Are there any open-source AI agents?
Yes. All six projects are open source, and each names its licence in its own repository. Kortix is the open-source CompanyOS: the agent's persona, skills and permissions are files in a repo you own.
Open-source SDR toolsaisalesagentsguide.com
Open source AI SDR tools stop the send in six places
Open source AI SDR tools are not one kind of software. Each of these six projects stops the send in a different place, and that place decides whether a company can run it. Kortix is the open-source CompanyOS, and it runs the same sales job with the send held at a human approval gate.
policy:
default_mode: risk
policies:
- match: "helpdesk.delete_*"
action: blockThe agent's persona, skills and permissions are files in the repo, versioned and changed through a reviewed change request.
Each project hands the send to a different place
These six projects are not versions of one product. SalesGPT is a Python library you embed in your own service, so the code that calls it decides whether an email leaves (SalesGPT). Knotie-AI is a Python app that answers leads itself, and its README documents no approval gate (Knotie-AI). Relaticle is a CRM that agents read and update and that sends no outreach (Relaticle).
DeskcommCRM answers on WhatsApp and stops at a sending window, a pacing rule and a cap (DeskcommCRM). OpenSales writes the campaign and parks every draft in an approval queue until you click send (OpenSales). OpenCloser places the call from a desktop app (OpenCloser).
Read them together and the feature lists blur: research, drafting, a sequence, a CRM. What survives the comparison is the stopping place. That is where a buyer has to put a person, a rule or a log, and it is the only part of the stack that reaches a customer.
Comparison
Where each project stops the send
| Product | What it does | Where the send stops | License | How you run it |
|---|---|---|---|---|
| SalesGPT | Sales agent across voice, email and messaging, with a product-knowledge tool | In code you write; the library sends when your service calls it | MIT | Embed the library in a service you host |
| Knotie-AI | Inbound and outbound sales agent in Python that talks to leads | Nowhere documented: the agent answers, and the README shows no approval gate | GPL-2.0 | Run the Python app; last commit January 2025 |
| Relaticle | Self-hosted CRM that agents read and update over MCP or REST | No send: it is the record the agent reads and writes | AGPL-3.0 | Self-host the CRM; connect an agent over MCP |
| DeskcommCRM | WhatsApp CRM whose agents answer, qualify and move the lead | A WhatsApp message under a sending window, pacing and a cap | MIT | Self-host the CRM and its WhatsApp connection |
| OpenSales | Multi-agent outbound system that plans, finds and writes cold email | An approval queue: you review the drafts and click send | MIT, declared in the README; no LICENSE file in the repo | Run it locally with your own SendGrid key |
| OpenCloser | Desktop app for AI cold calling and lead generation | The call it places from your desktop | MIT | Install the Tauri desktop app |
Each cell comes from the repository named in its column, as of November 2026: SalesGPT, Knotie-AI, Relaticle, DeskcommCRM, OpenSales, OpenCloser.
What none of the six gives you
Each project owns one slice. SalesGPT owns the conversation, Relaticle the record, DeskcommCRM the WhatsApp inbox, OpenSales the queue, OpenCloser the dialer. Knotie-AI owns the exchange with the lead and stops there. A company that adopts one still has to build the place where the agent gets its job, the tools it may call, the approvals it must pass, and the record of what it did. None of the six is that place, and installing a second one does not close the gap.
The file that names the agent and rules its tools
Kortix's own example, cut to the registry and one agent. A sales project renames the agent, points it at its own prompt, and writes the rules for its tools.
# yaml-language-server: $schema=https://kortix.com/schema/kortix.schema.jsondefault_agent: support# Registries: what the project has. A registry entry gives nothing to any agent.skills: refund-policy: skills/refund-policyagents: support: prompt_file: agents/support.md permission: bash: "*": allow "git push *": askWhere Kortix keeps the send
Kortix is the open-source CompanyOS, and the platform a company runs this job on. Hand a sales agent the research, the pre-call brief, the CRM write-back and the follow-up draft. Nothing goes out on its own: every follow-up stops at a human approval gate as a draft for a person to review and send. Every action is recorded.
The gate is a rule in the file that declares the agent. Each tool call is ruled allow, ask or block, down to the arguments it was given, so the address a message goes to can decide whether it waits. The persona, the skills and the permissions live in the same repository, and a change to any of them arrives as a diff a reviewer reads (what the repo holds). If the stack runs on your own infrastructure instead, the self-hosted setup covers what the install needs before the first session.
Questions
Questions buyers ask about open-source sales agents
Short answers on what exists, and on what each one costs.
Are there any AI sales agents?
Yes. AiSDR sells a managed AI SDR that runs email and LinkedIn sequences from one inbox, with a two-way HubSpot sync. Those six projects run on infrastructure you control, and Kortix runs a sales agent you own and holds every follow-up at a human approval gate.
Does Open AI have an AI agent?
Yes. OpenAI publishes an Agents API that runs a managed Codex harness and an Agents SDK that runs inside your own application (OpenAI's agent docs). It is a hosted runtime for building agents, with no sales application of its own.
How much do AI sales agents cost?
Most AI SDR platforms cost $200 to $5,000 a month, by AiSDR's own comparison, and AiSDR itself bills $1.50 per person contacted (AiSDR). The commercial tools and their pricing shapes are set out in the AI SDR tools comparison.
Put one send behind a rule you can read.
Start with new-lead follow-up and one tool rule, then read the draft before it goes.