Hermes gets better at you. intentic gives every agent a machine.
Both are MIT, self-hosted and answerable from a chat app you already have. Hermes spends its effort on what an agent carries between sessions; intentic on what an agent runs inside.
“The Agent That Grows With You” — Nous Research's self-improving agent, “the only agent with a built-in learning loop”: it creates skills from experience, improves them in use, and builds a model of you across sessions.
Hermes is a general-purpose agent with a learning loop: it writes its own skills, remembers across sessions, schedules its own work, and can put its terminal on Docker, SSH or a serverless sandbox. On memory and self-improvement it does something intentic does not attempt.
intentic is narrower on purpose — software work, and everything underneath it: a container built from an image you approve, one worktree per agent, capabilities that carry your systems, and a diff before anything is yours.
Head to head
3 of 9 rows go to Hermes — ▸ marks them
Where you agree · both are MIT and self-hosted, both run on model accounts that are yours, both hold that an agent should work on a machine you own, both can put the work in a container, both start on a schedule, and both reach you away from the desk.
| Dimension | intentic | Hermes |
|---|---|---|
| What it is | the machine and workplace an agent works in | a general-purpose agent that learns as it goes |
| Memory across sessions | context you author: repo instructions, skills, task lists that survive a rebuild | ▸ a learning loop: generated skills, persistent memory, a model of you |
| Where the work runs | always a Docker sandbox on hardware you own | local, Docker, SSH or a serverless sandbox, as configured |
| Isolation | one container and one git worktree per agent | isolated subagents inside a conversation |
| How you reach it | any browser over a private tunnel, phone included; Discord and Slack when invited | ▸ a CLI plus Telegram, Discord, Slack, WhatsApp, Signal and 20+ more |
| Which models | 5 native harnesses on subscriptions you already pay for, plus any ACP agent | ▸ any endpoint — Nous Portal, OpenRouter, OpenAI or your own |
| Reviewing changes | a diff reader over an isolated worktree, then land or discard | git, in the terminal it ran in |
| Starts on an event | automations: cron, webhook, chat, email, CI/CD, workspace events | a natural-language cron scheduler |
| Licence | MIT throughout — sandbox, platform and CLI | MIT |
What it comes down to
What is learned, and what is written down
The learning loop is Hermes' product: skills generated from experience, memory that outlives a session, a model of you that deepens. intentic's context is authored instead — repo instructions, the skill a capability brings, task lists that survive a rebuild — on the bet that when a coding agent has it wrong you want a file to edit, not a memory to correct.
Isolation is the floor, not a backend
Hermes lets the terminal choose local, Docker, SSH or a hosted sandbox. Here there is no local option: every agent works inside its container and on a worktree cut for it, and landing replays its delta onto your tree as ordinary git you can amend or revert.
One machine, two jobs
Both run on hardware you own, and an intentic automation answers a webhook — so Hermes can hand the repository work down to an agent that gets a container, a worktree and a diff, then report back in whichever chat you asked from.
Pick Hermes when
You want one agent that keeps getting better at you — writing its own skills, remembering how it solved something last month, reachable in whatever chat app you happen to be in — and your work is much broader than repositories.
Checked against hermes-agent.nousresearch.com the week this shipped — these products move fast. Spot something out of date? Open an issue and it is fixed in the next build.