If you are running an AI voice agent in India at any real volume, the per-minute rate is the number that decides whether the project pencils out. Most managed platforms quote an all-in price — telephony sometimes included, AI providers always bundled — and that bundling is where the markup hides. The honest answer to "what is the cheapest way to run a voice agent?" is: stop paying a markup on AI usage you could buy directly.
That is exactly what a bring-your-own-keys (BYOK) model does. You wire in your own speech-to-text, language model, and text-to-speech API keys, and Edesy charges a platform fee from ₹1/min with all your own keys (₹1.20 with some, ₹1.50 standard) for orchestration — nothing on top of your AI usage. This post breaks down how the BYOK platform fee works, when it actually saves money, the full cost breakdown, a worked example against a standard ₹4-6/min all-in rate, and how to set it up in a workspace.
Want the capability details, not just the cost angle? The full breakdown of which providers you can bring and how key resolution works lives on the BYOK & BYOP Voice AI page. This post is the money-math companion to it.
Quick verdict (TL;DR)
- BYOK price: from ₹1/min to Edesy with all your own keys (₹1.20 with some, ₹1.50 standard) as a platform and orchestration fee only. It does not include your AI provider costs and does not include telephony.
- Your true per-minute cost = platform fee (₹1, ₹1.20 or ₹1.50/min) + your AI provider cost/min + your telephony cost/min.
- Standard (platform keys) price: from ₹6/min pay-as-you-go with standard AI providers bundled in; paid plans bring it to ₹4-5/min. Telephony is billed separately on every plan.
- BYOK wins when you have high volume or already-negotiated AI contracts. The standard plans win for a fast, fully managed start at low volume.
| Scenario | Cheapest option |
|---|---|
| Under ~5,000 min/month, no AI contracts | Standard plan (Pro/Max) — simpler, managed |
| 10,000+ min/month | BYOK at ₹1-1.50/min platform + your own AI |
| Existing OpenAI / Gemini committed spend | BYOK — reuse your negotiated rates |
| Data residency requirements | BYOK — AI usage stays on your accounts |
| Fastest possible go-live, low ops | Standard plan — nothing to manage |
Why the per-minute rate is misleading
When a platform quotes "₹5/min, all-inclusive," that single number is really three costs stacked together: the platform's orchestration, the AI providers (STT + LLM + TTS), and — sometimes — telephony. Bundling is convenient, but it means you pay the platform's rate for AI usage rather than the provider's rate. If you already buy GPT-4o or Gemini at a negotiated price, you are paying twice for the same tokens: once in your own contract you are not using, and once inside the bundled minute.
The BYOK model unbundles those three costs so you can see and control each one:
- Platform / orchestration — Edesy's real-time voice runtime, multilingual stack, session routing, dashboards, and analytics. This is the BYOK platform fee: from ₹1/min with all your own keys, ₹1.20 with some, ₹1.50 standard.
- AI providers — STT, LLM, TTS. Runs on your keys, billed to your provider invoices at your rates.
- Telephony — inbound/outbound minutes, billed by your carrier or a platform telephony provider, separately.
Crucially, the platform fee covers only the first bucket. Anyone who tells you the BYOK fee is "all-in" is wrong — the AI and telephony buckets are real and you pay them separately. The saving is not that those costs vanish; it is that Edesy adds no markup on them.
The full cost breakdown
Here is how the two models line up. Note that telephony is separate in both columns — it is never folded into a headline rate.
