product / 02

Probe.

Seven agents in a pipeline. Lead CSV in, three-email sequence out. Personalised per signal, codified rules, vendor-agnostic. Scales from ten to ten thousand without losing voice.

the pipeline

Each agent has one job.

1. 01Research
2. 02Angle
3. 03Critique
4. 04Ideate
5. 05Draft
6. 06Refine
7. 07QA

Each agent is small, focused, replaceable. The pipeline survives upgrades. Swap the model behind any single agent without rewriting the others.

codified rules

Voice consistency at scale.

• Banned words list. 59 patterns AI defaults to. Stripped before send.
• Length constraints. 75 to 125 words per body. No wall of text.
• Subject line rules. Lowercase, no questions, under 6 words.
• QA retry loop. 3 attempts per email. 7-of-10 score threshold or it doesn't ship.
• No emoji. No "I noticed". No "I hope this finds you well".

stack

Vendor-agnostic.

TypeScript. Claude agents. Runs on top of any sequencer (Smartlead, Instantly, Reachinbox). Open source on GitHub. Writes the copy on every client build I run.

---

Want this writing your sequences?

The Build engagement forks the pipeline, tunes it to your voice, wires it into your sequencer. Multi-month. $10–25k+. Yours to keep.