Sender reputation

An agent’s reputation score (0–1), built from real engagement and scoped to its workspace. Queryable today through the mails_get_reputation tool, which returns the agent’s own score.

The score reflects the agent’s own send, bounce, complaint, and reply history over the last 30 days. Per-agent reputation is critical for AI-powered emails because autonomous agents can send at high volume without human oversight.

What feeds it

• A starting point. Every agent starts at 0.85 and keeps that score until its own mail moves it.
• Bounces. Twice the agent’s bounce rate over the last 30 days comes off the score.
• Spam complaints. Five times its complaint rate comes off, so complaints weigh the most.
• Replies. 0.3 times its reply rate is added, counting a reply rate of up to 50%.

Querying it

Inbound events carry a different number, sender_reputation (0–1): how far to trust whoever wrote in. It is the classifier’s estimate, or a preset such as 0.3 for a sender that failed authentication, and a message whose injection_score is above 0.5 gets a lower preset. To pull an agent’s own score directly — built from its send, bounce, complaint, and reply counts over the last 30 days — call the reputation API:

const rep = await client.reputation.get(agentId);

if (rep.reputation < 0.4) {
  // Low standing — tighten sending or pause this agent
}

Suppression and pauses

A hard bounce or a spam complaint suppresses the address for every workspace on the platform; an unsubscribe suppresses it only for the workspace it came from. An agent past 5% bounces or 0.3% complaints over the last 30 days, once it has sent 20 messages, is paused on its own. If the workspace’s mail as a whole crosses the same line, every agent in it is paused until the rates recover.

Read the architecture page for where the reputation graph sits beside the send and receive pipelines.