# Dialt

> Voice assistants, dialed in. Dialt is a conversational voice engine. Tune how your voice assistants sound, speak and act. Model training on your data currently requires working with our team.

## Start here

- [Developer documentation](https://dialt.com/docs/api/): SDKs, recipes and API references.
- [Get an API key](https://dialt.com/account#api-keys): Keep persistent keys on your backend; issue scoped session credentials to browsers and native apps.
- [Talk to us about training](mailto:hello@dialt.com?subject=Training%20with%20Dialt) to train a model on your data.

## API contracts

Use these as the source of truth for requests, events, defaults and errors. SDK guides describe language-specific behavior; recipes compose the SDKs into applications.

- [HTTP reference](https://dialt.com/docs/api/reference/http/): Credentials, voices, sessions and recording segments.
- [OpenAPI JSON](https://dialt.com/docs/api/openapi.json): Machine-readable HTTP contract.
- [Realtime reference](https://dialt.com/docs/api/reference/realtime/): Session configuration, control messages, tools and events.
- [AsyncAPI JSON](https://dialt.com/docs/api/asyncapi.json): Machine-readable realtime contract.
- [WebSocket guide](https://dialt.com/docs/api/websocket/): Authentication, audio, playback obligations and connection lifecycle.
- [Background tools](https://dialt.com/docs/api/background-tools/): Deferral, progress, decisions, cancellation and completion.

## SDKs

- [TypeScript / JavaScript SDK](https://dialt.com/docs/api/browser/): Install with `npm install @dialt/sdk`; includes browser media, headless Node.js realtime and a generated HTTP client.
- [Python SDK](https://dialt.com/docs/api/python/): Install with `uv add dialt-sdk`; your application owns capture and playback.
- [Swift SDK](https://github.com/dialt-ai/dialt-swift): Experimental native macOS/iOS client; install through Swift Package Manager. Read its validation notes before choosing a device integration.
- [Pipecat plugin](https://github.com/dialt-ai/pipecat-dialt): Dialt-maintained alpha integration; setup instructions, a minimal pipeline and supported features and limitations.

## Recipes

- [Recipe repository and setup](https://github.com/dialt-ai/dialt-recipes): Runnable applications using the public SDKs.
- [Agent handoff](https://github.com/dialt-ai/dialt-recipes/tree/main/examples/agent_to_agent_handoff): Python intake-to-specialist workflow.
- [Policy guidance](https://github.com/dialt-ai/dialt-recipes/tree/main/examples/policy_agent): Python session policies and guidance events.
- [Guided research](https://github.com/dialt-ai/dialt-recipes/tree/main/examples/guided_customer_research): Python and browser interview examples.
- [Twilio guide](https://dialt.com/docs/api/twilio/): Inbound telephony integration.
- [Twilio recipe](https://github.com/dialt-ai/dialt-recipes/tree/main/examples/integrations/twilio): Maintained Python Media Streams bridge.

## Evaluate and improve

- [Sessions](https://dialt.com/sessions): Inspect your transcripts, recordings and tool calls; requires an account.
- [Conversation evals](https://dialt.com/docs/api/evals/): Case structure, checks, text/voice runs and result interpretation.
- [Eval-building skill](https://dialt.com/docs/skills/dialt-evals/SKILL.md): Create a suite, validate cases with Dialt Smart in text, then evaluate in voice using the native dashboard.
- [Python evals client](https://dialt.com/docs/api/python/#python-evals): Manage hosted cases and runs.
- [Simulation recipes](https://github.com/dialt-ai/dialt-recipes/tree/main/examples/simulations): Runnable conversation cases.

## Migration

- [Converse to Dialt](https://dialt.com/docs/api/migration/): Package, class, credential and endpoint changes.

## Blog

- [Blog](https://dialt.com/blog/): Announcements, benchmark results and guides ([RSS](https://dialt.com/blog/feed.xml)).
- [Eleven Realtime joins our voice benchmark](https://dialt.com/blog/eleven-realtime/) (2026-10-21): Mini and Max results on ten cross-customer voice scenarios, with a review of tool use and response times.
- [Voice Isolation for Voice Agents](https://dialt.com/blog/voice-isolation/) (2026-10-19): Background speaker suppression for competing speech from television, radio and announcements, with optional agent control during a call.
- [Language Switching with Voice Agents](https://dialt.com/blog/language-switching-voice-agents/) (2026-10-15): Two ways to switch languages mid-call: one multilingual voice or a handoff to a voice and prompt for the caller's language, with a video demo and recipe.
- [Interactive Voice Design](https://dialt.com/blog/interactive-voice-design/) (2026-10-13): You can now design a virtual assistant by talking to it and adjust how it speaks, what it says, and its character in real time.
- [More Consistent Response Times for Dialt Fast and Smart](https://dialt.com/blog/more-consistent-response-times/) (2026-10-05): Lower p90 response times provide a more consistent caller experience while preserving benchmark-leading accuracy.
- [How to Build Evals for your Voice Agent](https://dialt.com/blog/build-voice-agent-evals/) (2026-10-01): Build an eval suite for your own voice agent with one prompt to a coding agent and the Dialt Evals skill, then compare models on it, with a video walkthrough.
- [3 Tips for Better Voice Agent Evals](https://dialt.com/blog/three-tips-voice-agent-evals/) (2026-09-30): Test in text mode first, use a strong simulated caller, and use a strong judge with criteria and a rubric.
- [Dialt Preview: Fast and Smart Real-Time Voice Models](https://dialt.com/blog/dialt-preview/) (2026-09-28): Dialt Fast and Dialt Smart against Gemini 3.8 Live and GPT Live-1 on our internal benchmark, with the method.