Frequently asked.
Can't find what you need? The Help Center goes deeper, or talk to a human.
Pricing & billing
How does outcome-based pricing actually work?
You pay a flat rate (default $0.80) per Resolved Outcome Unit. A ROU only counts after it clears every quality gate — at least two turns, confidence above the floor, no reopen within 72 hours, and a 7-day CSAT that isn't negative. Anything that fails a gate is never billed.
Is there a free tier?
Every new tenant gets the first 100 ROUs free. After that you only pay for outcomes that resolve — there are no per-seat or per-message charges.
What if a customer reopens the same issue?
The original outcome is automatically voided and credited. You can see every clawback in Billing → Clawbacks, and dispute any charge with its full evidence trail.
Languages & dialects
Which languages do you support?
50+ languages out of the box, with first-class dialect quality for Banglish, Hinglish, Khaleeji and Levantine Arabic — including honorific awareness (apni / tumi / tui). Dialect quality is tuned, not just translated.
Can the voice agent pronounce Bangla numbers correctly?
Yes. Digit strings — amounts, phone numbers, NIDs, dates — are pre-processed through a tool layer before TTS, so the agent never mispronounces raw numerals.
Security & compliance
Where is my data stored?
In the region you choose — EU, US, or APAC at launch. Each tenant's conversations, vector index, and audit logs are pinned to that region to satisfy EU AI Act, India DPDP, and UAE PDPL residency rules.
Can one tenant ever see another tenant's data?
No. Isolation is enforced at the database with Postgres Row-Level Security (FORCE RLS) — not just in application code. Even platform staff can't read across tenant boundaries for most tables.
How do you handle the EU AI Act?
Disclosure (Article 50) renders on every EU end-user channel and can't be disabled. Every autonomous action writes to a tamper-evident audit chain (Article 26). Consent is checked per interaction, not once.
Product & integration
What does "MCP-native" mean?
Fahmix AI is both an MCP server and client. You can expose your Fahmix AI agent to Claude, ChatGPT, or Cursor, and connect the agent to your own tools — Notion, Salesforce, Zendesk, HubSpot — over MCP with OAuth 2.1 + DCR + PKCE.
How long does onboarding take?
An SMB tenant can be live in under an hour. Connect a knowledge base, set a persona, drop the widget on your site, and you're resolving outcomes. Enterprise rollouts are white-glove with checkpoints.
Can I bring my own model?
Fahmix AI runs an all-Google stack by default, with Anthropic Claude and OpenAI wired into the multi-model router and enable-able per tenant — no code changes required.