Platform Comparison
Honest, sourced comparison of letustalk.ai vs Vapi, Retell, Bland, and open-source DIY voice frameworks (May 2026).
This page is maintained as an honest, sourced comparison. Every claim links to either an official platform doc or a 2026 third-party review. If you spot something stale, open a docs PR.
| Layer | What you can do |
|---|
| Dashboard UI | Pick LLM/STT/TTS providers, save custom org-scoped templates, tune VAD/noise/turn-detection per agent, configure post-call analysis schema, set webhook URL + event filter, manage HTTP & client tools, set transfer destinations |
| REST API + SDKs (TS, Python) | Full CRUD on agents, calls, campaigns, templates, settings. Per-call dynamic variables, assistant overrides, agent versioning, Idempotency-Key support |
| MCP server | Claude Desktop / any MCP client can list agents, place calls, fetch templates, check provider health, get cost presets |
| BYO API keys | Per-call provider keys passed in provider_keys — route through your own OpenAI / Anthropic / ElevenLabs accounts at zero markup |
| Self-host | The full letustalk.ai stack — bot pipeline + Hono API + Next.js dashboard — runs on your infrastructure with Postgres + Redis + WebRTC media layer + Twilio |
| Feature | letustalk.ai | Vapi | Retell | Bland | OSS DIY frameworks |
|---|
| LLM provider choice | ✅ | ✅ All major | ✅ All major | ✅ + custom | ✅ DIY |
| STT/TTS provider choice | ✅ | ✅ | ✅ | ✅ BYOK | ✅ DIY |
| Sub-600ms latency | ⚠️ Tunable | ✅ | ✅ ~600ms | ✅ sub-400ms | ✅ |
| Noise suppression (RNNoise + Krisp) | ✅ Both | ⚠️ Krisp paid | ✅ | ❌ | ⚠️ DIY |
| Multi-language (30+) | ✅ EN + 29 via ElevenLabs | ✅ | ✅ 31+ languages | ✅ | ✅ DIY |
| Channel-aware VAD defaults | ✅ Web vs phone tuned | ⚠️ Manual | ⚠️ Manual | ❌ | ⚠️ Manual |
| Feature | letustalk.ai | Vapi | Retell | Bland | OSS DIY |
|---|
| Visual drag-and-drop flow builder | ❌ Prompt-driven | ✅ Flow Studio | ✅ | ⚠️ Code-first | ❌ |
| Industry templates | ✅ 13 built-in across 8 categories | ✅ ~20 | ✅ | ⚠️ | ⚠️ Examples |
| Org-scoped custom templates | ✅ Save + share within team | ⚠️ | ⚠️ | ❌ | ❌ |
| Quality presets with live cost ranges | ✅ (unique) | ❌ | ❌ | ❌ | ❌ |
| AI-generated prompts | ✅ | ✅ | ⚠️ | ❌ | ❌ |
| Multi-agent squads / transfer | ✅ | ✅ Squads | ✅ | ✅ | ⚠️ DIY |
| Simulator + LLM-as-judge eval | ✅ | ✅ | ⚠️ | ❌ | ⚠️ |
| Channel | letustalk.ai | Vapi | Retell | Bland | OSS DIY |
|---|
| Web (WebRTC) | ✅ | ✅ | ✅ Web SDK | ⚠️ | ✅ |
| Phone in/outbound (SIP) | ✅ Twilio | ✅ Twilio/Vonage/Telnyx | ✅ | ✅ | ⚠️ DIY |
| WhatsApp voice (Meta Calling API) | ✅ | ✅ Meta partner | ❌ | ❌ | ❌ |
| WhatsApp text in same agent | ✅ (unique) | ❌ | ❌ | ❌ | ❌ |
| Chat (text-only LLM in same agent) | ✅ | ❌ | ❌ | ❌ | ❌ |
| Embeddable widget | ✅ | ✅ | ✅ | ⚠️ | ⚠️ DIY |
| Feature | letustalk.ai | Vapi | Retell | Bland | OSS DIY |
|---|
| Self-host / on-prem | ✅ Full stack | ❌ Managed only | ⚠️ Enterprise (verify) | ✅ Air-gapped | ✅ OSS |
| SOC 2 / HIPAA | ⚠️ Roadmap | ✅ SOC 2 | ✅ SOC 2 | ✅ SOC 2 + HIPAA | ⚠️ DIY |
| BYO API keys (per call) | ✅ | ✅ | ✅ | ✅ | ✅ DIY |
| BYO S3 / private storage | ✅ Per-org | ❌ | ❌ | ✅ Enterprise | ✅ DIY |
| Recording multi-channel | ✅ | ⚠️ | ⚠️ | ❌ | ⚠️ DIY |
| Recording retention controls | ✅ Per-org days | ⚠️ | ⚠️ | ⚠️ | ❌ |
| Concurrency at 1M+ | ⚠️ Atomic gate, un-load-tested | ⚠️ ~1000 | ⚠️ Thousands | ✅ Proven 1M | ⚠️ |
| Feature | letustalk.ai | Vapi | Retell | Bland | OSS DIY |
|---|
| Live transcript | ✅ | ✅ | ✅ | ✅ | ✅ DIY |
| Per-call cost breakdown (LLM/STT/TTS/SIP) | ✅ | ✅ | ✅ | ⚠️ | ❌ |
| Per-call latency breakdown (TTFB stacked) | ✅ (unique UI) | ⚠️ Numbers only | ⚠️ Numbers only | ❌ | ❌ |
| Provider health badges (live, 60s cached) | ✅ (unique) | ❌ | ❌ | ❌ | ❌ |
| Per-agent analytics (spend + latency + volume) | ✅ | ✅ | ✅ | ⚠️ | ❌ |
| Custom dashboards | ⚠️ Roadmap | ✅ | ✅ A/B test built-in | ⚠️ | ❌ |
| Sentiment + CSAT tracking | ✅ Post-call analysis | ✅ | ✅ | ⚠️ | ❌ |
