Image: GitHub
Why it mattersA small team doing cold outbound can now try an open-source agent on the AI plan they already pay for.
A small team doing cold outbound pays a sales tool. Then it pays again for the AI calls that tool makes. Franco cuts the second bill. The repository reached rank 2 on Trendshift's daily list on 23 October. It has 205 stars, 47 forks and an MIT licence. Gortex, a company in Sunnyvale, California, created the project on 15 October 2026. The README calls Franco "an outbound sales agent that runs on your own machine".
How it uses an AI subscription
The README's own words set out the billing idea. "There is no model API bill." Model calls go through the claude command on a Claude Pro or Max plan. A team can also opt in to an isolated codex runtime on a ChatGPT plan. Franco is a plain command on the user's own machine. It calls the same CLI the user already pays a subscription for. A team on Claude Pro as a daily driver can run Franco on the same plan.
The three modes
Franco has three ways to run a campaign. In Generated mode it "finds businesses, reads their websites, picks the right person, and drafts a personalized sequence". In Approved copy mode a team "imports accounts you researched yourself" and "Franco sends your approved sequences word for word. No model writes them". The third, Raad, is funnel-based: Franco "places every account in the narrowest funnel it fits" and writes the first email from that funnel.
What it does not send on its own
The README is clear on who signs off. "By default every outgoing email waits in the outbox for your approval." Nothing goes out until a human says yes. The team can turn on autopilot to change that. Franco imports a CSV, verifies email addresses, schedules sends on business days, handles replies, and tracks bounces. It also enforces a suppression list and a block list on claims that cannot be made.
What this changes
A team already paying for Claude Pro, Claude Max or a ChatGPT plan can run an outbound campaign on top. The tool is open source and MIT-licensed. The team can read what the agent sends before it sends it. Franco is less than two weeks old. 205 stars in nine days is a thin crowd. Trendshift rank 2 is a far stronger sign of early use. The README names no customers. There is no public measurement of reply rates or deliverability. A team trying this today is a tester, not a steady user.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
