
  backupModels
  optional
  array | null
  Ordered backup tiers when the primary voice model fails. Null restores the runtime default. Omitted preserves the current runtime setting. An empty list is not allowed because calls need a fallback.
  
  null
}
error
null
required
metadata
object
required
- requestId
  string
  required
  Request correlation identifier, also sent as X-Request-Id.
}

400

INVALID_INPUT: The request does not match the published schema.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

401

UNAUTHORIZED: A valid access token is required.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

403

FORBIDDEN: The token does not authorize this operation.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

404

NOT_FOUND: The requested operation does not exist.
RESOURCE_NOT_FOUND: The requested resource does not exist in this organization.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

405

HTTP_METHOD_NOT_ALLOWED: The HTTP method is not supported for this resource.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

409

AGENT_IN_USE: The Agent still has channels or active campaigns. Unassign them before deleting it.
AGENT_MODALITY_LOCKED: The Agent's modality cannot change while it has channels or is published for calls. Create a new Agent instead.
EVAL_GATE_UNMET: This Agent requires a completed evaluation run of exactly this content, in which every case passed, before what its model reads changes. Create one with createEvalRun and send its id as evalRunId.
VERSION_CONFLICT: The resource changed after it was read. Read it again and retry with the new version.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

413

REQUEST_TOO_LARGE: The request exceeds the allowed size.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

415

UNSUPPORTED_MEDIA_TYPE: The request content type is not supported.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

422

AGENT_PROCEDURES_INVALID: A procedure is incomplete, references a tool, document or procedure the Agent does not have, or breaks the structured step rules. issues says where.
AGENT_SETTINGS_INVALID: The settings do not apply to this Agent's modality, or two settings contradict each other.
INTEGRATION_PROCEDURE_UNAVAILABLE: A procedure does not exist in this organization's connections, Raíz cannot run its connection, or the Agent is not a text Agent.
INTEGRATION_TOOL_UNAVAILABLE: An integration tool does not exist in this organization's connections, or Raíz cannot run it.
KNOWLEDGE_DOCUMENT_UNAVAILABLE: A Knowledge document does not exist in this organization.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

429

RATE_LIMITED: The request rate limit was exceeded. Try again later.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

500

INTERNAL_ERROR: The service could not complete the operation.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

502

OUTCOME_UNKNOWN: The operation may have been applied; check its result before repeating it.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

503

UPSTREAM_UNAVAILABLE: The upstream service could not complete the request.
VOICE_UNAVAILABLE: Calls are unavailable right now. Try again in a few minutes.
INTEGRATIONS_UNAVAILABLE: The organization's integrations cannot be read right now. Try again shortly.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}

504

UPSTREAM_TIMEOUT: The upstream service did not respond before its deadline.

JSON

error
required
One of the following:
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
object
}