Myna

Myna is built around the idea that a customer-facing AI agent needs more than a good model — it needs guardrails. The platform separates tool work from customer-facing text: a plan stage handles calls to systems like HubSpot, Stripe, Zendesk, or Shopify, while a composer stage writes every reply from confirmed results with dialable tone controls. Instructions compile into stage-specific prompt modules so you can tune objectives, persona, and policy per workflow step without prompt engineering.

Under the hood, Myna adds deterministic pre- and post-gates. Pre-gates stop opt-outs, human takeovers, after-hours sends, and message caps before the model runs, while post-gates catch false booking claims and prompt leakage. Every tool action has typed inputs and outputs, policy rules, critical flags, idempotency, and throttling, so side effects are explicit and safe in production. The same runtime powers SMS, WhatsApp, email, web chat, and voice, with per-channel compliance, warm-up, daily limits, and throttles built in.

Myna targets service businesses and agencies. Workflow Studio lets teams wire sub-agents, triggers, and operations or start from templates, Memory Hub teaches an agent a business from its website, and Radar supports outreach. Agencies can white-label the platform, resell agents under their own brand, and manage reporting and rebilling from their own domain. It reports scale metrics such as 100 million+ operations processed and a 99.9% uptime target.