OpenHermit
Operable agents platform
Start new thread
William Wang

OpenHermit - Operable agents platform

by
Note Anywhere
•
OpenHermit is an open-source platform for deploying AI agents as production services. Most CLI agents keep state in files — fine for one user, but it breaks at scale. OpenHermit puts internal state (memory, sessions, schedules, secrets, etc) in Postgres, while the workspace stays a sandboxed Docker container. Fleet ops become trivial: install a skill on 100 agents with one command, push instructions, audit sessions, all operable from one admin UI and CLI commands.

Add a comment

Replies

Best
William Wang
Maker
📌

Hey Product Hunt 👋 — I'm William, the maker of OpenHermit.

OpenHermit is an open-source runtime that runs AI agents as deployable, operable services — not as personal CLI assistants.

The core idea: most CLI agents (Claude Code, OpenClaw, Hermes…) keep their state in files or per-agent SQLite DBs. That's perfect for one human at one machine, but it falls apart the moment you need to run a fleet — for a team, a SaaS, or a swarm of specialized agents.

So OpenHermit splits state in two:

- Internal state — sessions, messages, events, memories, instructions, skills, MCP, schedules, secrets, users — lives in a shared Postgres, scoped by agent_id.

- External state — the workspace files the agent is actually working on — stays in a sandboxed Docker container per agent.

Once internal state is centralized, fleet operations that were painful become trivial:

- push a new instruction or skill to 100 agents with one command

- one user, multiple identities across Web / CLI / Telegram / Discord / Slack — recognized as the same person

- platform-side visibility into every session, event, and memory

- gateway-managed auth, routing, and lifecycle

The goal isn't "your personal local assistant" — it's agents as deployable services.

🔗 Site: https://openhermit.ai

📖 The story: https://openhermit.ai/blog/agents-but-operable

💻 Repo: https://github.com/williamwa/openhermit (MIT)

I'd especially love feedback on the internal/external state split — does it match how you'd want to run agents at scale? I'll be in the comments all day.