
Everything your
voice agents need.
Natural conversations, mid-call actions, built-in telephony, analytics, and a visual agent builder.
Voice & AI Models
Three engines, 20+ language models, 41 voices and 32+ languages, all configured from the dashboard.
Call Experience
Every call aspect is configurable — greetings, noise suppression, mid-call actions, and transfers.
Knowledge & Intelligence
Knowledge base, call analytics, and automatic post-call analysis.
Channels & Reach
Web widget, phone numbers, SIP trunking, and outbound calling API.
Agent Builder
Build, test, and publish agents entirely from the dashboard.
Platform
Teams, security, admin tools, and a full developer platform.
Voice & AI Models
Cascading engine
Speech-to-text, a language model and text-to-speech as separate stages. Pick every one.
- LLM, OpenAI: GPT 4.1, 4.1 Mini, 4.1 Nano, 4o, 4o Mini, 5, 5 Mini, 5 Nano, 5.1, 5.2, 5.4, 5.4 Mini, 5.4 Nano, 5.5, 5.6 Sol, 5.6 Terra, 5.6 Luna
- LLM, Anthropic: Claude Sonnet 5, Sonnet 4.6, Opus 4.8, Haiku 4.5. Google: Gemini 3.6 Flash, 3.5 Flash-Lite
- STT: Deepgram Flux (English and Multilingual), Nova 3, Nova 2, and xAI Grok STT, with language per agent
- TTS: OpenAI, ElevenLabs or Grok voices. Mix and match providers
Speech-to-speech engine
One realtime model hears the caller and answers in audio, with no transcription step.
- OpenAI: GPT Realtime, Realtime Mini, Realtime 2, Realtime 2.1, Realtime 2.1 Mini, with semantic VAD and eagerness levels
- GPT-Live-1
- Grok Voice Agent: 5 voices (Ara, Rex, Sal, Eve, Leo), 40+ languages, server-side VAD
Full duplex engine
Both sides talk simultaneously with graceful interruption handling.
- Real-time turn-taking with configurable interruption sensitivity
- Minimum word threshold before allowing caller interruptions
- Turn detection with 23-language support and ~12ms inference
41 voices across 3 providers
OpenAI, ElevenLabs and xAI Grok, with accent, gender and speed options.
- OpenAI TTS: gpt-4o-mini-tts with 10 voices (alloy, ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer)
- ElevenLabs: 26 voices on v3, v3 Conversational and Flash 2.5
- Grok TTS: 5 voices (Ara, Rex, Sal, Eve, Leo)
32+ languages
Multilingual voice support across all providers out of the box.
- Configurable language per agent for STT, TTS, and LLM independently
- Voice best practices injection optional per language
Natural personality
Listening cues like "uh-huh" and subtle office ambience for realism.
- Configurable acknowledgment word list and frequency
- Pre-synthesized cues at session start for zero latency
- Background audio (e.g., call-center ambience) with adjustable volume
Per-agent tuning
Temperature, turn detection, and silence thresholds per agent.
- Temperature from 0 to 2 (default 1.0) for creativity vs. determinism
- Smart turn detection with configurable activation threshold and silence duration
- Dynamic system prompts with template variables resolved at call start
Call Experience
Configurable greetings
AI-generated, pre-defined message, or silent mode.
- Dynamic: AI generates greeting from system prompt context
- Static: pre-defined message (e.g., "Hi, I am Kelly...")
- Silent: agent waits for the caller to speak first
Noise suppression
Deep-learning noise cancellation for clearer conversations.
- DTLN deep-learning noise cancellation plugin
- Configurable strength from 0 (off) to 1 (maximum)
- Applied to caller audio for clearer agent comprehension
Cost & spam controls
Max duration, silence hangup, and inactivity reminders.
- Inactivity reminders with configurable interval (e.g., 10-15s) and max count
- Max call duration up to 1 hour (3,600,000ms)
- End call after configurable silence timeout (180s-600s)
- Configurable ring duration before timeout (default 30s)
Look up information
Query any system for order status or account details mid-call.
- Custom HTTP tools with configurable method, headers, body, and timeout
- Response variables capture values via JSONPath extraction
- Values injected into conversation context for follow-up use
Book appointments
Real-time availability via Calendly and Cal.com.
