diff --git a/fern/assistants/call-recording.mdx b/fern/assistants/call-recording.mdx index fb285fa84..50f87ee90 100644 --- a/fern/assistants/call-recording.mdx +++ b/fern/assistants/call-recording.mdx @@ -205,6 +205,39 @@ Configure different types of artifacts with these options: - **`pcapEnabled`**: Enable SIP packet capture for phone calls (stored in `call.artifact.pcapUrl`) Default: `true` - **`transcriptPlan`**: Configure transcript generation with speaker names +For separate recording in your own Twilio account, see [Twilio outbound transport settings](/phone-numbers/import-twilio#outbound-transport-settings). + +## Video recording + +Set `artifactPlan.videoRecordingEnabled` to `true`, or enable **Video Recording** in the assistant's Advanced settings. It defaults to `false` and applies only to web calls using Daily, not phone or Vapi WebSocket calls. After the call ends, find the video URL at `call.artifact.videoRecordingUrl`. + +Video recording is disabled if the organization or any participating assistant has Zero Data Retention (ZDR) or PCI enabled. Daily video recordings use Vapi's storage, so adding custom storage does not enable video recording for those calls. + +### Choose who starts recording + +Set `videoRecordingStartMode` at the top level of the `POST /call/web` creation request, not inside `artifactPlan`. It only applies when video recording is enabled: + +| Mode | Behavior | +| --- | --- | +| `client` (default) | Recording starts when the web client joins. The returned meeting token can start it automatically; older clients can use client-initiated recording. | +| `server` | Vapi starts recording once when the user first joins, so starting does not depend on the client's connection. The returned `transport.callToken` does not start recording, and your client must not start it separately. | + +For example, create a web call using a stored assistant and enable server-started recording for that call: + +```json +{ + "assistantId": "ASSISTANT_ID", + "assistantOverrides": { + "artifactPlan": { + "videoRecordingEnabled": true + } + }, + "videoRecordingStartMode": "server" +} +``` + +In `server` mode, if Vapi cannot set up recording, the response has no `transport.callToken`. If starting recording fails during the call, the client is notified. There is no automatic fallback to a client start. + ## Storage Configuration diff --git a/fern/assistants/model-intelligence/understanding-cost.mdx b/fern/assistants/model-intelligence/understanding-cost.mdx index 3bf4956ed..a030525bf 100644 --- a/fern/assistants/model-intelligence/understanding-cost.mdx +++ b/fern/assistants/model-intelligence/understanding-cost.mdx @@ -61,6 +61,8 @@ Vapi estimates 5 requests per turn. We use number of requests rather than conver **Effective input rate** accounts for prompt caching. Cached input is stable prompt content that providers reuse across requests. Many providers offer a discounted rate for this input. +For OpenAI configuration options, see [Prompt caching](/providers/model/openai#prompt-caching). + When a model supports caching, Vapi blends its standard and cached rates based on an assumed 50% cache hit rate. Otherwise, Vapi uses the standard rate. The calculation uses the actual rates for each model and provider. #### Output tokens per minute diff --git a/fern/phone-numbers/import-twilio.mdx b/fern/phone-numbers/import-twilio.mdx index eb00c0067..347e0b910 100644 --- a/fern/phone-numbers/import-twilio.mdx +++ b/fern/phone-numbers/import-twilio.mdx @@ -60,3 +60,28 @@ As you scale your agents, you may want to use other telephony providers, like Tw If outbound calls fail before the phone rings, see [Troubleshoot call connection errors](/calls/troubleshoot-call-connection-errors#outbound-calls). + +## Outbound transport settings + +Set `transportConfigurations` on the assistant through the API to configure Twilio's outbound calls and warm-transfer legs. These settings have no Dashboard control. + +```json +{ + "transportConfigurations": [ + { + "provider": "twilio", + "timeout": 30, + "record": true, + "recordingChannels": "dual" + } + ] +} +``` + +| Field | Default | Behavior | +| --- | --- | --- | +| `timeout` | `60` | Seconds to let the destination ring before treating the call as unanswered. This is not a conversation-duration or silence timeout. Some Twilio call flows can add about five seconds. | +| `record` | `false` | Enables recording in your Twilio account. Requires your own Twilio credentials on the phone number; Vapi forces this off otherwise, including for Vapi-managed numbers without your credentials. | +| `recordingChannels` | `mono` | `mono` combines both call legs in one channel. `dual` puts the parent call in the first channel and the child call in the second. Applies when Twilio recording is enabled. | + +Twilio-side recording is separate from Vapi's [`artifactPlan` recording](/assistants/call-recording). Configure `artifactPlan` to control Vapi's recordings; these transport settings control recordings in Twilio. diff --git a/fern/providers/model/openai.mdx b/fern/providers/model/openai.mdx index 3b4a8371d..8b0cf1a46 100644 --- a/fern/providers/model/openai.mdx +++ b/fern/providers/model/openai.mdx @@ -82,6 +82,30 @@ If you've added an [Azure OpenAI credential](/providers/model/azure-openai) for | GPT 5 Mini | `gpt-5-mini:eastus2`