| Cost bucket | Standard (platform AI keys) | BYOK (your own AI keys) |
|---|---|---|
| Edesy platform / orchestration | Bundled into the rate | from ₹1/min (₹1.20 with some keys, ₹1.50 standard) |
| STT + LLM + TTS | Included at platform's rate | Your provider invoices, your rates |
| Edesy markup on AI | Built into the bundled rate | None |
| Telephony | Billed separately | Billed separately (or bring your own carrier) |
| Headline number | From ₹6/min pay-as-you-go | from ₹1/min + your AI + your telephony |
| Paid plans | Pro ₹1,499/mo (₹5/min), Max ₹4,999/mo (₹4.50/min), Ultra ₹14,999/mo (₹4/min) | Platform fee from ₹1/min (₹1.20 with some keys, ₹1.50 standard) |
| What you manage | Nothing — fully managed | Your own AI provider keys |
The BYOK providers you can bring are broad, so you can shop each slot for the cheapest capable option:
- LLM: OpenAI (GPT-4o, GPT-4o-mini), Anthropic Claude, Google Gemini, Azure OpenAI, Groq, Cerebras
- STT: Deepgram, Google Chirp, Azure Speech, ElevenLabs Scribe, AssemblyAI, OpenAI Whisper, Sarvam Saarika, Soniox
- TTS: ElevenLabs, Cartesia, Google, Azure, Deepgram, OpenAI, Sarvam Bulbul, Soniox
You can also mix: bring your own LLM and let Edesy handle STT, for example. Key resolution is slot-level, so you only self-manage the providers where it actually saves you money. The mechanics of that — including which model options exist per slot — are covered on the LLM providers page.
Worked example: BYOK vs standard at 10,000 minutes/month
Let's make the math concrete. Assume a mid-market team running 10,000 outbound minutes a month, telephony held constant so we compare like-for-like on platform + AI.
Standard plan (Ultra, ₹14,999/mo, ₹4/min):
- Platform + bundled AI: 10,000 min x ₹4 = ₹40,000 (the monthly plan fee offsets against usage)
- Telephony: billed separately
- Platform + AI subtotal: ~₹40,000/month
BYOK (₹1.50/min standard platform fee + your own AI; ₹1/min if every provider uses your keys):
- Edesy platform: 10,000 min x ₹1.50 = ₹15,000 (₹10,000 at ₹1/min)
- Your AI usage (STT + LLM + TTS on your own accounts, at your rates): variable, but for a lean stack — a fast STT, a mini-class LLM, and a cost-efficient TTS — this often lands well under the ~₹2.5/min gap the standard bundle was charging for AI
- Telephony: billed separately (same as standard)
- Platform + AI subtotal: ₹15,000 platform + your AI bill
The BYOK column wins as long as your own AI cost per minute is less than the ~₹2.5/min the standard plan was effectively charging for AI on top of the platform fee. With mini-class models and efficient STT/TTS, or with a negotiated enterprise contract, that is usually comfortably true — which is why BYOK is the cheapest path at volume. At low volume, the platform fee and self-management overhead make the standard managed plans the simpler call.
Not sure which side of the line you fall on? The BYOK pricing page lays out the Standard-vs-BYOK comparison table and the plan cards side by side so you can model your own numbers before committing.
Decision framework: pick BYOK if...
Check the boxes that apply to you:
- You run more than ~10,000 voice minutes a month
- You already have a negotiated or committed-spend AI contract (OpenAI, Gemini, etc.)
- You need AI usage to stay on your own accounts for data residency or compliance
- Your finance team wants AI and platform costs as separate, attributable line items
- You want to shop each provider slot for the cheapest capable model
If three or more apply, BYOK is very likely your cheapest option. If fewer apply — especially if you are at low volume and want zero ops — a standard plan is the more pragmatic choice, and you can switch to BYOK later without re-platforming.
How to set up BYOK pricing
Moving a workspace onto the BYOK platform rate is a settings change, not a migration:
- Open your workspace settings. Go to Settings > Integrations > Provider Keys.
- Paste your provider keys. Add keys for each slot you want to self-manage — LLM, STT, TTS. You do not have to bring all three; slot-level mixing is allowed.
- Keys are encrypted at rest and take precedence over platform defaults automatically. There is nothing else to toggle.
- Verify per call. Every call logs a
byokflag so you can confirm usage ran on your own keys and is billed at the platform rate. - Bring your own telephony too, if you want. You can add your own Twilio, Plivo, or Exotel account, or your own SIP trunk, and pay your carrier directly. See Bring Your Own SIP Trunk for the setup.
