Truffl
Voice OS control plane
Not a bot builder. A deployment engine with layers for context, compilation, runtime, quality, and learning.
Architecture
Every Truffl workflow runs through a layered OS — from raw business data to live voice calls to continuous improvement.
Your agent should know the business before it starts talking.
Inputs
Outputs
Recommend the highest-ROI workflow for your vertical.
Inputs
Outputs
From context to agent package — fully automated.
Inputs
Outputs
Orchestrate STT, LLM, TTS, and tools under 300ms.
Inputs
Outputs
Test the agent before your customers do.
Inputs
Outputs
Surface quality gaps and improve automatically.
Inputs
Outputs
Layer 1
Truffl ingests your existing data sources and assembles a structured business brain — no manual prompt writing required.
Ingestion checklist
Outputs
Layer 3
The Agent Compiler assembles all required layers automatically — no manual configuration.
System Prompt
Base agent persona and constraints
Conversational Prompt
Dynamic dialogue scaffolding
Response Guidelines
Tone, length, and escalation rules
FAQ / KB
Knowledge base injection
Custom Variables
Business-specific dynamic fields
Tool Schemas
CRM, payment, and API integrations
Escalation Rules
Human handoff triggers and paths
Output Fields
Structured data captured after call
Eval Criteria
Pass/fail rubric for simulation runs
Layer 5
Truffl runs your agent against 10 test personas before launch. Every edge case covered before going live.
Deployment
Deploy across five channels with built-in rules for retry, fallback, handoff, and compliance.
Channels
Outbound phone
Trigger calls from CRM events, webhooks, or schedules
Inbound number
Assign a DID and handle inbound customer queries
Website voice widget
Embed a voice CTA directly on your site or landing page
WhatsApp callback
Trigger voice from WhatsApp thread events
Marketing journey action
Drop a voice call step into any MoEngage or CleverTap journey
Rules engine
Dashboard
Every call, every outcome, every quality signal — tracked in a single Revenue Dashboard.
1,240
Calls triggered
This week
68%
Connected
Connection rate
183
Qualified leads
Post-call
₹4.2L
Revenue influenced
Recovered + influenced
287ms
Avg latency
P50 response
Voice stack
Swap any layer without rebuilding your workflow. Truffl handles provider routing automatically.
| Layer | Providers |
|---|---|
| STT | Deepgram· nova-2, low latencySarvam· Indic languagesWhisper· Multi-languageAzure Speech· Enterprise grade |
| TTS | ElevenLabs· Natural voiceCartesia· Ultra-low latencyAzure Neural· Broad language supportGoogle WaveNet· High fidelitySarvam Bulbul· Hindi-first |
| LLM | GPT-4o· OpenAI flagshipGPT-4o-mini· Fast + affordableGemini 1.5 Flash· Google low-latencyGemini 1.5 Pro· Google high-qualityClaude Sonnet· AnthropicLlama 3· Open-sourceMistral· European open model |
| Telephony | Twilio· GlobalPlivo· GlobalExotel· India-firstSIP providers· Custom SIP trunk |
| VAD | Silero· Default presetDeepgram VAD· Paired with Deepgram STTCustom presets· Tune silence threshold |
Version control
Deploy a new agent version to production traffic
Revert to any previous stable version instantly
Diff two versions side-by-side with eval scores
A/B test versions with configurable traffic allocation
Start free with 100 calls. No prompt writing, no flow mapping.