Get a call

GET /api/v1/calls/{callId}

Full detail for one call — everything zoxaAI persists about that call's lifecycle. The endpoint resolves recording storage keys to signed URLs at read time so you can play audio directly from the response.

Authentication

X-API-Key: zsk_...

Path parameters

Param	Type	Description
`callId`	string	The `callId` returned by `POST /call` (e.g. `call_4e4e571f8c9b3cf8e99d`).

Response

Returns HTTP 200 OK. The full CallDetail shape:

{
  "callId": "call_4e4e571f8c9b3cf8e99d",
  "transport": "outbound",
  "pipelineSource": "agent",
  "status": "completed",
  "connectionStatus": "completed",
  "endedReason": "user_hangup",

  "agentId": "550e8400-...",
  "agentName": "Sales bot",
  "campaignId": null,
  "campaignName": null,

  "platformNumber": "+14155550100",
  "customerNumber": "+14155559999",
  "telephonyProvider": "twilio",
  "providerCallSid": "CAxxxxxxx",
  "callSource": "api_outbound",
  "telephonySnapshot": { "provider": "twilio", "fromNumber": "+14155550100", "toNumber": "+14155559999", "credentialsFingerprint": "a1b2c3" },
  "bindingId": null,

  "requestSnapshot": {
    "type": "outbound",
    "callConfig": { "provider": "twilio", "phoneNumber": "+14155550100" },
    "toNumber": "+14155559999",
    "agentId": "550e8400-...",
    "contextVariables": { "customerName": "Aman" }
  },

  "startedAt": "2026-06-16T10:30:12Z",
  "endedAt":   "2026-06-16T10:32:01Z",
  "durationSeconds": 109,
  "createdAt": "2026-06-16T10:30:10Z",
  "updatedAt": "2026-06-16T10:32:01Z",

  "cost": 0.0185,
  "costBreakdown": { "llm": 0.012, "tts": 0.005, "stt": 0.0015 },

  "configSnapshot": { "name": "Sales bot", "systemPrompt": "...", "llm": { "provider": "openai", "model": "gpt-5.4-mini" } },
  "contextVariables": { "customerName": "Aman" },

  "recordings": [
    { "url": "https://signed-url...", "format": "wav", "storage_backend": "minio" }
  ],
  "transcript": { "url": "https://signed-url..." },

  "logs":        { "telephony_status_callbacks": [/*...*/] },
  "usageInfo":   { "llm_tokens_in": 320, "llm_tokens_out": 410, "tts_chars": 1820, "stt_seconds": 47 },
  "analysis":    { "summary": "User confirmed appointment for next Tuesday." },
  "latencySummary": { "userToBotMs_p50": 850, "userToBotMs_p95": 1320 },

  "tags": ["api", "outbound"],

  "turns":  [/* per-turn rows, see CallTurn below */],
  "events": [/* fine-grained pipeline events, see CallEvent below */]
}

Top-level fields

Same as list item plus:

Field	Type	Description
`connectionStatus`	string \| `null`	Did the call connect — `completed`, or the reason it never did. See `connectionStatus` values.
`endedReason`	string \| `null`	Why a connected call ended (`user_hangup`, `agent_hangup`, `silence_timeout`, `max_duration`, `voicemail`, `transferred`, `pipeline_error`, `cancelled`, `unknown`). `null` when the call never connected. See `endedReason` values.
`providerCallSid`	string \| `null`	Provider's own call id (Twilio `CallSid`, Vobiz `call_uuid`). Use it to cross-reference in the provider console.
`telephonySnapshot`	object \| `null`	Provider/number/fingerprint metadata. Credentials are stripped server-side — only the fingerprint remains.
`bindingId`	int \| `null`	Set on inbound API-first calls that matched an inbound binding.
`requestSnapshot`	object \| `null`	The original API request body (auth removed). For dashboard/WebRTC calls, a composed view of `agent` + `contextVariables` + provider info. The unified "Config Snapshot" panel in the dashboard reads this.
`configSnapshot`	object	The resolved Agent Config that drove the call — prompt, greeting, llm/stt/tts, tools, call guards, etc. Provider auth is NOT included (telephony credentials live separately on `telephonySnapshot` and are redacted; LLM/TTS/STT API keys are resolved server-side and never serialized into the snapshot).
`contextVariables`	object \| `null`	Variables substituted into the snapshot.
`recordings`	array \| `null`	One entry per recording. `url` is a signed CDN URL valid for ~1 hour.
`transcript`	object \| `null`	`{ url: signed-url }`. Format depends on storage backend; JSONL is standard.
`logs`	object \| `null`	Pipeline-level logs (telephony callbacks, etc.).
`usageInfo`	object \| `null`	Token counts, TTS chars, STT seconds.
`analysis`	object \| `null`	Post-call analysis when enabled. Includes `summary` and any custom extraction.
`latencySummary`	object \| `null`	Aggregate latency stats across all turns.
`turns`	array	Per-turn latency + token breakdown. See below.
`events`	array	Fine-grained pipeline events (turn start/end, interruptions, tool calls). See below.