- Calendly integration with event type URI and timezone configuration
- Cal.com integration with full availability checking
- Real-time slot availability and instant booking confirmation
Call any API
HTTP endpoints with configurable auth and request templates.
- Configurable HTTP method, headers, auth, body templates, and query params
- Body templates with {{variable}} placeholders resolved at call time
- Supports JSON and SOAP/XML response parsing
- Strict JSON schema compliance mode for function calls
Transfer calls
Zero-gap audio handoff to humans or external numbers.
- Cold transfer via SIP REFER for zero-audio-gap handoff
- Transfer to human agents or external phone numbers
- Transfer destination automatically logged in call record
Agent-to-agent transfer
Hand off a live call to a different AI agent with a completely new persona.
- New agent fully takes over — different voice, LLM, prompt, tools, and knowledge base
- Caller stays on the same line with no interruption
- Configurable context passing: full conversation history, AI-generated summary, or fresh start
- Multiple transfer targets per agent — route to billing, support, or sales based on conversation
- New agent greets the caller using its own voice and personality
- Transfer tracked in call logs with target agent link
Voicemail detection
Detect machines and leave messages or generate AI responses.
- Automatic answering machine detection (CPA/AMD)
- Three actions: hang up, send static message, or LLM-generated message
- Audio resampled 48kHz to 8kHz with G.711 ULAW encoding
- Available on outbound calls only
Knowledge & Intelligence
Multiple sources
Crawl web pages, upload PDFs / Word docs, or paste text.
- URL crawling with auto-discovery of linked pages, include/exclude path patterns
- PDF and Word document upload with size tracking
- Direct text input with custom titles
Auto-refresh
Scheduled re-crawling keeps web content current.
- Configurable re-crawl interval per URL source
- Automatic status tracking: pending, processing, completed, failed
- Error messages and progress details per source
Vector search
Semantic retrieval with automatic chunking and deduplication.
- OpenAI text-embedding-3-small (1536d) or AWS Bedrock (256d)
- Automatic content chunking with token counting and metadata
- Content hashing for deduplication across sources
- Full-text keyword search fallback alongside semantic search
Analytics dashboard
Volume trends, performance, daily costs, disconnection analysis.
- Call volume trends (line chart), calls by agent (bar chart)
- Sentiment and outcome distribution (pie/bar charts)
- Cost per day with consolidated cost/duration metrics
- Disconnection reason analysis and agent performance comparison
Full call detail
Transcript, audio, cost breakdown, latency, and event timeline.
- Transcript with speaker labels and tool call visibility
- Audio playback synced to transcript — click to seek, auto-highlight during playback
- Cost breakdown: LLM input/output tokens, TTS characters, STT minutes
- Latency metrics: LLM TTFT, TTS TTFB, end-to-end delay
- Events timeline, raw logs, dynamic variables, KB retrieval URLs
Post-call analysis
Automatic summary, sentiment, success evaluation, and outcome.
- Call summary (1-3 sentences) and success evaluation (boolean)
- Sentiment: Positive / Negative / Neutral / Unknown
- Outcome: Contained / Partially Contained / Business Rule Transfer / Early Abandon / Hangup / Agent Escalation / Confusion Transfer / Timeout
- Configurable analysis model and custom analysis prompts
Custom extraction
Define additional data points to extract per agent.
- Per-agent fields with types: string, boolean, enum, number
- Extracted automatically from transcript after each call
- Results available via API, dashboard, and webhooks
Webhooks
call_started, call_ended, call_analyzed events with auto retry.
- Configurable URL with timeout per webhook
- Automatic retry logic with delivery tracking
- Response logging for debugging integrations
Channels & Reach
Web widget
Voice on any website in an inline embed. Chat and the floating bubble are coming soon.
- Modes: voice today; chat coming soon
- Display: inline (embedded in page) today; floating bubble coming soon
- Real-time audio streaming via WebRTC
- Try the embed demo
Fully customizable
Colors, themes, icons, branding, and per-page overrides. White-label ready.
