AgentSDR
Kandid · AgentSDR: Open-Source, Self-Hosted AI SDR for Email, LinkedIn and WhatsApp Outreach
AgentSDR is a free, MIT-licensed AI SDR you host yourself: it runs email, LinkedIn and WhatsApp outreach from one lead database, enriches lists with your own data-provider accounts, and has an AI CRM that classifies every reply and drafts the answer for a person to approve. It suits founders, agencies and small sales teams who want to own their outbound stack and are willing to connect and pay for the underlying services themselves.
Best for
Technical founders, small sales teams and outbound agencies who want email, LinkedIn and WhatsApp outreach plus AI reply triage in one tool they own, with no per-seat fees, and who can run a server and manage their own Google Workspace, Unipile and model-provider accounts
Not ideal for
Teams that want a managed AI SDR that researches prospects and writes and sends cold outreach on its own, anyone without the time to configure service accounts, webhooks and API keys, Microsoft 365 email users, and organizations that cannot accept the account risk of LinkedIn or WhatsApp automation
Who it's for
Founders doing their own outbound, small B2B sales teams and outbound agencies that want to self-host their sales engagement stack
Capabilities
- AI CRM that reads every reply on email, LinkedIn and WhatsApp, classifies it into Interested, Customer, Not interested or Other plus your own subcategories, and shows a confidence and the reason
- Drafted replies grounded in a versioned knowledge base of pricing, product, scheduling, objection and case-study documents, with a Knowledge used list showing which excerpts each draft relied on
- Action required queue where every draft and follow-up waits for a person to approve, edit, regenerate with feedback or discard
- Pipeline board that the AI keeps current: it moves a lead forward on its own only above a confidence threshold (85% by default) and holds backward moves, Customer and Other for review
- CRM sequences of up to 30 AI-drafted reply and follow-up steps, attached to reply subcategories and cancelled automatically when the lead replies again
- Email sequences from your own Google Workspace mailboxes with merge fields, spin text, mailbox rotation, per-mailbox daily caps and sending windows, one-click unsubscribe and bounce suppression
- LinkedIn campaigns across several accounts through Unipile: connection requests with notes, a message after acceptance and follow-ups, plus LinkedIn search that adds up to 400 leads a day per account by default
- WhatsApp message campaigns with a 24-hour warm-up for new numbers and 25 new chats a day per number by default
- WhatsApp calls dialed from AgentSDR through a Chrome extension, recorded to your own Cloudflare R2 bucket and transcribed by your chosen audio model
- Enrichment Tables with 15 data providers, AI columns that prompt a model per row, and HTTP columns that call any web API per row
- Sending rules per organization with defaults, hard limits, warnings past a safe edge and per-account overrides
- Multi-organization deployments with owner, admin and member roles, teams, Do Not Contact across every channel, and analytics for replies, positive replies, meetings and customers by channel
Limitations
- Self-hosted only. The project describes no shared cloud behind it, and no managed or paid hosted plan is offered
- The software is free, but the services it needs are billed separately by their vendors: Google Workspace for email, Unipile (a paid service) for LinkedIn and WhatsApp, OpenRouter plus your own model-provider keys for every AI feature, Cloudflare R2 for call recordings, and credits on each enrichment provider you connect
- AI features work only in OpenRouter bring-your-own-key mode: you need an OpenRouter account with credits, at least one provider key in OpenRouter's BYOK settings, and shared-capacity fallback turned off, or AgentSDR will not send any model requests
- Email sending supports Google Workspace mailboxes through a service account with domain-wide delegation. Microsoft 365 is not supported yet
- The project's own docs say LinkedIn's User Agreement does not allow automated messaging and that any automation, AgentSDR included, can get an account restricted or banned. WhatsApp also bans numbers that send unsolicited bulk messages; the built-in limits reduce the risk but do not remove it
- Must run as a single app instance, because the outreach sender has no distributed lock and two copies would each send. Webhooks from Unipile and Gmail need a public HTTPS address
- No contact database of its own: leads come from your imports, LinkedIn search through Unipile, or your own enrichment-provider accounts
- First-touch outreach is template-based: the AI does not research prospects and compose cold messages on its own. You can generate per-lead lines with an AI column in Tables and use them as merge fields, but you design that step. Email opens and clicks are not tracked
- WhatsApp calling needs the Call Recorder Chrome extension and relies on WhatsApp Web's English interface, and call transcription is off until you pick an audio model
- A very young project: first public release on October 6, 2026, still pre-1.0, and the project says minor releases may include breaking changes
Use cases
- Running cold email sequences from several Google Workspace mailboxes and working every reply from one queue with AI drafts ready
- Combining LinkedIn invitations and email for the same lead so a reply on either channel lands in one conversation and one pipeline
- Building a prospect list from a CSV, enriching emails and phone numbers with Apollo, Hunter or Findymail, adding an AI column for a pitch angle, and turning the rows into a campaign
- Running outbound for several clients from one deployment, with a separate organization, accounts and sending rules for each
- Calling leads over WhatsApp and getting a recording and transcript attached to the lead's record
Our take
AgentSDR's honest pitch is ownership, not autonomy. It gives you the plumbing of a modern outbound stack, with multichannel sequences, enrichment and a unified reply queue, plus an AI layer that is deliberately conservative: it sorts replies and drafts answers, and a person sends them. That makes it a good fit for founders and agencies who are tired of paying per seat across five tools and are comfortable running a server, but the cost moves rather than disappears. Expect to set up a Google service account, Pub/Sub, Unipile, OpenRouter BYOK keys and possibly R2 before everything works, and to pay those vendors directly. Treat the LinkedIn and WhatsApp channels with the caution the project's own docs ask for, and since the code is only days old, try it on a small campaign before moving your whole pipeline.