`gpt-5-mini:germanywestcentral`
`gpt-5-mini:polandcentral`
`gpt-5-mini:spaincentral`
`gpt-5-mini:swedencentral`
`gpt-5-mini:westeurope` | | GPT 5 Nano | `gpt-5-nano:eastus2`
`gpt-5-nano:swedencentral` | +## Prompt caching + +Prompt caching lets OpenAI reuse the unchanged beginning of a prompt, such as your assistant's instructions and tool definitions, instead of processing it again on every request. Cache hits can reduce input-token costs and latency, but are not guaranteed. + +For supported GPT-4.1 and GPT-5.x models, open the assistant's model **Advanced settings** in the Dashboard to configure **Prompt cache retention** and **Prompt cache key**. Check the Dashboard or the [`OpenAIModel` fields](/api-reference/assistants/create#request.body.model.OpenAI) for the current supported model list. + +- **`promptCacheKey`**: An optional hint to help OpenAI reuse matching prompt prefixes across requests. It does not store a prompt or guarantee a cache hit. In the standard assistant model configuration, Vapi leaves this unset unless you provide it; it does not generate a key automatically. +- **`promptCacheRetention`**: Choose `in_memory` for OpenAI's in-memory cache policy or `24h` to keep cached prefixes available for up to 24 hours on models that support extended caching. + +Both settings are unset by default. For these models, omitting retention uses OpenAI's default `in_memory` policy; leaving the key unset does not disable automatic prompt caching. + +For example, include these fields in your assistant's model configuration: + +```json +{ + "model": { + "provider": "openai", + "model": "gpt-4.1", + "promptCacheKey": "support-assistant-v1", + "promptCacheRetention": "24h" + } +} +``` + ## Reasoning effort Reasoning models think before they answer. Set `model.reasoningEffort` to control how much. A lower effort reduces latency and cost, which suits most voice calls. A higher effort can help with harder tasks. diff --git a/fern/providers/transcriber/soniox.mdx b/fern/providers/transcriber/soniox.mdx index ee6088932..854895b53 100644 --- a/fern/providers/transcriber/soniox.mdx +++ b/fern/providers/transcriber/soniox.mdx @@ -108,6 +108,32 @@ The following languages are currently accepted by Vapi for Soniox. See [Soniox s | Vietnamese | `vi` | | Welsh | `cy` | +## End-of-turn detection + +For `stt-rt-v5` and `stt-rt-v6`, you can tune when Soniox decides the caller has finished speaking. These settings do not apply to `stt-rt-v4`. + +In the Dashboard, open the assistant's transcriber **Advanced settings** to adjust **Endpoint sensitivity** and **Endpoint latency level**. Through the API, set: + +| Field | Default | Behavior | +| --- | --- | --- | +| `endpointSensitivity` | `0.3` | Higher values make Soniox end turns sooner. Negative values make it more patient with pauses mid-sentence. | +| `endpointLatencyAdjustmentLevel` | `2` | Controls how aggressively Soniox reduces the wait before ending a turn, from `0` (most patient) to `3` (most aggressive). Higher levels can split speech into more segments. | + +These are Vapi's defaults, not Soniox's own defaults. For callers reading long numbers or pausing between parts of a name, lower the latency level first. If turns still end too early, lower sensitivity. Soniox advises against negative sensitivity with a latency level above `0`, because the settings work against each other. + +For example, this assistant waits more patiently for pauses: + +```json +{ + "transcriber": { + "provider": "soniox", + "model": "stt-rt-v6", + "endpointSensitivity": -0.3, + "endpointLatencyAdjustmentLevel": 0 + } +} +``` + ## Low-confidence filter You can discard transcripts that Soniox returns with low confidence, so the assistant does not respond to uncertain speech. In the dashboard, open the transcriber **Advanced settings** and turn on the **Low-confidence filter** to set a threshold. Via the API, set `transcriber.confidenceThreshold` to a value between `0` and `1`.