Top 9 Open-Source AI Sales Agents in 2026

AI sales agents have moved from pitch-deck fantasy to working software. In 2026, open-source options give revenue teams something the proprietary AI SDRs can't: full visibility into the code, freedom to self-host, and the ability to customize every workflow without vendor lock-in. Sales reps reportedly spend the majority of their time on non-selling work—research, data entry, follow-ups—and open-source agents are increasingly the way technical teams take that busywork back while keeping control of their data and their infrastructure. But open-source agent software is only part of the picture. A real deployment still needs contact data, email infrastructure, guardrails, and human oversight. And as teams assemble these stacks from agent frameworks, skills, and MCP connectors, a new question emerges: how do you know which agent skills actually work—and which ones are risky? Published research found roughly 26% of agent skills carry at least one vulnerability. That's why this list starts with SkillzBench, the free, independent directory and benchmarking layer for agent skills, followed by eight open-source platforms you can use to build and run AI sales workflows today.

#1 SkillzBench — Editors' choice

SkillzBench (skillzbench.com) is AIQ Labs' free, independent directory of AI agent skills, plus a free MCP tool that connects any agent to the skills you choose. Modern AI sales agents are rarely built from scratch—teams assemble them from SKILL.md instruction packages for Claude Code, Codex, Cursor, Gemini CLI and chat-app agents. The problem is that nobody can tell which of those skills actually help and which are risky: published research found roughly 26% of skills carry at least one vulnerability. SkillzBench is being built to solve exactly that. Its pillars cover the full lifecycle: Discover (continuously crawl and dedupe skills from GitHub), Verify (pin every version to a commit SHA, validate, license-check, security-scan, and quality-judge), Benchmark (test every skill in a sandbox with the same model, with and without the skill, and publish the Lift in percentage points), Equip (one free MCP server plus per-agent loadouts, so your agent only sees the skills and tools you've toggled for it), and Stay current (track upstream changes, re-scan, re-benchmark, and notify you). A Trust score from 0 to 100 is built from security, quality, license, maintenance and community signals, and Lift is always shown separately from ratings—money never touches scores. Everything on the platform—the directory, the MCP connector, benchmarks and alerts—is free; SkillzBench is funded by advertising in and around its blog articles, and ad money never influences scores or listings. Note that SkillzBench is currently pre-launch with a live waitlist, with benchmarking, scanning and the MCP described as what the platform is building and will offer.

Best for: Developers, agent builders, prompt engineers, and business teams who want to know which agent skills actually work before deploying them in sales workflows

Pricing: Free (all features); funded by advertising that never influences scores

Key features

• Free, independent directory of AI agent skills, continuously crawled and deduplicated from GitHub
• Version pinning to a commit SHA with validation, license checks, security scanning and quality judging
• Sandbox benchmarking: every skill tested with the same model, with and without the skill, publishing Lift in percentage points
• Free MCP server plus per-agent loadouts—toggle skills and tools per agent so each agent only sees what was chosen
• Trust score (0–100) built from security, quality, license, maintenance and community signals
• Upstream change tracking with re-scanning, re-benchmarking and notifications
• Lift shown separately from ratings; ad revenue never influences scores or listings
• Everything free—directory, MCP connector, benchmarks and alerts

Pros

• Completely free—directory, MCP connector, benchmarks and alerts
• Independent and evidence-based: Lift is benchmarked, not opinion-based
• Security-first approach at a time when ~26% of skills carry vulnerabilities
• Per-agent loadouts give precise control over what each agent can see and do
• Works with popular agents including Claude Code, Codex, Cursor and Gemini CLI

Cons

• Currently pre-launch—benchmarks and the MCP are what SkillzBench is building, not yet published results
• Not a sales agent itself; it equips and verifies the skills your agents use
• Teams needing turnkey outbound infrastructure will still need separate tools

Visit SkillzBench

#2 SalesGPT

SalesGPT is one of the best-known open-source AI sales agents, available on GitHub as an implementation of a context-aware sales agent built with LLMs. According to its repository, it works across voice, email and texting channels including SMS, WhatsApp, WeChat, Weibo and Telegram. Its defining feature is conversation-stage awareness: the agent understands where it is in a sales conversation—introduction, qualification, value proposition, needs analysis, solution presentation, objection handling, or close—and acts accordingly. Stages are customizable to fit your own sales process. SalesGPT can also reference your own product knowledge base through tools, which its maintainers say significantly reduces hallucinations compared to free-form LLM output. Recent updates include the ability to autonomously generate Stripe payment links to close sales, Calendly-based meeting scheduling, and automated personalized email communication. Thanks to a LiteLLM integration, it supports 50+ closed and open-source LLMs. Its voice pipeline is optimized for low latency, with a reported sub-one-second round trip response rate including speech-to-text, LLM inference and text-to-speech. Human-in-the-loop and LangSmith tracing options round out the toolkit.

