Frequently asked questions

Q: Where is customer data stored?

Wassla runs on Supabase Postgres in the `ap-northeast-1` (Tokyo) region today, with the React app deployed on Vercel `fra1` (Frankfurt) and the Python voice agent on Fly.io `fra1` (Frankfurt). All tenant data is row-level-security isolated at the database layer, every staff write is recorded to an append-only audit table, and customer data is never hard-deleted — deletions use a hide-and-backlog pattern so audit trails stay intact. EU-resident, UK-resident, and KSA-resident single-tenant deployment options are available on the roadmap for enterprise customers with hard data-residency constraints; talk to hello@wassla.io if region pinning is a hard requirement for your business and we will scope it.

Wassla is an AI customer-support platform for high-volume customer teams worldwide that unifies WhatsApp, Instagram, Facebook Page, SMS, voice calls, and an embeddable web widget into one shared inbox staffed by named AI agents. WASSLA LTD is UK-incorporated (England and Wales). Bilingual EN + AR UI is built in. This page answers the twenty most common questions we hear from new operators, in one place, with concrete links into the deeper docs.

Is Wassla for the Saudi market only?

No. Wassla is a global product. Bilingual EN + AR UI is built in for any business that supports Arabic-speaking customers, but our channels and infrastructure work for any market. WASSLA LTD is UK-incorporated (England and Wales), our compliance posture covers UK GDPR + EU SCCs + Saudi PDPL, and the platform is designed for high-volume customer teams anywhere in the world.

Does Wassla actually support Arabic, or is it bolted on?

Arabic is a first-class language across the entire product, not a translation layer. The UI ships with verified English and Arabic parity on every key, the layout is right-to-left at the CSS level using logical properties, and agents reply in the customer's language without configuration. The AI models we route to (OpenAI and Anthropic frontier models) handle MSA, Saudi and other Gulf dialects, and English code-switching natively. You can switch the interface language from the user menu at any time, and the staff and tenant audit pages render correctly in both directions.

Can your AI agents handle real phone calls, not just chat?

Yes. Voice is a first-class channel. Inbound calls on a Twilio number, or on a WhatsApp Business number with WhatsApp Calling enabled, are answered by a LiveKit-backed voice agent that greets the caller, transcribes both sides in real time, generates a reply, speaks it back, and handles natural turn-taking including barge-in. The voice agent shares the same persona and knowledge base as your text agent. See Voice agents vs text agents for the full breakdown.

What channels can Wassla connect to today?

Six inbound and outbound surfaces are supported in production: WhatsApp Business Cloud API (text and voice on the same number), Instagram Direct Messages, Facebook Page messages and comment-replies, Twilio SMS, Twilio Voice, and an embeddable web widget you drop into any site with a single script tag. All six channels feed the same unified inbox, write to the same conversation table, and bill against the same credit balance. One agent can answer across every channel or you can dedicate one agent per channel.

How much does Wassla cost?

Pricing is usage-based, not seat-based. New workspaces start free on the pre-live tier where you can wire up channels, train the agent, and run internal tests with AI spend gated off. When you go live you pay for credits that are consumed per AI reply, per voice minute, per knowledge embedding, and per OCR page. Every metric is priced at a documented 5x markup over real provider cost and the margin audit is exposed to operators. Read Wassla billing and plans for the current rates.

Is there a free trial?

There is no time-limited trial — there is a permanent free pre-live tier. Every new workspace defaults to infra_only billing mode where all AI spend is gated off at the server before any LLM, Whisper, or embedding fetch can run. You can connect channels, invite your team, hire and train agents, and walk the full product without burning credits. When you are ready, you flip the workspace to demo (limited monthly allowance) or live (production) from the Billing page.

Where is customer data stored?

Wassla runs on Supabase Postgres in the ap-northeast-1 (Tokyo) region today, with the React app deployed on Vercel fra1 (Frankfurt) and the Python voice agent on Fly.io fra1 (Frankfurt). All tenant data is row-level-security isolated at the database layer, every staff write is recorded to an append-only audit table, and customer data is never hard-deleted — deletions use a hide-and-backlog pattern so audit trails stay intact. EU-resident, UK-resident, and KSA-resident single-tenant deployment options are available on the roadmap for enterprise customers with hard data-residency constraints; talk to [email protected] if region pinning is a hard requirement for your business and we will scope it.

Does Wassla integrate with my existing helpdesk?

Direct integrations with Zendesk, Freshdesk, Intercom, and HubSpot are on the roadmap and not yet shipped. In the meantime, Wassla exposes a webhook surface (tenant_webhooks) that fires on conversation create, message create, ticket create, and handoff events, plus a REST API gated by tenant API keys. Most teams that need to mirror conversations into an existing helpdesk wire it up through those webhooks. If a deeper native integration would unblock you, file a request from the Support page so we can prioritize it.

Can I white-label Wassla for my own customers?

White-label is on the roadmap and not yet generally available. Today every workspace renders with your own logo and a primary brand color on the agent surfaces customers see (the web widget, transactional emails, and the invite acceptance page), but the parent product chrome is Wassla-branded. Reseller and full-white-label tiers are planned for enterprise. If you need this before it ships, email [email protected] to be added to the design partner list.

