Docs menu
Answer email in the app
Flinkk Mail can run an agent for you, with no code and no API key. When a new email arrives in one of the agent's inboxes, it reads the thread, follows your instructions and decides: write a reply, hand the email to a person, or leave it. Replies wait for a person to approve them unless you turn on auto-send for that inbox.
If you want to connect your own agent (Claude, ChatGPT, Cursor or your own code) instead, see the agent inbox API and the MCP server. Both can be used on the same inbox.
Turn it on
- Open Agents and create an agent, or open an existing one.
- Under Inbox access, grant it an inbox with at least Read and Draft. Keep Draft only if a person should approve every reply (recommended to start).
- In Answer new mail in the app, write the instructions and switch it On.
From then on, every new email in that inbox is handled within a few seconds. Only owners and admins can change these settings. AI must be configured for your workspace for the agent to run.
Write good instructions
Say who the agent is, how it should sound and where its limits are. Short and specific works best.
You answer questions from guests about our hotel.
- Be warm and brief. Sign off as "The Front Desk team".
- Check-in is 2pm, check-out 11am.
- For refunds, complaints or anything you are unsure about, hand it to a person.Per-inbox instructions are optional and are added to the general ones for mail in that inbox only, for example a different sign-off for bookings@ than for hello@. Open the agent, find Per-inbox instructions and choose Add instructions next to the inbox. An agent can run on per-inbox instructions alone.
What it knows: FAQ and templates
FAQ. Write down the facts the agent may rely on: prices, opening hours, policies, answers to common questions. The agent is told not to state facts that are in neither the FAQ nor the email thread, so a good FAQ is the best way to get accurate replies. Up to 20,000 characters.
Templates. Pick up to 20 of your templates, yours or built-in, that the agent may reply with. When one fits better than a written reply (a booking confirmation, a price list), the agent fills in the template's variables from the email and the FAQ. The template keeps its design and values are inserted as plain text. If a required variable is missing, the agent hands the email to a person instead of sending a half-filled template. The reply stays in the thread with a Re: subject.
Test it before you switch it on
Test it on an email runs the agent on a pasted email, using the instructions, FAQ and template choices exactly as they are on the screen, even before you save them. If the agent has more than one inbox, choose the one the email arrives in. Nothing is drafted, sent or stored. Try a few real questions, including ones it should hand off.
What it does with each email
| Outcome | What happens |
|---|---|
| Reply | A reply is drafted in the thread. In Draft only it waits in Agents > Approvals and at the top of the thread; in Auto-send it is sent, within the inbox's daily limit. |
| Reply with a template | Same as a reply, with the template rendered and filled in. |
| Hand to a person | Nothing is drafted. The thread gets a Needs a person label (if the agent may label in that inbox) and the reason is shown under Recent activity. |
| Skip | Nothing happens. Used for newsletters, notifications, receipts and messages that need no answer. |
Recent activity on the agent's page lists every email it handled and what it did. The Agents item in the sidebar shows how many replies are waiting for your approval.
Safety
The agent works under the same rules as any agent connected through the API, enforced on the server:
- Draft only (the default): a person approves every reply.
- Reply only (the default): it answers people already in the thread, nothing else.
- Daily send limit per inbox, and at most three replies per thread per day.
- It only sees the inboxes you granted, and only the messages in a thread that involve those inboxes.
- It never answers your own workspace's addresses or automated senders such as
no-reply@, so two inboxes cannot loop. - Email content is treated as untrusted: instructions written inside an email are ignored.
- Pause the agent to stop it immediately. Every action is in its audit log.
On this page: Turn it on · Write good instructions · What it knows: FAQ and templates · Test it before you switch it on · What it does with each email · Safety