How Talkar Works
High-level platform model, Master/Sub-Org account hierarchy, lifecycle phases, UI gating, Playground vs. Live rules, and exact tier & margin-floor governance.
Talkar pairs a real-time Voice & Text Agent Engine (handling WebRTC, PSTN/SIP telephony, LLM orchestration, Silero VAD, RAG, tools, and campaigns) with a Customer & Billing Platform Layer (handling onboarding briefs, prepaid wallets, Razorpay top-ups/auto-recharge, DID leases, quotas, and alerts).
Voice & Text Chat Modalities
Every agent workflow you build runs across:
- Inbound & Outbound Phone Calls (PSTN / SIP trunks across 7 supported production telephony providers + Mock)
- Browser WebRTC & Website Voice Widgets
- API & Embed Text Chat Sessions (using the exact same graph nodes, prompts, KB retrieval, extraction, and post-call webhooks — useful when clients want omnichannel voice + chat behavior)
Account Topologies: Independent, Master & Sub-Org
Organizations can operate either standalone or in a shared fleet hierarchy:
| Topology | Wallet & Billing | Plan & Concurrency | Typical Use Case |
|---|---|---|---|
| Independent / Master Org | Owns its own prepaid wallet, Razorpay invoices, GST billing, and Auto-Recharge mandate | Owns tier plan (starter, growth, pro, custom) and fleet concurrency cap | Single company or parent client org |
| Sub-Org (Shared Fleet) | Shares the Master org's prepaid wallet (no separate wallet) | Inherits Master tier (or sub-org approved tier capped by master tier rank, unless master is custom which locks inheritance) and draws from the shared fleet concurrency pool; can optionally set a per-sub-org monthly spend cap or concurrency sub-cap | Separate client departments, campaigns, or regional business units |
- Switching modes: Moving a Sub-Org to Independent creates its own empty wallet and places it in
pending_deposituntil funded; moving an Independent org to Shared transfers any positive stranded balance (credit_transfer) into the target Master wallet. - Additional agent creation: Sub-orgs created via a new-agent brief start in
under_reviewand—when attached to anactiveMaster Organization—activate fromagent_buildingas soon as the Master wallet balance is strictly> ₹500(rather than needing another fullactivation_deposit; otherwise>= activation_depositis still required). - Whole-minute rounding & floors: Connected calls always round up to whole minutes (
ceil(duration_seconds / 60)), while<= 0scalls,textchat,mock, andagent_buildingWebRTC calls bill₹0.
Client Lifecycle Phases & UI Gating
A client workspace moves through statuses that control what dashboard pages and calling channels are accessible:
- UI Route Gating: Accounts in
pending_approval,info_requested,rejected, or brief-less sub-orgagent_buildingredirect automatically to onboarding; accounts inpending_depositorsuspendedredirect to wallet/billing pages so funds can be added cleanly.
Playground (agent_building) vs. Live (active) Rules
- Allowed for free: Browser WebRTC voice tests, Text Chat sessions, and Mock soak tests bypass wallet balance checks (capped at 2 concurrent playground sessions) and log at
₹0cost. - Blocked: Real PSTN/SIP outbound or inbound phone calls cannot run while status is
agent_building. - Suppressed Alerts: Automated low-balance, tier-upgrade, credit-grant, and Automation
/notifyalerts are suppressed so clients aren't flooded during setup. - UI Banner: Clients see a non-blocking banner indicating their agent is currently being built while still being able to test drafts.
Tier Managed Credits, Margin Floors & BYOK
Talkar Managed Tiers
| Tier | Rate | Activation Deposit | Fleet Concurrency | Max Call Duration | Free DIDs | Managed Stack |
|---|---|---|---|---|---|---|
starter | ₹6/min | ₹6,000 | 2 | 15 min (900s) | 1 | gpt-4o-mini / Deepgram STT / Deepgram aura-2 TTS |
growth | ₹6/min | ₹6,000 | 2 | 15 min (900s) | 1 | gpt-4o-mini / Deepgram STT / SmallestAI lightning_v3.1_pro TTS |
pro | ₹18/min | ₹8,000 | 10 | 30 min (1800s) | 2 | gpt-4o-mini / Deepgram STT / ElevenLabs eleven_flash_v2_5 TTS |
custom | Custom (default ₹18/min) | Custom (default ₹8,000) | Custom (default 10) | Custom (default 30 min) | Custom (default 2) | Admin-configured custom pricing & overrides |
(Note: elite exists in tier definitions as disabled / non-self-serve and cannot be picked via self-serve checkout. Excess DIDs beyond a tier's free fleet allowance default to ₹500/month per DID.)
Voice Margin Floors
To prevent negative unit margins on premium TTS engines, Talkar enforces minimum per-minute floors:
- ElevenLabs / Cartesia TTS: Minimum floor
₹12/min, and runtime quota/deduction checks automatically floor active ElevenLabs/Cartesia agents to at least₹18/min(protier rate) unless an explicit per-agent custom rate override is configured. - SmallestAI TTS: Minimum floor
₹5/min. - Deepgram / Default TTS: Minimum floor
₹4/min. - BYOK (Bring Your Own Key): Organizations or individual workflows can supply their own API keys for Pipeline or Realtime S2S models—see Voice, Models & STT/TTS Selection.
Welcome to Talkar Builder Docs
Internal engineering & delivery handbook for building client voice/chat agents, connecting CRMs, running campaigns, and debugging live calls.
Quickstart: Build & Test Your First Client Agent
Go from blank canvas (or AI template) to a tested, published client agent in minutes.