Who should use it
Founder-led sales teams, small outbound teams and agencies with some technical capacity who want one self-hosted workspace for email, LinkedIn and WhatsApp outreach, prefer to approve every AI reply, and want to keep lead data on their own infrastructure.
Who should skip it
Teams looking for a hands-off AI SDR that researches prospects and writes first-touch outreach for them, non-technical users without someone to run a server, Microsoft 365 email users, and companies whose policies rule out LinkedIn or WhatsApp automation.
Strengths
- Free and MIT-licensed, with no per-seat, per-contact or per-mailbox fees
- Lead data, conversations and recordings stay in your own database and storage bucket
- Email, LinkedIn and WhatsApp replies share one lead record, one queue and one pipeline
- Clear human-in-the-loop rules: nothing the AI drafts is sent without approval, and backward pipeline moves are held for review
- Model calls are pinned to the provider you choose, with fallbacks off
Weaknesses
- Needs several paid outside accounts and a public server before every feature works
- LinkedIn and WhatsApp automation carries account-restriction and ban risk
- Google Workspace only for email; no Microsoft 365
- Brand new and pre-1.0, with breaking changes already in its first week
AgentSDR pricing
Open source (self-hosted)
Free
- MIT License, including commercial use and hosting for others
- Email, LinkedIn and WhatsApp campaigns, Tables and the AI CRM
- Several organizations, with members, roles and teams, on one deployment
- Community support through GitHub Discussions and issues
Note: AgentSDR charges nothing and adds no markup. Running it means paying your own server plus each connected service directly: Google Workspace for mailboxes, Unipile (a paid service with its own pricing) for LinkedIn and WhatsApp, OpenRouter credits and your model-provider keys for AI, Cloudflare R2 for call recordings, a Resend account if you invite teammates, and credits with each enrichment provider you use. No managed or hosted plan is offered.
Technical specs
Modalities
text, audio
Available models
Where AgentSDR excels
Founder-led outbound without an SDR hire
A founder can run email and LinkedIn sequences from one workspace and spend a few minutes a day in the Action required queue, approving AI drafts that are already classified and grounded in the founder's own pricing and scheduling notes.
Agency running outbound for several clients
One deployment holds a separate organization per client, each with its own mailboxes, LinkedIn accounts, sending rules and data, and no per-seat fees as the client list grows.
Keeping prospect data in-house
Leads, conversations and analytics stay in your PostgreSQL database, recordings in your own R2 bucket, and AI requests go only to the model providers you approve, which matters to teams that do not want a vendor holding their pipeline.
AgentSDR vs. competitors
AgentSDR vs. 11x
11x is a hosted, paid service whose AI SDR Alice researches prospects and writes and runs personalized outreach on autopilot, with 11x's own lead database and mailbox deliverability infrastructure included. AgentSDR is free, self-hosted software where you supply the mailboxes, Unipile account, data providers and model keys, and its AI drafts replies for a person to approve rather than running outreach end to end.
AgentSDR vs. Artisan
Artisan's AI BDR Ava searches a built-in database of 250M+ B2B contacts and writes personalized multichannel sequences that it can send on its own or in approval mode, as a hosted, paid product. AgentSDR has no database of its own, enriches through provider accounts you connect, and keeps all data on your server, at the cost of setting up and paying for those services yourself.
AgentSDR vs. Apollo.io
Apollo combines a large B2B contact database with sequencing, a dialer and AI agents, priced per seat. AgentSDR can use Apollo as an enrichment provider inside its Tables, and adds LinkedIn and WhatsApp channels and an AI reply CRM on your own server, but it does not supply contact data itself.
AgentSDR vs. Clay
Clay is a hosted data and workflow platform for building and enriching prospect lists from 200+ data providers, with AI research agents and a native sequencer. AgentSDR's Tables cover a smaller version of that job with 15 providers you connect yourself, AI columns and no waterfall between providers yet, and sit beside its own email, LinkedIn and WhatsApp sending, reply inbox and AI CRM on your server.
Frequently asked questions
What is AgentSDR?
AgentSDR is an open-source AI SDR that you host yourself. It runs email, LinkedIn and WhatsApp outreach from one lead database, enriches leads in Tables with data providers and AI columns, and puts every reply into an AI CRM that classifies it and drafts an answer for a person to approve. It is one Next.js app and one PostgreSQL database, and the project is led by Kandid.