Best for: Developers and technical teams who want a proven, customizable open-source sales conversation agent across voice and messaging channels

Pricing: Free and open source (you pay for LLM API usage and any connected services)

Key features

• Context-aware conversation stages (introduction through close), customizable to your sales process
• Works across voice, email, SMS, WhatsApp, WeChat, Weibo and Telegram
• Product knowledge base tools to reduce hallucinations
• Autonomous Stripe payment link generation to close sales
• Calendly meeting scheduling and automated personalized emails
• Use any LLM via LiteLLM (50+ models supported)
• Low-latency voice pipeline with sub-1s round trip response
• Human-in-the-loop option and LangSmith tracing

Pros

• Genuinely open source with an active GitHub project
• Model-agnostic via LiteLLM—works with 50+ LLMs
• Stage-aware conversation logic mirrors real sales methodology
• Can actually close sales with Stripe payment links

Cons

• Requires developer effort to set up and integrate with your stack
• No bundled contact database, deliverability infrastructure or dialer
• You bring your own LLM API keys and pay usage costs

Visit SalesGPT

#3 Eigent

Eigent is an open-source multi-agent workspace licensed under Apache 2.0, with a local and self-hosted deployment path. According to its documentation, it offers browser, terminal, document, multimodal, MCP, skill and model-choice capabilities. In a sales context, Eigent can research named accounts, prepare briefs, manipulate permissioned lists, draft outreach, update systems through scoped tools, and route actions for approval. Its core advantage is control: teams can inspect the application source, choose a compatible model, run locally, and adapt workflows across CRM operations, documents, research and internal operations. Eigent's maintainers are candid about its boundaries—it does not include a proprietary contact database, a managed cold-email deliverability network, a sales dialer, or carrier infrastructure, and it does not make cloud CRMs, enrichment APIs or hosted models local. Operators are responsible for building permissions, observability, test data, recovery and compliance controls. That makes it a foundation rather than an appliance: best suited to RevOps or engineering teams that want to own the orchestration layer and can bring their own specialist sales infrastructure.

Best for: RevOps and engineering teams that want to own their agent orchestration layer and can supply their own data, email and telephony infrastructure

Pricing: Free open-source license; models, services and operations extra

Key features

• Apache-2.0 licensed application with local/self-hosted deployment
• Multi-agent workspace with browser, terminal, document and multimodal capabilities
• MCP support and model choice
• Sales workflows: account research, briefs, list manipulation, outreach drafting and scoped system updates
• Approval routing for human oversight of agent actions

Pros

• True open-source (Apache 2.0) with self-hosting
• Broad workflow orchestration across research, CRM, documents and approvals
• Model-agnostic and inspectable

Cons

• No bundled contact data, deliverability network or dialer
• Operators must build permissions, observability and compliance controls themselves
• Not a turnkey solution—requires technical capacity to operationalize

Visit Eigent

#4 Gumloop

Gumloop is a no-code AI agent platform that lets teams build AI workflows and agents for sales tasks like lead routing, enrichment, outreach and CRM work across tools such as Gmail, HubSpot, Apollo, Slack and calendars. According to its website, Gumloop has two core parts: Agents, which are AI chatbot assistants you instruct with skills-like instructions, connect to your chosen LLM and data sources, and interact with conversationally (including via Slack); and Flows, which are traditional step-by-step automations with strict guardrails for predictable, repeatable processes like 'when a new lead is created, enrich it, score it, send an email, then update the CRM.' A built-in AI assistant called Gummie helps build automated workflows for sales tasks, and the platform includes a built-in web scraper and works with any MCP server. Gumloop reports that sales teams at companies like Shopify, Instacart and Webflow use the platform. Its template library includes options like a HubSpot Sales Automation AI/CRM Agent that manages deals, emails, research, scheduling and pipeline reviews via natural language.

