AI customer agents

The problem

Small businesses want AI to handle routine customer conversations, but they need to control what the agent says, use their own model accounts and get a clear outcome from each chat rather than a transcript to read.

What it does

• Accounts with email confirmation, password reset and lock-out, one workspace each, isolated in the database
• Agents from templates (order confirmation, cart recovery or blank), published as frozen versions
• A streaming chat engine with tool use, and a live test chat beside the agent settings
• Bring your own model keys: Anthropic, Gemini or any OpenAI-compatible service, stored encrypted
• Conversation transcripts and 30-day numbers
• Shareable chat links and a website widget
• An events API, signed webhooks and a staff inbox for hand-off
• A knowledge base with Bangla search, and test runs before publishing
• WhatsApp, then voice agents