Is AgentSDR free?
The software is free under the MIT License, with no seats, tiers or per-contact fees. You pay for your server and for the accounts you connect: Google Workspace mailboxes, a Unipile account for LinkedIn and WhatsApp, OpenRouter and your model-provider keys for AI, Cloudflare R2 if you record calls, and credits with any enrichment provider. AgentSDR adds no markup to any of them.
Does AgentSDR send messages on its own?
Outreach campaigns you launch send automatically on schedule, within your sending rules. Replies are different: AI drafts, including every follow-up step in a CRM sequence, wait in the Action required queue until someone approves or edits them. The AI can move a lead forward in the pipeline on its own when it is confident (85% by default), and it marks a person Do Not Contact immediately when they ask to stop.
How do I self-host AgentSDR?
Download the docker-compose.yml from the repository, replace the secrets marked CHANGEME and run docker compose up -d. That starts PostgreSQL and the app on port 3000, creates the schema and runs the scheduled jobs inside the app. You can also run it from source with Bun 1.2+ and PostgreSQL 16+, or deploy it on a VPS with Dokploy. Run only one app instance, and give it a public HTTPS address so Unipile and Gmail webhooks can reach it.
Which AI models does AgentSDR use?
Any model available on OpenRouter, using your own keys in OpenRouter's bring-your-own-key mode. You pick the allowed providers and models, a default model for classification and drafts, and optionally an audio-capable model for call transcription. Each request is pinned to the provider you chose with fallbacks turned off.
Is LinkedIn automation with AgentSDR safe?
No tool can guarantee it. AgentSDR's own responsible-use page says LinkedIn's User Agreement does not allow automated messaging and that accounts using automation can be restricted or banned. Its defaults keep to 30 invitations a day on premium accounts and 5 on free ones, with random gaps and optional working hours, which reduce but do not remove that risk.
AgentSDR vs 11x: what is the difference?
11x sells Alice, a hosted, paid AI SDR that researches prospects and writes and runs personalized outreach for you, with its own lead database and mailbox deliverability infrastructure included. AgentSDR is free software you run yourself, with your own mailboxes, Unipile account, data providers and model keys. Its AI focuses on classifying replies and drafting answers for approval, while first-touch outreach comes from templates you write.
Can I try AgentSDR before self-hosting it?
Yes. The agentsdr.ai website links to a read-only live demo of the app filled with a fictional company's sample data, with no sign-up. To use it with your own leads and accounts you need to install it on your own server; there is no hosted AgentSDR service.
Integrations & fit
Alternatives to consider
About AgentSDR
AgentSDR replaces a stack of outbound tools with one Next.js app and one PostgreSQL database that you run on your own server. You import leads from CSV or XLSX or pull them from LinkedIn search, enrich them in spreadsheet-style Tables with 15 data providers (Apollo, Hunter, Lusha, Findymail, ZeroBounce and others), AI columns and HTTP calls, then add them to email, LinkedIn or WhatsApp campaigns. Campaigns are templates with merge fields (which can include per-lead text generated in an AI column) that send automatically on schedule, paced by sending rules such as 30 emails a day per mailbox and 30 LinkedIn invitations a day on premium accounts. The agentic part starts when someone replies. Replies from every channel land in one conversation per person, and the AI CRM classifies each one (Interested, Customer, Not interested or Other, with your own subcategories), records a confidence and a reason, and drafts an answer grounded in a knowledge base you write. Drafts and follow-up steps always wait in an Action required queue until a person approves them. The AI may move a lead forward in the pipeline on its own when its confidence is at least 85% by default, and an explicit request to stop marks the person Do Not Contact at once, but backward moves, Customer and Other are held for review. WhatsApp calls are dialed through a Chrome extension, recorded to your Cloudflare R2 bucket and transcribed by a model you choose. The software is free and the project is led by Kandid, but the services behind it are not. Email requires Google Workspace mailboxes, LinkedIn and WhatsApp run through a paid Unipile account, every AI feature runs through OpenRouter in bring-your-own-key mode with your own model-provider keys, and enrichment spends credits on your own provider accounts. AgentSDR is new, with its first public release on October 6, 2026, and its docs state plainly that LinkedIn's terms do not allow automation and that accounts can be restricted.
Updates from AgentSDR
Docker Compose now runs two containers, the database and the app. Scheduled jobs such as the daily rollover and LinkedIn outreach run inside the app, the schema is created on first start, and each organization's daily limits reset at the start of its own day. Version 0.3.1 followed the same day with fixes for the Compose image tag and a first-run setup loop.
Version 0.2.0 relicensed AgentSDR under MIT, allowing use in closed-source products and hosted services. The release also added a first-run setup page, a copy-and-run docker-compose.yml using the published image, and a read-only public demo mode.
The first public release shipped multi-organization accounts, per-organization integrations for Unipile, Google Workspace, Cloudflare R2 and OpenRouter, email, LinkedIn and WhatsApp campaigns, the AI-assisted CRM with reply classification and drafts, and Tables with enrichment providers.
Are you the founder? Claim this listing →