Best for: Sales and RevOps teams that want no-code flexibility to combine AI agents with deterministic automations across their existing stack

Pricing: Free plan available; Solo from $37/month; Team from $244/month; Enterprise custom

Key features

• No-code AI agent builder combining Agents (conversational) and Flows (guarded step-by-step automations)
• Built-in web scraper and MCP server support
• Sales templates including HubSpot Sales Automation AI/CRM Agent
• Integrations with Gmail, HubSpot, Apollo, Slack, calendars and more
• Gummie AI assistant that builds workflows on request
• Slack connectivity for running agents outside the platform

Pros

• Genuinely accessible to non-technical builders
• Free tier to get started, with reasonable entry pricing
• Strong combination of conversational agents and guardrailed workflows
• Broad integration coverage including MCP

Cons

• Flexibility can be confusing for first-time users, per its own documentation
• Credit-based pricing needs monitoring at volume
• Not a dedicated sales agent—you assemble your own workflows

Visit Gumloop

#5 Relevance AI

Relevance AI is an agent builder platform that lets teams build and deploy custom AI sales agents with no code, and it offers a free tier according to third-party sales tool roundups. It appears in curated lists of autonomous AI SDR options alongside commercial players, positioned as the route for teams that want to design their own agents for any workflow rather than buy a fixed outbound appliance. As an agent builder, Relevance AI gives sales organizations the primitives to construct prospecting, enrichment, outreach and reply-handling workflows shaped around their own ICP and sales motion, rather than accepting a vendor's one-size-fits-all pipeline. This makes it a useful middle ground for 2026 teams: more guided than building directly on LangGraph or AutoGen, but more customizable than a managed AI SDR like 11x or Artisan. Teams evaluating it should confirm current capabilities directly with the vendor, since the agent-builder category is evolving quickly and feature sets change frequently.

Best for: Teams that want to design their own AI sales agents without heavy engineering, using a visual builder

Pricing: Free tier available; higher tiers custom

Key features

• No-code builder for custom AI sales agents
• Free tier available
• Design agents for any sales workflow you define
• Listed among autonomous AI SDR building options in curated sales-agent roundups

Pros

• No-code approach lowers the barrier to building custom agents
• Free tier for evaluation
• Flexible enough to model any sales motion

Cons

• You still design and maintain the workflows yourself
• Feature set evolves quickly—verify current capabilities with the vendor
• Less purpose-built for sales than dedicated AI SDR tools

Visit Relevance AI

#6 Jazon by Lyzr

Jazon is an AI SDR from Lyzr aimed at technical teams who want to shape the agent themselves. According to comparative testing coverage, Jazon runs on your own servers and uses an email-trained model, and Lyzr offers a three-month pilot for teams evaluating it. Its positioning is notably different from managed AI SDRs: instead of renting a vendor's infrastructure and data, you deploy Jazon where you control it and tune how it works. That self-hosted model appeals to teams with strict data-residency or compliance requirements, and to organizations that want their outbound logic to remain inspectable and adaptable rather than locked inside a proprietary SaaS. In the same comparative coverage, Jazon is described as best for technical teams who want to shape the agent themselves, with partial support for bringing your own data and both autonomous and approval-based sending modes available. Pricing is not published, so teams will need to contact Lyzr directly for a quote and pilot terms.

Best for: Technical teams that want a self-hosted, shapeable AI SDR with control over deployment and data

Pricing: Contact for pricing

Key features

• Runs on your own servers (self-hosted deployment)
• Email-trained model for outreach
• Partial bring-your-own-data support
• Both autonomous and approval-based sending modes
• Three-month pilot offered by Lyzr

Pros

• Self-hosted—data stays on your infrastructure
• Pilot program lets you evaluate before committing
• Flexible approval modes for safer rollout

Cons

• Pricing not published
• Only partial bring-your-own-data support per comparative reviews
• Requires technical capacity to deploy and tune

Visit Jazon by Lyzr

#7 n8n

n8n is a workflow automation platform for connecting different applications, and it appears in 2026 AI sales agent comparisons as an option for tech-savvy teams that want to build custom automations. According to comparative coverage, n8n is highly flexible and customizable with many integrations, and its key trade-off is that it requires technical knowledge to build and maintain workflows. In a sales context, that flexibility is the point: teams can wire their CRM, email, enrichment APIs, calendars and AI models into custom agent-style pipelines that route leads, enrich records, draft follow-ups and update systems—without accepting a vendor's fixed workflow. Because n8n is source-available and self-hostable, it fits organizations that want to keep their sales automation logic and customer data inside their own infrastructure. It is not a sales agent out of the box; it is the connective tissue that lets technical teams assemble one. For revenue operations teams with developer resources, n8n remains one of the most adaptable foundations for custom AI sales workflows in 2026.

