For founders

An AI SDR for founder-led outbound

You are the SDR, the closer and the founder. AgentSDR runs email, LinkedIn and WhatsApp from one workspace, and the AI reads every reply so the queue you work each day is short and already drafted.

Star on GitHub Self-host it

agentsdr.your-company.com/crm/actions

Action required

Every reply the AI has classified, with the answer it drafted. Approve, edit or reclassify.

In the queue

replies and follow-ups due

Drafts to review

written by your model

Follow-ups due

past their due date

Errors

nothing failed today

Queue

Newest reply first. Hover a reply or a draft to read it whole.

The problem

Founder-led outbound breaks in the gaps

Doing it yourself is the right call early. What hurts is the glue: tools that do not talk to each other and replies nobody sees in time.

The patchwork

A sequencer for email, a separate tool for LinkedIn, another for WhatsApp
Replies in three inboxes, found when you remember to look
A spreadsheet that is the only record of who was contacted where
A per-seat bill for a team of one, plus per-contact limits

One workspace

One workspace for email, LinkedIn and WhatsApp campaigns
One Action required queue, every reply classified with a draft attached
One lead record per person, whichever channel they answer on
Free software; you pay your server, your accounts and your AI usage

A day in the workflow

Sequences run. You answer the replies.

This is what a day looks like once campaigns are live: a few hours of waiting, then a short queue with drafts already written.

09:00
Sending hours open
The first email goes out. The next from this mailbox follows 18 to 24 minutes later.
10:40
Dana Whitfield, Ledgerly, replies
Information Requested
“Can you send pricing?” The remaining emails in her sequence are cancelled.
10:41
Draft ready in Action required
Classified and answered from your knowledge base. Nothing sends until you approve.
13:15
Tomas Reyes accepted your invite
The first follow-up is queued inside his account's working hours.
15:30
Aisha Khan, Brightpath, replies
Meeting Requested
“Tuesday works for a call.” Her email and LinkedIn steps stop.
16:05
You work the queue
Review the drafts, edit or approve them, and send.

Action required

Queue empty. Sequences keep sending on their own.

Done for the day

An illustration with sample data. Sending hours default to Monday to Friday, 09:00 to 18:00.

The workflow

Triage that keeps replies from piling up

The part of outbound that eats a founder’s week is not sending. It is reading, sorting and answering.

AI triage

Every reply classified, every answer drafted

The AI puts each reply in a category you control, records how sure it is and why, and writes a draft using your instructions and knowledge base. You see a ranked queue, not a pile of threads.

Interested, Customer, Not interested or Other, with subcategories
Applied on its own only above 85% confidence by default
Do Not Contact is honoured on every channel at once

Fast to clear

Clear the queue in one sitting

Open a row, read the lead’s message beside the draft, and approve it as written or edit it first. The reply goes out from the account the conversation is on.

Approve & send, or edit and send
Switch channel and the reply is drafted separately
Nothing is sent without a person approving it

One workspace

Three channels, one lead record

You do not run three outbound programmes. You run one, and pick the channel per step.

Email from your own mailboxes
Multi-step sequences from Google Workspace, 30 a day per mailbox by default, inside sending hours.
LinkedIn invites and follow-ups
Paced runs with random gaps, per-account daily limits and working hours.
WhatsApp messages and calls
Message campaigns with a warm-up period, and recorded calls placed from your own number.
A lead database for all of it
Import a CSV or enrich in a table. A person is one record whichever channel they answer on.
One inbox for replies
Email, LinkedIn and WhatsApp conversations in a single list, with drafts ready.
Stops when they answer
A reply on any channel ends that lead’s other sequences, so nobody gets a follow-up after replying.

What it costs

Free software. You pay for what you connect.

AgentSDR is open source under the AGPL-3.0. There are no seats, tiers or per-contact fees, so a team of one pays the same as a team of ten: nothing for the software.

per seat, per contact or per mailbox: $0
channels in one workspace: email, LinkedIn, WhatsApp: 3
Action required queue for every reply: 1
Docker Compose services: db, setup, app, cron: 4

What you pay for when you self-host AgentSDR
You pay for	What it is
Your server and PostgreSQL	One app process and one database; Docker Compose starts both.
Google Workspace mailboxes	The mailboxes you already own, for cold email.
Unipile	The LinkedIn and WhatsApp accounts you link.
OpenRouter usage	Your own key, for every AI step. Billed to you, by the model provider.
Resend	Sign-in, password-reset and invitation emails (required in production).

Your first week

From clone to first reply

01
Run it
Start the app and PostgreSQL with Docker Compose, sign up and create your organization.
02
Connect your accounts
Google Workspace, Unipile and your OpenRouter key, each from its own guided Settings page.
03
Add leads, write sequences
Import a CSV or build a table, then write an email, LinkedIn or WhatsApp sequence with merge fields.
04
Work Action required
Replies arrive classified with a draft. Approve, edit or change the category, and the lead moves on.

FAQ

Questions, answered

Can one person really run outbound with AgentSDR?

That is the case it is built for. Sequences send on their own inside sending hours, replies from every channel land in one Action required queue, and the AI classifies each one and drafts an answer. Your job is to review drafts and send them, not to chase tabs.

What does an AI SDR for founders cost?

The software is free: it is open source under the AGPL-3.0 and you host it yourself, so there are no seats, tiers or per-contact fees. You pay for your own server, the accounts you connect (Google Workspace mailboxes, Unipile for LinkedIn and WhatsApp) and your own AI usage on OpenRouter. A Resend account is needed to email sign-in and invitation messages in production.

Will the AI send messages without my approval?

No. The AI classifies replies and drafts answers, but nothing is sent until a person approves it. It applies a classification on its own only when it is confident (85% by default), and holds anything else as a proposal for you to accept or change.

Do I need a team or a developer to set it up?

You need to run one app and one PostgreSQL database, which the repository's Docker Compose file starts for you. After that, every service is connected from the Settings pages with guided setup: Google Workspace, Unipile, OpenRouter and Cloudflare R2. See the open-source page for the exact steps.

Is it safe to run LinkedIn and email from my own accounts?

No tool can promise an account is never restricted, but the defaults are conservative: 30 emails a day per mailbox with an 18 to 24 minute gap, 30 LinkedIn invitations a day on Premium accounts and 5 on free ones, and sending only inside working hours. All of it is adjustable in Sending rules, with a warning when you pass the safe edge.

What happens when a lead replies on a different channel?

A person is one record across channels. A reply on email, LinkedIn or WhatsApp marks their campaign enrollments as replied, so you stop writing to someone who has answered, and the reply reaches Action required.

Be the whole SDR team, without the glue

Self-host it, connect your accounts and let the AI keep your reply queue short. Free software, your data, your keys.