`turns[]` — per-turn breakdown

One row per conversation turn, ordered by turnNo.

Field	Type	Description
`turnNo`	int	Turn number, starting at 1.
`startedAt` / `endedAt`	ISO 8601 \| `null`	Turn boundaries.
`durationMs`	int \| `null`	End − start.
`wasInterrupted`	bool	True if the user spoke over the agent.
`userToBotMs`	int \| `null`	Round-trip latency — user stops speaking → bot first audio.
`llmTtfbMs`	int \| `null`	Time-to-first-byte from the LLM.
`ttsTtfbMs`	int \| `null`	TTFB from the TTS provider.
`sttTtfbMs`	int \| `null`	TTFB from the STT provider on this turn.
`textAggMs`	int \| `null`	Time spent aggregating tokens into a sentence before TTS.
`tokensIn` / `tokensOut`	int \| `null`	LLM tokens this turn.
`tokensCacheRead` / `tokensCacheCreate` / `tokensReasoning`	int \| `null`	Detailed token accounting where the provider reports it.
`ttsCharsInput`	int \| `null`	Characters sent to TTS this turn.
`botSpeakingMs`	int \| `null`	How long the bot's TTS played.
`llmModel` / `ttsVoice` / `sttModel`	string \| `null`	Models used (lets you compare cost/latency across providers).
`toolCallCount`	int	Function calls invoked this turn.
`toolCallTotalMs`	int	Total time spent inside tools.
`payload`	object	Raw turn metadata (transcript snippet, etc.).

`events[]` — fine-grained pipeline events

Ordered by tMs (milliseconds since call start). Use for visualizing the timeline.

Field	Type	Description
`tMs`	int	Event timestamp in ms since call start.
`kind`	string	Event type. Examples: `turn_start`, `turn_end`, `interruption`, `tool_call_start`, `tool_call_end`, `error`.
`processor`	string \| `null`	Which pipeline processor emitted it.
`turnNo`	int \| `null`	Associated turn, if any.
`payload`	object	Event-specific data.

Examples

curl https://dashboard.zoxa.ai/api/v1/calls/call_4e4e571f8c9b3cf8e99d \
  -H "X-API-Key: zsk_..."

const res = await fetch(
  `https://dashboard.zoxa.ai/api/v1/calls/${callId}`,
  { headers: { "X-API-Key": "zsk_..." } },
);
const call = await res.json();
console.log(call.connectionStatus, call.endedReason, call.durationSeconds, call.cost);
if (call.recordings?.[0]?.url) {
  // signed URL valid ~1h
  playAudio(call.recordings[0].url);
}

import httpx

call = httpx.get(
    f"https://dashboard.zoxa.ai/api/v1/calls/{call_id}",
    headers={"X-API-Key": "zsk_..."},
).json()

print(call["connectionStatus"], call["endedReason"], call["durationSeconds"], call["cost"])
for turn in call["turns"]:
    print(turn["turnNo"], "user→bot:", turn["userToBotMs"], "ms")

Errors

Status	`detail`	When
`400`	`"No organization selected"`	Auth missing org context.
`401`	auth errors	See Errors.
`404`	`"Call not found"`	`callId` doesn't exist in your organization, or belongs to another org.

{ "detail": "Call not found" }

404 covers both "id doesn't exist" and "exists in a different org" — your code treats them the same.

Notes on behavior

Signed URLs expire. Recording / transcript URLs are signed at read-time and expire within ~1 hour. Fetch fresh URLs by re-calling this endpoint.
Snapshot redaction is unconditional. telephonySnapshot.credentials and requestSnapshot.callConfig.auth are stripped server-side before serialization — even in debug mode, you'll never see raw provider tokens here.
turns and events can be empty for calls without per-turn observability data.

GET /calls — list with filters
WebSocket protocol — live-call alternative to fetching history
Webhook events — push delivery of the same lifecycle data

On this page