| Agent versioning + rollback | ✅ | ⚠️ | ⚠️ | ❌ | ❌ |
| Feature | letustalk.ai | Vapi | Retell | Bland | OSS DIY |
|---|
| HMAC-signed payloads | ✅ SHA-256 | ✅ | ✅ | ⚠️ | ❌ |
| Retry with exponential backoff | ✅ 1m→12h, 6 attempts | ✅ | ✅ | ⚠️ | ❌ |
| Dead-letter queue | ✅ | ⚠️ | ⚠️ | ❌ | ❌ |
| Delivery log UI (last 100) | ✅ Redis-backed | ⚠️ | ⚠️ | ❌ | ❌ |
| Event-level subscription filter | ✅ | ✅ | ⚠️ | ❌ | ❌ |
| Test ping button | ✅ | ✅ | ⚠️ | ❌ | ❌ |
| Idempotency-Key (Stripe-style) | ✅ 24h cache | ⚠️ | ⚠️ | ❌ | ❌ |
| MCP server | ✅ 8 tools | ✅ Hosted + SSE | ❌ | ❌ | ⚠️ |
| Knowledge base / RAG | ⚠️ Via tools | ✅ Native | ✅ | ⚠️ | ⚠️ DIY |
| Pre-built integrations marketplace | ❌ | ✅ Cal.com, HubSpot, etc. | ✅ | ❌ | ❌ |
| Feature | letustalk.ai | Vapi | Retell | Bland | OSS DIY |
|---|
| TypeScript / Node SDK | ✅ Zero deps | ✅ | ✅ | ✅ | ✅ |
| Python SDK (sync + async) | ✅ | ✅ | ✅ | ⚠️ | ✅ |
| MCP server for Claude Desktop | ✅ | ✅ | ❌ | ❌ | ⚠️ |
| Public REST API | ✅ | ✅ | ✅ | ✅ | ⚠️ DIY |
| Public templates endpoint | ✅ (unique) | ❌ | ❌ | ❌ | ❌ |
| OpenAPI / Postman | ⚠️ Roadmap | ✅ | ✅ | ✅ | ❌ |
| Webhook signature verifier in SDK | ⚠️ Docs only | ✅ | ✅ | ⚠️ | ❌ |
| Aspect | letustalk.ai | Vapi | Retell | Bland | OSS DIY |
|---|
| Platform fee/min | Hidden in USD price | $0.05/min | $0.07/min | $0.20–0.30/min BYOK | OSS |
| Live cost range (low–typical–high) | ✅ (unique) | ❌ Single | ❌ Single | ❌ Single | n/a |
| Per-component breakdown in UI | ✅ LLM/STT/TTS/SIP | ⚠️ | ⚠️ | ❌ | n/a |
| Decimal.js billing precision | ✅ | ✅ | ✅ | ⚠️ | n/a |
| USD billing + Stripe | ✅ | ✅ | ✅ | ✅ | n/a |
| Concurrency upcharge | ❌ Atomic gate, no $/line | ⚠️ $10/extra line | ⚠️ | ⚠️ | n/a |
- Unified channel stack — USD billing, Stripe, and WhatsApp voice + text in the same agent. No competitor combines these.
- Cost transparency — only platform that renders a live low/typical/high range from the actual rate tables. Competitors quote a misleading single number.
- Webhook reliability surface — log UI + event filter + Idempotency-Key together. Competitors have pieces; not the whole set.
- Provider health badges — only platform that surfaces live OpenAI/Anthropic/Groq/Deepgram/ElevenLabs status BEFORE you place a call.
- Per-call latency breakdown chart — stacked-bar TTFB visualization to spot which provider is the bottleneck. Competitors expose raw numbers; not the visual.
- Open-source-style customization — full self-host of the entire letustalk.ai stack on your own infrastructure. Vapi/Retell are closed managed services.
- No visual flow builder. Vapi (Flow Studio) and Retell both ship drag-and-drop. Big gap for non-technical users.
- No KB / RAG out of the box. Vapi and Retell ship native knowledge bases. letustalk.ai requires building HTTP tools.
- No marketplace. Vapi has direct Cal.com / HubSpot / Salesforce integrations. letustalk.ai's templates help with prompts, not connectors.
- No public SOC 2 / HIPAA. Vapi has SOC 2; Bland has SOC 2 + HIPAA. Enterprise sales blocker until letustalk.ai closes this.
- No proven 1M concurrency. Bland advertises 1M concurrent calls. letustalk.ai's atomic capacity gate is correct but un-load-tested.
- No branded caller-ID / SHAKEN/STIR. Retell and Vapi have US-specific telephony hardening letustalk.ai lacks (relies on Twilio defaults).
| If you need… | Choose |
|---|
| Voice agents with WhatsApp voice + text in one agent | letustalk.ai |
| US-first regulated industries (healthcare, finance) with SOC 2 + HIPAA | Bland or Vapi |
| Drag-and-drop conversation flows for non-developers | Vapi (Flow Studio) or Retell |
| Lowest-latency English calls with native multi-language | Retell |
| 1M+ concurrent outbound campaigns | Bland |
| Full DIY framework, no managed platform | An open-source DIY voice framework |
| Per-call cost transparency + per-provider health visibility | letustalk.ai |
| Open-source-friendly stack you can self-host | letustalk.ai |
Last verified: May 2026. Open a PR if anything's drifted.