Hermes is a personalized email platform that connects to your database, reads user activity, and sends a unique email to every recipient based on triggers you define. Not a broadcast, not a template — a different message for each person, at scale. InsForge is the backend that makes it run.
The Challenge
Hermes started on self-hosted Supabase running on a VPS. It worked early on. But as the team built more with AI agents, a recurring problem emerged.
Agents would read schema from the dev environment and assume prod was the same. It wasn't. Migrations would error out, and someone had to step in manually to fix the SQL and re-run it.
"The migration SQL it produced would often error out or need manual fixes, so it wasn't reliable enough for us to trust directly."
Carson Lin, Co-founder, Hermes
The agent wasn't wrong. The infrastructure just couldn't give it an accurate picture of its own environment. Every fix meant a human relaying context the agent should have had in the first place.
Why InsForge
1. Agent Native
InsForge is built around a different assumption: the agent always has access to the full context it needs to operate, including schema state, runtime information, and environment status, without a human in the relay chain. The agent is the primary operator of the platform — it can inspect the real state of the backend, generate correct migrations, and apply them autonomously.
"With InsForge, the agent just handles it. We stopped being the ones debugging infrastructure."
2. Less Complexity, Faster Iteration
Previously the team managed Auth, Database, and model access across a fragmented self-hosted setup. With InsForge, all three come out of the box. Fewer platforms, less coordination overhead, and agents that can operate the full stack without context switching between services.
"We're spending 80% less time on infrastructure and backend work."
Jaehoon Son, Co-founder, Hermes
The Solution
Hermes built their entire product on InsForge. Customers connect their own database to Hermes, define triggers based on user behavior, and the platform takes it from there.
When a trigger fires, Hermes pulls activity data from the customer's database, runs it through the Model Gateway to generate a personalized email for each recipient, and sends. Each recipient gets a different email, tailored to what they actually did. No templates, no manual segmentation.
Because InsForge is agent-native, the Hermes team manages schemas, runs migrations, and handles environment changes through their agent directly. No switching between platforms, no broken context across environments.
"The agent can actually run the whole thing now. That's what we were trying to build from the beginning."
Germain Hirwa, Co-founder, Hermes
InsForge is the backend for agentic builders
What made InsForge work for Hermes comes down to one thing: the agent is the operator. Not a helper, not a shortcut — the actual operator. It can read the real state of the backend, make changes, and move on. No human needed to fill in the gaps.
For a team building an agentic product, that alignment matters. When the infrastructure works the same way the product does, everything gets faster. Hermes spends 80% less time on backend work. Schema changes go through the agent. Environments stay in sync. The team focuses on what they're actually building, not on keeping the infrastructure running.
That is what it looks like when the backend is built for the way agents work.
What's Next
Hermes is continuing to build a platform where teams can define behavioral triggers and automatically send highly personalized emails at scale. With InsForge as the backend, the team can spend less time managing infrastructure and more time improving personalization, expanding trigger workflows, and shipping faster.