Finds the passages that actually answer this, with citations
Operations · 4 agents
Support Desk
installed in 72 hours.
Every question triggers a parallel retrieval across your docs and account data, then a drafter and a separate confidence judge decide whether to answer or escalate. Escalations arrive summarized, with what was already tried.
$2,500 install $400 a month after 30 days Cancel anytime
Jamie / operations systemSupport Desk
Live Trigger
Email, 2:14 AM: "How do I change the card on file, and will I lose my annual discount?" Doc retriever and account reader run in parallel.
Output
You can update the card under Settings, then Billing, then Payment method. It takes about a minute. Your annual discount stays attached to the subscription, not the card, so nothing changes. Source: Billing FAQ, section 3. Anything else?
Confidence judge cleared it: both facts had full source coverage. Answered in 40 seconds, seven hours before anyone opened the inbox.
4agents in the system
2run at the same time
3ways it fires itself
1approval gate you control
The agent graph.
This is what you are buying. Not one prompt in a loop: a trigger that fires without you, agents that run at the same time, a chain that only starts once they all report, and one gate where the work meets a person.
Goes off Fires on every question
Including the ones that arrive at 2 AM in a timezone you do not staff.
- New ticket, email, or chat message
- Documentation change triggers re-index
- Weekly gaps report
In parallel Two retrievals at once
Policy and account state are different sources; asking them separately is what makes the answer specific.
Doc retriever Account reader Reads this customer's plan, order, or status, read-only
In series Then, in order
The judge is a separate agent from the drafter. A model grading its own answer is not a control.
Drafter Writes the answer from retrieved sources only, citing them
Confidence judge Decides send or escalate based on source coverage, not on tone
Approval gate Where it meets you
Escalation is a success state, not a failure. The judge is tuned to escalate when in doubt.
- Auto-answer: topics you allow, with full source coverage
- Always route to a person: billing disputes, cancellations, anything legal or medical
- Immediate escalation: negative sentiment, or the customer asking for a human
Hands off Then it closes its own gaps
The weekly gaps report is how the resolution rate goes up without anyone rewriting the docs blind.
- Escalations arrive summarized with what was already tried
- Weekly list of questions your docs do not answer
- Re-indexes automatically when docs change
What fires it.
- A ticket, email, or chat message arrives
- A doc changes and re-indexes itself
- Weekly gaps-report schedule
Your tools, not ours.
- Help docs in Notion, Drive, or your site
- Shared inbox, Intercom, Zendesk, or Slack
- Order or account system, read-only
Where it stops and asks.
You choose which topics can be auto-answered and which always route to a person: billing disputes, cancellations, anything legal or medical. Saying "human" routes immediately, every time.
Best for
- Software companies
- E-commerce
- Membership businesses
- Property managers
- Schools and courses
Illustrative time math
The arithmetic, labeled as a scenario.
Every number here is illustrative. On kickoff we replace it with your inquiry volume, quote value, list size, or ticket count, and the system reports observed results after that.
- Formula
- 300 tickets a month × 55% resolved × 6 minutes each
16.5 hours a month backResolution rate depends on how well your docs cover the questions. The gaps report closes that over time.
What you receive.
- Knowledge ingestion and auto re-index
- Tone guide
- Escalation rules and topic allowlist
- Weekly gaps report
- Simple dashboard
$2,500
- 90-minute kickoff: accounts, voice, rules, and approval boundaries
- Triggers wired to your forms, inbox, calendar, and tools
- Every agent in the topology configured on your real history
- A dry run on past cases before anything reaches a customer
- An approval desk you can run from your phone
- Live within 72 hours of kickoff
$400 after 30 days
- Hosting, scheduling, and the trigger watchers, 24/7
- Claude usage up to 1,000 agent runs a month
- Rule, prompt, and routing changes whenever you ask
- A 30-minute tuning call every month
- Weekly report: what ran, what it decided, what it escalated
- Cancel anytime; the bots run to the end of the paid month
It hands off to these.
Bots chain. Support Desk finishes its run by handing the result to the next system, so one trigger can carry a customer from first contact to a paid, reviewed job with no one retyping anything.
InstalledSupport Desk
ThenInvoice ChaserA ledger watcher with its own escalation clock, and a classifier that knows a dispute from a delay.
ThenReview EngineA watcher on your listings plus a timed asker on your job list, running on two independent clocks.
Fleet of 3$6,000 install, then $900 a monthThree chained systems installed together. Saves $1,500 up front and $300 a month against three separate installs.
Questions about Support Desk
Plain answers.
Will it make things up?
The drafter can only use retrieved passages, and a separate judge checks source coverage before anything sends. No coverage means escalation, not a guess.
Can customers reach a human?
Always. Saying "human" or "person" routes immediately, and the bot says so up front if asked.
What about angry customers?
Sentiment triggers escalation regardless of confidence. The person gets a summary and a suggested first line.
Install Support Desk
Buy the system.
Keep the desk.
One install for one business. The managed desk starts after 30 days, so the first month has one job: prove this system deserves to stay.
Today$2,500installation
thenAfter 30 days$400per month · cancel anytime
- Live within 72 hours of kickoff
- All 4 agents tuned on your history
- Triggers wired to your tools
- Approval desk on your phone
- Monthly tuning call
- Direct founder support