Metrichain
An always-on AI agent that runs cold email outreach end to end: sourcing, copy, sending, replies. We took it from positioning to production.
Metrichain is an always-on AI agent that runs cold email outreach end to end. It finds and verifies the leads, writes the sequences, manages the sending, and surfaces every reply worth a founder's time. From €49 a week, no contracts, no agency retainer.
We partnered across the whole surface: the positioning and the words, the product and its onboarding, the marketing site, and the agent architecture underneath. One system, designed and engineered as a single argument: outbound should feel like delegating, not operating.
Acalmproductinaloudcategory
Outbound tooling competes on horsepower: more dashboards, more integrations, more knobs. Founders do not want a better cockpit. They want the flight handled. So Metrichain sells calm: one agent you talk to, one workspace that reads like a morning briefing, pricing you can try for the cost of a lunch.
The visual world follows the promise. Soft sky over warm paper, one typeface working hard, and interfaces with the density of a well-set page rather than a trading terminal. Every screen is written in plain speech: what the agent did, what it found, what it needs from you.
Onesentencein.Acampaignout.
The whole product collapses into a single interaction: tell the agent what you want. It sources verified, deliverability-checked leads, drafts a personalised multi-step sequence, tests the angles, and files every reply by intent, so meetings rise to the top and the noise stays out of the way.
Onboarding asks three questions before it ever asks for an account, and the agent starts planning while you watch. By the time you sign up, there is already something to see.
Aworker,notachatbot
Under the hood, the agent is deterministic first. Roughly ninety percent of every run is pure code: the model is invoked only at a few narrow, capped, cached points, with template fallbacks. That keeps runs fast, auditable, and cheap enough to price by the week.
It is also built to be trusted with someone else's inbox. The agent drafts and never sends: launching a campaign is an explicit human action. Every client's context, memory, and replies are isolated at the database layer.
- 01Deterministic core
- About ninety percent of an agent run is plain code. The model fills a few capped, cached slots, with template fallbacks when it is off.
- 02Versioned skills
- Every capability is a versioned playbook. Improve a skill once and every client's agent gets better at the same moment.
- 03A second brain per client
- Structured profile plus episodic memory, assembled into a compact prompt. Sharp recall without a runaway context bill.
- 04Drafts only, by design
- The agent never adds real leads, never schedules, never starts sending. Launch is an explicit human action, every time.
- 05Hard tenant isolation
- One sending backend, a native client per customer, and our database as the source of truth, enforced with row-level security.
- 06A reply pipeline that sorts itself
- Webhooked replies are mapped to their owner, deduplicated, classified by intent, and pushed to the top of the right inbox.
Pricedtobetried
Distribution was designed with the product, not after it. Weekly pricing from €49 turns a considered purchase into an experiment, and the marketing site, onboarding, and product share one design system, so the promise on the landing page is the product you get.
- 90%
- of an agent run is deterministic code, not model calls
- 5
- layers between an instruction and a finished campaign
- €49
- a week to start. No contracts, no agency retainer
- 0
- emails sent without an explicit human action
Builtsoafoundercanhandovertheirpipelinewithouthandingovercontrol.