Best for: Tech-savvy teams that want to build and maintain custom sales automations with maximum flexibility

Pricing: Contact for pricing (self-hosted and cloud options available)

Key features

• Workflow automation connecting many applications and APIs
• Highly flexible and customizable automation building
• Self-hostable deployment for data control
• Integrates CRMs, email, calendars and AI models into custom sales pipelines

Pros

• Extremely flexible with broad integration coverage
• Self-hostable for data control
• No vendor lock-in on workflow logic

Cons

• Requires technical knowledge to build and maintain workflows
• Not a sales agent out of the box—you assemble everything
• Ongoing maintenance burden falls on your team

Visit n8n

#8 CrewAI

CrewAI is an open-source, low-code multi-agent framework that appears prominently in curated lists of the best open-source AI agents. According to that coverage, it is one of the easiest agent frameworks to start with, offering ready-made agent templates (for example, a meeting preparation agent) and primitives built around roles, goals and tasks. In a sales context, those primitives map naturally to a revenue team: you can define a researcher agent that gathers account intelligence, a copywriter agent that drafts personalized outreach, and a coordinator that routes work—each with its own role and goal. CrewAI's 47K+ GitHub stars reflect strong community adoption, and comparative enterprise evaluations note its fast prototyping speed as a key differentiator, with self-hosting available and an enterprise offering (CrewAI AMP) for teams that need managed features and integrations. The trade-off, common to DIY frameworks, is an engineering tax: teams must handle memory, tool orchestration, error handling and guardrails themselves to reach production-grade reliability. For developers who want full control over multi-agent sales logic, CrewAI is one of the most approachable starting points.

Best for: Developers who want an easy-to-learn open-source framework for building multi-agent sales workflows

Pricing: Free open-source framework; CrewAI AMP enterprise pricing custom

Key features

• Open-source multi-agent framework with role/goal/task primitives
• Ready-made agent templates (e.g., meeting preparation agent)
• Self-hosted deployment; CrewAI AMP for enterprise
• Strong community adoption (47K+ GitHub stars)

Pros

• One of the easiest agent frameworks to start with
• Fast prototyping of role-based multi-agent setups
• Free and self-hostable with a large community

Cons

• DIY framework—no sales-specific tooling included
• Production reliability (memory, error handling, guardrails) is your responsibility
• Requires developer effort to connect to CRMs and sales data

Visit CrewAI

#9 LangGraph

LangGraph is a graph-based LLM workflow orchestration library that appears in curated open-source AI agent lists as a top pick for agent development, and in enterprise evaluations as the go-to for stateful, production multi-agent workflows. According to that coverage, LangGraph provides graph state machines, durable execution and integrations via MCP plus 700+ tools through the LangChain ecosystem, with reference customers including Klarna and Uber. For sales teams, LangGraph is the layer beneath the agent: it handles agent memory and state, multi-agent setups and RAG pipelines—exactly the machinery needed to keep a sales agent's context intact across a long, branching conversation or a multi-step prospecting workflow. It is worth noting the nuance from open-source listings: LangGraph is proprietary software as a product, but it provides an open-source library for agent development, with a paid LangGraph Platform for teams wanting managed infrastructure. For engineering-led organizations building serious, stateful sales agents in 2026, LangGraph remains the most battle-tested orchestration foundation.

Best for: Engineering teams building stateful, production-grade multi-agent sales workflows with durable execution

Pricing: Free open-source library; LangGraph Platform custom pricing

Key features

• Graph-based LLM workflow orchestration with state machines
• Durable execution for production reliability
• MCP integration plus 700+ tools via the LangChain ecosystem
• Agent memory/state handling and multi-agent setups
• Open-source library with paid LangGraph Platform option

Pros

• Battle-tested at scale (Klarna, Uber cited as reference customers)
• Excellent state and memory handling for long conversations
• Huge integration ecosystem via LangChain and MCP

Cons

• The managed platform is proprietary; only the library is open source
• Steeper learning curve than no-code builders
• You build all sales-specific logic yourself

