Docs/Getting Started

Hermes Documentation

Hermes reads your production data and turns real user behavior into hyper-personalized emails — automatically.

Hermes sits on top of the data you already have — your production database, your analytics, your billing. It understands your schema, finds the users who matter right now, and writes each of them an email from their actual data — no segments to maintain, no templates to fill in.

Driving Hermes with an AI agent instead? Point it at docs.tryhermes.dev — the agent-facing reference, written to be read by agents directly — and connect it via MCP or the CLI.

Getting to your first campaign takes three steps.

Connect your data

Connect a read-only data source: PostgreSQL, MySQL, SQL Server, Firestore, PostHog, Stripe, or a CSV upload. Hermes maps your schema and understands your users in seconds.

Connect your data →

Create a trigger

Describe who you want to reach and why — in plain English. Hermes writes the query, evaluates each matched user individually, drafts the email, and runs it on a schedule. Transactional or marketing, your call.

Create a trigger →

Review and send emails

Every email is drafted from real user data — no mail merge. Review drafts before they go out, or let established triggers send automatically, and track performance from the dashboard.

Review and send emails →

How Hermes works

Hermes runs two loops on top of your data.

The campaign loop. When a trigger runs, Hermes executes its detection query against your data source to find matching users. For each match, an evaluator agent decides whether now is the right moment for that specific person — checking what they've already received, whether they replied, whether they opted out. Only then does Hermes read that user's actual activity and context and draft a one-of-a-kind email. Drafts either wait in your review queue or send automatically — you choose per trigger.

The discovery loop. Every few days, Hermes investigates your data on its own and surfaces signals — non-obvious cohorts and behavior patterns your dashboards would miss, each scored by surprise and confidence. You review them and turn the good ones into triggers with one conversation.

Hermes never writes to your data. Connections are read-only and credentials are encrypted with AES-256 the moment you add them.

One platform, four surfaces

Everything in Hermes — connections, triggers, drafts, signals, templates, analytics — is available through four interchangeable surfaces:

Surface	What it is
Dashboard	The web app: review queue, signals deck, campaign analytics, inbox for replies, and an in-app agent that can operate all of it in chat.
MCP server	A hosted MCP endpoint so claude.ai, Claude Code, Cursor, or any MCP client can run your outreach.
CLI	`hermes` in your terminal — every operation, scriptable, with `--json` output.
REST API	The `/api/v1` control plane that all of the above are built on.