- Theme: light / dark / auto with accent color (hex)
- Position, visualizer type (aura/wave), layout (side-by-side/stacked)
- Custom bubble icon and label, panel width and height
- Control visibility: microphone, chat, and leave button toggles
- White-label: toggle Blowfish branding off per embed
- Security: allowed origins (CORS), rate limiting, Cloudflare Turnstile bot protection
Number marketplace
Search, purchase, and manage phone numbers from the dashboard.
- Twilio integration for search, purchase, and management
- Filter by region/country with E.164 validation
- Assign numbers to agents with inbound/outbound direction tracking
SIP trunking
Inbound/outbound trunks with IP or credential authentication.
- Inbound: IP-based (allowed addresses) or credential-based (username/password) auth
- Outbound: configurable SIP server, transport (UDP/TCP/TLS/WS/WSS), phone number pool
- Custom headers and metadata per trunk
Dispatch rules
Route incoming calls to agents by trunk and phone number.
- Route by trunk, inbound number, and metadata
- Rule types: individual routing (to agent) or direct routing (to trunk)
- Auto-cleanup: outbound trunk deleted when last number removed
Outbound API
Launch calls with per-call prompt, greeting, and metadata overrides.
- E.164 format for from_number and to_number
- Per-call agent override, greeting, prompt, goodbye, voicemail behavior
- Metadata and dynamic variables injected at call start
- Form-based key/value builder alongside raw JSON mode
Agent Builder
Visual configuration
Voice, model, prompt, tools, knowledge, and telephony in one place.
- Voice catalog organized by provider, gender, accent, age with language filtering
- LLM catalog by provider, type (traditional vs. S2S), and billing model
- STT catalog with language support metadata
Draft & publish
Edit freely without affecting live agents, publish with one click.
- Draft/published state machine for safe editing
- Published configs synced to production cache for instant reads
- Cron job keeps configs in sync across environments
Dynamic prompts
Template variables like {{caller_name}} resolved at call start.
- Variables: {{caller_name}}, {{current_time_{{timezone}}}}, {{user_number}}, {{franchise_id}}
- Resolved at call start from metadata and system context
- Can reference values captured from previous tool calls
Browser testing
Push-to-talk or open-mic with support for test variables.
- Push-to-talk button alongside open-mic mode
- Pass custom variables to simulate real call metadata
- Full transcript and tool call visibility during test
Voice preview
Listen to voice samples before choosing.
- Cached voice samples with signed URLs for instant playback
- Listen directly in the voice selector
- Organized by provider, gender, accent, and age
Retell migration
Import wizard maps existing Retell agents, prompts, and voices.
- Automatic mapping of agents, prompts, tools, voices, and dynamic variables
- Provider and voice matching across platforms
- One-click migration from Retell AI
Platform
Org-based isolation
Each org sees only its own agents, calls, numbers, and data.
- Organization-scoped agents, calls, numbers, and knowledge bases
- Self-service org creation from the dashboard with org switching
- Cross-tab logout synchronization across browser sessions
Auth & roles
Email/password, magic link, Google SSO, admin/member RBAC.
- Email/password, magic link (passwordless), and Google SSO
- Email verification and password reset flows
- Admin and member roles with settings restrictions
- Signup approval workflow: pending, approved, or rejected
Super admin
Manage all orgs, user impersonation with audit trail.
- Dashboard to manage all organizations and users across the platform
- User impersonation with required reason, session expiration, and revocation
- Audit log: auth events, admin actions, user activity with IP and user agent
REST API & docs
Full API with interactive OpenAPI reference.
- CRUD endpoints for agents, calls, numbers, KBs, telephony, webhooks, voices, LLMs, embeds
- Auto-generated OpenAPI reference with interactive API explorer
- Internal API for secure agent-to-dashboard communication
MCP server
AI assistants can manage agents, calls, and numbers.
- List agents, get call details, list phone numbers, make outbound calls
- OAuth 2.1 provider: AI assistants authenticate via standard OAuth flow
- Connect to third-party MCP servers with per-server auth and timeout
Webhooks & cron
Call event subscriptions with retry and automated background jobs.
- Events: call_started, call_ended, call_analyzed
- Configurable URL, timeout, delivery tracking, automatic retry
- Automated background jobs: stale call cleanup, KB refresh, config sync
Tell us what you are building
Tell us what you are making. We will walk you through the engines, the API, and what it takes to run it in production.