Build your own AI SDR with an AI agent

You can assemble a working AI SDR by connecting WarmySender to Claude, ChatGPT, Cursor, Codex, OpenClaw or Hermes Agent — or any agent that speaks MCP. In plain language it finds leads in the built-in business database, verifies them, builds a multi-step campaign and launches it, then reports on how it did. What it never does is send a single message itself or lift a limit. WarmySender's scheduler paces every send, and the deliverability ceiling every mailbox provider enforces — a spam-complaint rate that must stay under 0.3% — is what really governs how fast an SDR can grow.

Think of it as a division of labour. The agent is excellent at the slow, clicky parts — researching a list, cleaning it, wiring up a sequence, pulling a weekly report. The moment anything is due to actually leave a mailbox, it is WarmySender that decides what goes and when, using the same daily caps, gradual warmup ramp and sending window it applies when you press the buttons yourself. That line is the whole reason an "AI SDR" built this way is safe to run.

What does an "AI SDR" actually do here?

A human SDR turns a target market into booked conversations: they build a list, clean it, write the outreach, send it on a schedule, watch the replies and adjust. An AI SDR assembled on WarmySender does every part of that except the sending and the reply-reading — those two stay with the platform and with you. So the honest description is: your agent runs the pipeline, and WarmySender runs the sending.

The pipeline has five stages, and your agent can carry out all five from a single conversation: find leads, verify them, build the campaign, launch it, and report on it. The table below is that pipeline, stage by stage.

What can the agent do at each stage?

The AI SDR pipeline: what your connected agent can and cannot do at each stage.
Stage Your agent can Your agent cannot
Find leads Search the built-in business database by role, company, industry, size and location, save what it finds to a list, and export it — searching and filtering are free, and a credit is spent only when a new lead is saved Spend more of your lead allowance than the daily pace allows, or promise that a saved address will accept mail today
Verify Check a single address on the spot or submit a whole list, read the results back, cancel a running batch, check how much allowance is left first, and turn the confirmed addresses into a clean list Verify faster than your allowance and the same safe pacing permit, top up its own allowance, or discover an address you do not already have
Build the campaign Create a multi-step campaign, work from your saved templates, set the sending window, choose which mailboxes send, add and enroll prospects one at a time or in bulk, and configure warmup on those mailboxes first Set a daily limit above what your plan and your mailboxes allow, or add anyone past your plan's contact limit
Launch Start the campaign, pause it, resume it — and enrolling a large list is safe, because starting only hands the campaign to the scheduler Send any of the messages itself, or make them go out any faster than your safe daily pace
Report and adjust Answer "how did this do", "who replied", "which mailboxes look weak this week" and "why did this send nothing yesterday", then pause, re-enroll or build a follow-up Read a reply's contents or send a reply — reading intent and replying stay in the app, with you

Notice the shape of the "cannot" column: every entry is either "send" or "raise a limit" or "handle a credential" or "read a reply". Those are not switches left off — there is simply nothing an agent can call that does them.

Who actually sends the emails?

WarmySender does, and only WarmySender. When your agent starts a campaign, all that happens is the campaign is written down and handed over. From then on the scheduler decides what leaves and when: it spreads the day's sends across your sending window, respects each mailbox's daily limit, keeps the whole domain under its ceiling, honours your plan's allowance and skips anyone on your suppression list.

This is what makes an aggressive-sounding instruction safe. "Find 2,000 founders, verify them, and email them" does not produce 2,000 emails — it produces a clean list and a queue that drains at your normal pace, and a queue you can pause with one sentence. The agent's ambition and the account's safety live on opposite sides of that line, on purpose.

Why can't my AI SDR just send faster?

Because the real constraint on cold outreach is not how fast a tool can fire — it is a number the mailbox providers watch and you do not control: your spam-complaint rate. Google, Yahoo and the other providers ask every sender, at any volume, to keep that rate below 0.3%, and Google recommends staying under 0.1%. Cross it and your mail starts landing in spam no matter how clever the sending schedule is. That ceiling, not the agent, is what caps how quickly an SDR can scale.

Everything the safe design forces is aimed at staying under that ceiling:

For the actual per-mailbox and per-domain volume numbers, see the canonical table at how many cold emails you can send per day. The short version: 30–50 a day per warmed mailbox, and you grow by adding mailboxes and domains.

How do I set one up?

  1. Connect an agent. Create a key in Settings → API Keys (it starts with ws_) and add the WarmySender server to Claude, Cursor, Codex or ChatGPT. The step-by-step is in connect your AI agent, and the ChatGPT and Codex guide covers the one client with per-plan differences.
  2. Start read-only if you want to. When you create the key you choose exactly which permissions it carries, so a read-only key lets your SDR research and report for a week before you let it change anything.
  3. Connect at least one mailbox in the app first. Your agent works with the mailboxes you have already connected — it cannot connect one itself, because a mailbox credential should never pass through an agent.
  4. Then just talk to it. The whole pipeline is one conversation; the prompts below are a starting point.

What should I tell it?

Frequently asked questions

Can I really build an AI SDR with an AI agent?

Yes, for everything except the sending and the reply-reading. Connect WarmySender to Claude, ChatGPT, Cursor, Codex, OpenClaw or Hermes Agent, and it can find leads, verify them, build and launch campaigns, configure warmup and report back — the full pipeline, in plain language. WarmySender's scheduler does the actual sending, and reading and replying to responses stays in the app with you.

Can my AI SDR send the emails on its own?

No. It builds and launches the campaign, and starting a campaign only hands it to WarmySender's scheduler, which paces every message inside your sending window, your per-mailbox daily limit, your domain's ceiling and your plan's allowance. Enrolling thousands of people is safe: it creates a queue that drains at your normal pace, not a burst.

Can my AI SDR raise my sending limits to hit a target?

No, and there is no way for it to ask. Daily caps, warmup stages and domain-wide ceilings cannot be moved by anything an agent can call. It can set a campaign's daily limit, but the number that actually applies is still the lowest of your mailbox limit, your mailbox's combined warmup-and-cold budget, your domain's ceiling and your plan's allowance. The real ceiling on growth is your spam-complaint rate, which every provider asks you to keep under 0.3%.

Does the AI SDR verify leads before sending?

It can, and you should have it do so on every list. Ask it to verify the leads it found and build the campaign only from the addresses that came back safe to send. Verification keeps your bounce rate down, and bounces are one of the fastest ways to push your complaint rate toward the 0.3% ceiling that gets mail filtered. Leads from the built-in database are a starting point, not a promise that a mailbox will accept mail today.

Can my AI SDR connect my mailbox or buy leads for me?

No. Connecting and disconnecting mailboxes stays in the app, so no credential ever passes through an agent, and anything that touches billing — buying lead credits or verification credits — also stays with you. Your agent works with the accounts and allowance you already have.

What if my AI SDR gets something wrong?

Far less happens than you would fear, because nothing goes out in a burst. A mis-briefed campaign leaves at your normal daily pace, so there is time to notice; pausing is one sentence and is never blocked; and everything your agent built is a normal campaign you can open, edit or delete. Start it on a read-only key if you want to watch it work before you let it change anything.

Can I give my AI SDR read-only access first?

Yes. When you create a key in Settings → API Keys you pick exactly which permissions it carries, so a read-only key lets your SDR research leads, report on campaigns and read verification results while being unable to create, change or launch anything. You can revoke or replace the key at any time and access stops immediately.

Building your first agent-run pipeline and want a second pair of eyes? Email [email protected] — we respond within 24 hours on business days.