Can I bring my own AI model?

The platform is model-agnostic at the routing layer — we currently route to OpenAI and Anthropic frontier models for replies and Groq Whisper-large-v3 for transcription with an OpenAI fallback. Tenant-level "bring your own model" (a custom OpenAI org key or an Azure OpenAI deployment per workspace) is on the roadmap. Self-hosted open-source model support (Llama 3.x, Mistral) is being evaluated for enterprise. If you have a specific model requirement, please reach out so we can sequence it correctly against the roadmap.

Do you offer on-premises or private-cloud deployment?

The standard offering is multi-tenant SaaS on Supabase (ap-northeast-1, Tokyo) and Vercel + Fly.io (fra1, Frankfurt). We do not ship an on-prem installer today. For regulated customers with a hard on-prem, sovereign-cloud, or specific-region requirement (EU, UK, KSA, GCC, US, or any other jurisdiction), a single-tenant private deployment with region pinning is available as a custom engagement — contact [email protected] with your data-residency and compliance constraints and we will scope it. Most teams find the SaaS deployment with RLS isolation, append-only auditing, and the published sub-processor list sufficient for their compliance posture.

How does pricing on credits actually work?

Every billable action consumes a documented number of credits. A text AI reply costs 3 credits, a voice minute costs 35 credits, OCR pages cost 3 credits each, and knowledge storage is metered monthly per GB. The full list lives in the credit_costs table and is surfaced in the Billing page. Credits are debited atomically through a sharded RPC, so you never lose accuracy under load, and every debit writes an immutable ledger row that powers your usage report and the month-to-date demo allowance calculation.

Will Wassla replace my human team?

No. Wassla treats the AI agent as the first line of response and your humans as the escalation path. The agent handles tier-one volume — order status, store hours, return policy, basic troubleshooting — and routes the rest to your team based on the escalation rules you configure. Human takeover is one click in the inbox; the agent silences itself until you release the conversation. The goal is to free your team from repetitive work, not to remove them from the loop.

How fast is the AI reply?

Text replies stream over Server-Sent Events as the tokens are generated, so the customer sees the first words within roughly one to two seconds. Voice replies typically begin within a second of the customer finishing their turn, with barge-in supported so the customer can interrupt mid-sentence. End-to-end latency depends on the channel, the model you route to, and the size of the knowledge base, but every site has been load-tested from the fra1 region against the production database.

How do I train the AI agent on my business?

You feed it a knowledge base. Upload PDFs, paste website URLs that we crawl for you, or paste structured Q&A, and the content is chunked, embedded with OpenAI 1536-dim embeddings, and stored in pgvector. The agent searches that index on every reply and grounds its answer in your content. You can also write a system prompt that frames tone, role, and policy, and you can add prompt sections (refund policy, escalation rules, business hours) that compose into the final prompt. See Train your AI agent for the walkthrough.

What languages does the AI understand?

The frontier models we route to are strong in English and Modern Standard Arabic, fluent in Saudi dialect and Khaleeji variants, and competent in most major world languages. Customers writing in code-switched Arabic-English are handled correctly without configuration. If a customer writes in French, Urdu, or another major language, the agent will reply in the same language as long as your knowledge base contains the relevant facts.

Can I bring my own WhatsApp number?

Yes. The self-serve flow connects an existing WhatsApp Business Cloud API number through Meta OAuth in a few minutes — open Channels, click WhatsApp, and follow the prompts. If your number is currently on the WhatsApp Business App (the consumer app) rather than the Cloud API, you will need to migrate it through Meta Business Manager first. See Connect WhatsApp to Wassla for the step-by-step.

What happens when I run out of credits?

You see a credit balance widget at the top of the Billing page and you can configure alert thresholds under Settings > Alerts. When the balance hits zero on a live workspace, new AI replies return an out_of_credits error code (HTTP 402) at the server, and the inbox surfaces the block to your team. Inbound messages still land in the queue and your humans can reply manually. You top up by purchasing more credits from the Billing page through Paddle checkout.

Is my data used to train AI models?

No. Wassla does not use customer conversations to train models, and the upstream providers we route to (OpenAI and Anthropic) are configured on their API tiers that exclude content from training by default. Knowledge base content stays inside your tenant; nothing is shared cross-workspace. Audit events are tenant-scoped and isolated at the database row-level-security layer.

How do I manage who on my team can do what?

Wassla has five tenant roles — owner, admin, finance, support, and auditor — plus a separate staff portal for Wassla operators. Owners and admins can invite teammates and assign roles from Settings > Team. Sensitive staff actions above defined money or impersonation thresholds require dual-control approval, and every staff action is written to an append-only audit log visible to the workspace owner. See Team roles for the role matrix.

How do I get help if I am stuck?

Search this help center first — every article is text-first and citable. If you cannot find your answer, reach the Wassla team through the Support page — signed-in customers can open a ticket directly from inside their workspace, and for general inquiries you can email [email protected]. We answer in English and Arabic within one business day. For urgent production issues on a paid workspace, the Billing page lists the escalation contact for your plan.