Once keys are in place, that usage is billed at the BYOK platform fee (from ₹1/min), with AI on your provider invoices and telephony billed separately. You can start a workspace and add keys here.
Frequently asked questions
Is the BYOK fee really the total cost of a voice agent call?
No — and it is important to be precise about this. The BYOK fee (from ₹1/min with all your own keys, ₹1.20 with some, ₹1.50 standard) is Edesy's platform and orchestration fee only. Your STT, LLM, and TTS usage is billed directly to you by your AI providers at your own rates, and telephony (inbound and outbound minutes) is billed separately by your carrier or a platform telephony provider. Your true per-minute cost is ₹1.5/min plus your AI cost per minute plus your telephony cost per minute. The value of BYOK is not that AI and telephony are free — it is that Edesy adds no markup on them.
How much cheaper is BYOK than the standard rate?
On standard plans, AI is bundled into the per-minute price: from ₹6/min pay-as-you-go, or ₹4-5/min on paid plans (Pro ₹1,499/mo at ₹5/min, Max ₹4,999/mo at ₹4.50/min, Ultra ₹14,999/mo at ₹4/min). BYOK drops the Edesy charge to a flat ₹1.5/min because AI runs on your own accounts. Whether your all-in total is lower depends on your own AI cost per minute — at volume or with negotiated contracts, it typically is. Telephony is billed separately in both models, so it does not change the comparison.
When does BYOK not make sense?
At low volume or when you want a fully managed setup with nothing to administer. If you are running a few thousand minutes a month and do not have existing AI contracts, the standard plans are simpler: one bill, standard providers included, no keys to manage. You can always migrate to BYOK later once volume grows — it is a settings change, not a rebuild.
Which AI providers can I bring to get the BYOK rate?
For LLM: OpenAI (GPT-4o, GPT-4o-mini), Anthropic Claude, Google Gemini, Azure OpenAI, Groq, Cerebras. For STT: Deepgram, Google Chirp, Azure Speech, ElevenLabs Scribe, AssemblyAI, OpenAI Whisper, Sarvam Saarika, Soniox. For TTS: ElevenLabs, Cartesia, Google, Azure, Deepgram, OpenAI, Sarvam Bulbul, Soniox. You can bring one slot or all of them, and you can also bring your own telephony provider and SIP trunk.
Where do I enter my keys and are they safe?
Keys go in your workspace under Settings > Integrations > Provider Keys. They are encrypted at rest, and customer keys take precedence over platform defaults, so once they are in place the runtime uses them first. Each call logs a byok flag so you can audit that usage ran on your own keys. Full details are on the BYOK & BYOP page.
Conclusion
The cheapest way to run an AI voice agent in India is not to find a lower bundled per-minute quote — it is to unbundle. Bring your own STT, LLM, and TTS keys, pay Edesy a platform fee from ₹1/min with all your own keys (₹1.20 with some, ₹1.50 standard), and keep your AI usage on your own accounts at your own rates. Just remember the platform fee covers orchestration only: your AI provider bill and your telephony are billed separately, and your real per-minute cost is the sum of all three.
If you have volume or existing AI contracts, that sum almost always beats a standard ₹4-6/min all-in rate. Start a workspace and add your keys to switch to BYOK platform pricing (from ₹1/min), and see the full comparison on the AI Voice Agent product page.
Related resources:
- BYOK & BYOP Voice AI — bring your own AI keys and telephony provider
- BYOK Pricing — the BYOK cost math (from ₹1/min) and comparison table
- LLM Providers — which models you can bring per slot
- Bring Your Own SIP Trunk — self-manage telephony
- AI Voice Agent — the full product overview
Related reading
- Vapi and Retell AI alternatives for India
- White-label AI voice agent for agencies
- AI voice agent pricing comparison tool — compare per-minute costs across platforms