AgentSDR

Product Introduction

1. Definition: AgentSDR is an open-source, self-hosted AI Sales Development Representative (SDR) workspace and CRM platform. Technically, it is a full-stack web application built on a modern stack (Next.js 16, React 19, TypeScript, Postgres, Drizzle ORM, Tailwind CSS, Bun) and deployed via Docker. It functions as a unified outbound sales automation and engagement platform.
2. Core Value Proposition: It exists to replace fragmented, expensive, and data-leaking SaaS sales tools (like Clay, Smartlead, and HubSpot Sales Hub) by providing a single, private workspace for multi-channel outreach (email, LinkedIn, WhatsApp) backed by an AI-powered CRM that automates reply triage and response drafting. Its primary value is data ownership, cost control (no per-seat or per-contact fees), and workflow consolidation for technical sales teams.

Main Features

1. Multi-Channel Outreach Engine: AgentSDR orchestrates sequences across email, LinkedIn, and WhatsApp from a single interface. For email, it uses your own Google Workspace mailboxes with configurable daily caps, sending windows, and mailbox pooling. For LinkedIn, it integrates via Unipile to automate connection invites and follow-ups within strict platform limits (30/day premium, 5/day free). For WhatsApp, it uses a Chrome extension to initiate calls within WhatsApp Web, records both sides, and transcribes the audio.
2. AI-Powered CRM & Inbox Triage: This is the core AI automation. The system automatically classifies every inbound reply (e.g., Interested, Not Interested, Customer, Other) using your own AI model key (via OpenRouter). It then drafts context-aware responses grounded in your provided knowledge base (pricing, FAQs, case studies). Crucially, it operates on a "draft, not auto-send" principle, holding all drafts for human review and approval.
3. Unified Lead Database with AI Enrichment (Tables): All channels feed from and into a single database of people and companies. The "Tables" feature allows for importing CSV/XLSX data and enriching it with AI-generated columns (e.g., "Pitch angle") or data from external APIs like Apollo. This enriched database can be used directly to create targeted campaigns.
4. Self-Hosted Deployment & Data Control: The entire application runs on your infrastructure using Docker Compose. All data—lead records, conversation history, and call recordings—resides in your own PostgreSQL database and Cloudflare R2 storage bucket. All AI calls use your own OpenRouter API key, ensuring no data passes through third-party AI SDR services.

Problems Solved

1. Pain Point: High cost and vendor lock-in of commercial sales engagement platforms (SEPs) and CRMs that charge per seat, per contact, or for AI features. AgentSDR eliminates recurring license fees, charging only for infrastructure and your own AI usage.
2. Pain Point: Data privacy and security concerns with SaaS platforms where sensitive lead data, communication history, and proprietary outreach strategies are stored on vendor servers. AgentSDR keeps all data in your private cloud environment.
3. Target Audience: Technical founders, sales ops engineers, and growth teams at startups and SMBs who prioritize data ownership, need to customize their sales stack, and have the capability to manage a Docker-based deployment. It's also suited for agencies managing outreach for multiple clients who require strict data separation.
4. Use Cases: Running integrated email and LinkedIn cold outreach campaigns for a B2B SaaS product launch. Managing a high-touch WhatsApp calling sequence for lead follow-up, with all calls recorded and transcribed for quality assurance. Consolidating reply management from multiple channels into a single AI-assisted inbox to improve SDR response time and consistency.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros:
     • Complete Data Sovereignty: Ultimate control and privacy as you host all data and AI logic.
     • Significant Cost Reduction: No per-user or per-contact licensing; costs are limited to server hosting and your chosen AI model usage.
     • Workflow Consolidation: Replaces the need for separate tools for email sequencing, LinkedIn automation, WhatsApp logging, and basic CRM.
     • Transparent & Controllable AI: AI actions are grounded, draft-only, and use your explicitly chosen LLM provider without silent fallbacks.
     • Fully Customizable: As open-source software, the codebase can be modified to add features, integrations, or custom logic.
   • Cons:
     • Technical Overhead: Requires in-house technical resources to deploy, maintain, update, and troubleshoot the Docker/Postgres stack.
     • Dependency on Third-Party Bridges: LinkedIn and WhatsApp automation rely on the Unipile integration, adding a dependency and potential point of failure/change.
     • Limited Native Email Provider Support: Currently only supports sending via Google Workspace, not SMTP or other providers like Outlook or Amazon SES.
     • Setup Complexity: Connecting channels (Google Workspace service account, Unipile, OpenRouter, R2) involves non-trivial configuration steps compared to signing up for a SaaS.

2. Key Alternatives & Differentiation:

   • Clay / Smartlead (SaaS AI SDR Platforms): These are hosted, all-in-one SaaS solutions. Differentiation: AgentSDR is self-hosted and open-source, offering superior data control and avoiding monthly fees. Clay/Smartlead offer easier setup and maintenance but lock you into their platform and pricing.
   • HubSpot Sales Hub / Outreach.io (Sales Engagement Platforms): These are mature, feature-rich enterprise SaaS platforms. Differentiation: AgentSDR is a fraction of the cost (infrastructure-only) and provides deeper, code-level customization. However, it lacks the vast ecosystem of native integrations, robust reporting, and enterprise support that the SaaS giants offer.
   • Building In-House: A team could build a similar system. Differentiation: AgentSDR provides a production-ready, fully-featured open-source codebase, saving potentially hundreds of developer hours. It offers a proven architecture for a complex problem, allowing teams to customize rather than build from scratch.

Frequently Asked Questions (FAQ)

1. Is AgentSDR really free to use? Yes, the software itself is free and open-source (AGPLv3 licensed). You only pay for your own infrastructure costs (server hosting, database), the accounts you connect (Google Workspace, Unipile subscription), and your own AI API usage on OpenRouter. There are no per-seat or per-contact fees charged by the AgentSDR project.
2. How does AgentSDR handle LinkedIn automation without getting accounts banned? AgentSDR enforces strict safety limits configured within the LinkedIn channel via Unipile: a maximum of 30 connection invites per day for premium accounts and 5 for free accounts, with randomized delays (30-60 seconds) between actions and configurable sending windows. It is designed to operate within LinkedIn's published guidelines, though no tool can guarantee restriction.
3. Can I use my own AI model, like OpenAI GPT-4 or Anthropic Claude, with AgentSDR? Yes, but indirectly. AgentSDR uses OpenRouter as its AI gateway. You provide your OpenRouter API key, and within OpenRouter's dashboard, you can pin your usage to specific providers and models (like OpenAI or Anthropic). This gives you flexibility and cost control while keeping the application's integration simple.
4. What technical skills are required to deploy and maintain AgentSDR? Deployment requires basic DevOps knowledge: ability to run Docker and Docker Compose, set up a PostgreSQL database, and configure environment variables. Ongoing maintenance involves monitoring the server, applying updates by pulling new code from GitHub, and managing the connected service integrations. It is not a "click-to-deploy" solution for non-technical users.
5. Does the AI automatically send replies to leads? No, a core design principle is "drafts, not auto-sends." The AI classifies replies and generates response drafts based on your knowledge base. All drafts are placed in a review queue ("Action required") where a human must approve and send them, either as-is or after editing. The AI can only automatically advance a lead to a positive pipeline stage (e.g., from "Interested" to "Meeting Requested") when it has high confidence.