It answers from your docs and the customer's actual account, fixes billing and access itself, files real bugs with repro, and follows up until it's closed.
It reads the customer's account before it answers, fixes billing and access within your limits, files real bugs to engineering with repro steps, and follows up until the customer says it's solved.
SaaS support has a shape the generic chatbots miss. Half the tickets aren't questions — they're states: the card declined and exports stopped, the invite never arrived, the API key got rotated, the webhook silently stopped retrying. A bot that only reads your help center can explain what a webhook is. It can't see that this workspace's endpoint has been returning 502s since Monday. Fasrad's support agent can, because it looks at the customer's account — plan, seats, billing status, recent errors — from your records or straight through your API, before it answers.
Then it does the thing. A declined card gets a payment link sent to the billing contact and the account un-paused when it clears. A missing invite is resent. A 'how do I' gets the answer from your docs with the exact settings path — and a follow-up two days later asking if it worked. A real bug gets filed to engineering with the workspace, the logs, the failing payload, and a repro — and the customer is told who has it and when they'll hear, not 'we'll look into it.' It runs in your support inbox and as chat on your site or inside your app, and it's the same agent with the same memory in both.
Tickets it closes without a human, and the ones it routes:
Everything it learns stays with the account. Every ticket is filed in the built-in CRM against the workspace — category, summary, who wrote in, what was done — so the third ticket from the same customer reads like it came from someone who remembers the first two. Rules keep it inside your lines: 'credits over $200 ask me,' 'enterprise accounts always copy their CSM,' 'never promise a ship date.' When a rule fires you get the prepared action and a button on Telegram. And when Thursday comes and the customer hasn't heard about their bug, the agent chases — them and your engineer — because 'closed' means the customer said so.
These pages cover the adjacent jobs buyers usually compare before choosing an AI agent.
It looks it up. Point it at your customer records — a datastore it keeps, a sheet you export, or your own admin API — and it reads the account before it answers: plan, seats, billing state, recent errors, open tickets. That's the difference between explaining what a paused export means and telling the customer their card was declined on the 12th and here's the link to fix it.
Fix them, inside limits you set. It sends payment links, resends invites, resets 2FA through your recovery flow, changes seats and plans, and reissues invoices with the right tax ID — through your API or your billing provider's. Anything over a line you wrote — 'credits above $200 ask me' — it prepares and holds for your tap.
It works out whether it is one — asks for what it needs, tries to reproduce from the account's logs — then files it to your tracker with the workspace, the failing payload, and steps, and pings on-call on Telegram. The customer is told who has it and when they'll hear back, and the agent follows up with both sides until the fix ships. No ticket goes quiet.
All of them, as one agent. Drop the chat embed on your site or inside your app, give it your support inbox, and it answers the same customer in either place with the same memory. Telegram and phone are the same agent too.
It answers from your docs and the account in front of it, and it's told to say 'that's not something we do yet' rather than invent. You can tell it what's on the roadmap and how to phrase it. Every answer it gives is logged, so you can review and correct — and a correction becomes a rule it remembers.
Yes — Zendesk, Intercom, Help Scout, Linear, Jira, GitHub Issues, through their APIs. It can read and post tickets there, or run standalone with its own CRM and ticket log if you'd rather not pay for seats. Slack and Telegram for the on-call ping either way.
$49/month or $490/year — cancel anytime. Flat — no per-resolution fee, no per-seat charge, no conversation cap, so a good month for signups isn't a bad month for the support bill. It's in public beta and takes about four minutes to connect.
One support agent that resolves tickets instead of deflecting them — on email, web chat, and inside your product — with plain-English rules for what asks you first.