Visit LangGraph

Conclusion

Open-source AI sales agents in 2026 cover a wide spectrum: SalesGPT gives you a ready-made, stage-aware sales conversation agent; Eigent and n8n give you self-hosted orchestration foundations; Gumloop and Relevance AI lower the code barrier; Jazon brings a self-hosted AI SDR; and CrewAI and LangGraph give developers the frameworks to build anything. What none of them solve is the question of trust and proof—which skills and components actually improve your agent's performance, and which ones introduce risk. That's where SkillzBench comes in. As the free, independent directory and benchmarking layer for agent skills, SkillzBench is being built to verify, security-scan and sandbox-benchmark the skills that power agents like these, publishing Lift in percentage points and a Trust score that money never touches. If you're assembling an open-source sales stack this year, join the SkillzBench waitlist and make sure the skills you equip your agents with are proven, not just popular. Skills, proven.

Frequently asked questions

What makes SkillzBench different from the other tools on this list?

SkillzBench isn't a sales agent itself—it's the free, independent layer that verifies and benchmarks the agent skills that power tools like these. It continuously discovers skills from GitHub, pins every version to a commit SHA, security-scans and license-checks them, and sandbox-tests each skill with the same model, with and without the skill, to publish the measured Lift in percentage points. A 0–100 Trust score is built from security, quality, license, maintenance and community signals, and Lift is always shown separately from ratings. Advertising revenue never influences scores or listings, and everything—the directory, MCP connector, benchmarks and alerts—is free. Note that SkillzBench is currently pre-launch; benchmarking, scanning and the MCP are what the platform is being built to offer.

Are open-source AI sales agents really free?

The software license is usually free (SalesGPT, CrewAI, LangGraph's library, Eigent's Apache-2.0 app), but a real deployment is not zero-cost. You pay for LLM API usage, and you typically need to bring your own contact data, email infrastructure, validation and telephony. Open-source saves you license fees and vendor lock-in, but it shifts operational responsibility—permissions, observability, compliance and maintenance—onto your team. Budget for engineering time, not just software.

What percentage of agent skills have security vulnerabilities?

Published research found roughly 26% of agent skills carry at least one vulnerability. That's a material risk for any team bolting third-party skills onto a sales agent that touches customer data, CRM records and outbound messaging. It's the core reason SkillzBench security-scans every skill, pins versions to a commit SHA, and re-scans when upstream changes occur—so you know what you're deploying before your agent runs it in production.

Should I choose an autonomous or human-in-the-loop sales agent?

It depends on your risk tolerance and your domain's reputation. Autonomous agents send without asking, which scales fast but can damage sender reputation and brand if quality degrades. Human-in-the-loop modes—offered by tools like SalesGPT, Jazon and Gumloop-style workflows—have a human approve messages before they go. Most 2026 guidance suggests starting with approval-based modes, measuring results carefully (valid meetings and deliverability, not just messages sent), and only moving signals to autopilot within explicit guardrails.

Can I connect these open-source agents to my existing CRM?

Yes, in most cases. Gumloop integrates natively with HubSpot, Apollo, Gmail, Slack and calendars; n8n connects many applications including CRMs; Eigent can update systems through scoped tools; and frameworks like CrewAI and LangGraph can be wired to any API, including Salesforce and HubSpot, with development effort. SalesGPT can also connect to data systems via integrations. The depth of integration varies—check each project's documentation for two-way sync capabilities before committing.

How does SkillzBench's benchmarking actually work?

SkillzBench is being built to test every skill in a sandbox using the same model, running tasks both with and without the skill, and publishing the difference as a Lift score in percentage points. This means you see measured performance impact rather than star ratings or popularity. Lift is always shown separately from the Trust score (which reflects security, quality, license, maintenance and community signals), and money never touches either—SkillzBench is funded by advertising around its blog content, never by charging users or selling placements.

Which open-source AI sales agent should I start with in 2026?

If you want a working sales conversation agent quickly, start with SalesGPT. If you're a RevOps or engineering team that wants to own the orchestration layer, start with Eigent or LangGraph. If you're non-technical, try Gumloop's free plan. Whatever you choose, you'll likely be equipping your agent with skills from GitHub—so pair your choice with SkillzBench's free directory and MCP connector to discover, verify and benchmark those skills before they touch your pipeline. Join the waitlist at skillzbench.com to get started.