Long-running AI agents that pick up where they left off

Long-running AI agents keep their files, tools and history between tasks, so each run builds on the last one. Agenhood runs them on your own server, pauses them when idle and lets you undo a bad run.

Why most AI agents start from zero

Most agent runs get a fresh sandbox that is thrown away at the end. The next run clones the repo again, reinstalls its tools and forgets what it learned. Fine for a one-off task. Wasteful for an agent that writes a weekly report, keeps a research log or looks after one codebase for months.

Agenhood keeps the agent instead. Its files, tools and setup are still there tomorrow, so every task builds on the last one.

What a stateful AI agent keeps

  • Its files. Notes, reports, cloned repos, scripts and data stay in the agent's workspace for weeks. You can browse and download them from the console.
  • Its setup. The model, tools, skills and MCP servers stay the same from task to task.
  • Its history. Every past task and what happened in it.
  • Its conversations, if you want. Related tasks can continue the same conversation.

This is about the agent's files and setup, not a vector memory store. When an agent needs to remember something, it writes it to a file, and the file is still there next week.

Long-running AI agents that cost nothing while idle

Keeping agents around usually means paying for idle containers. Not here. An agent with nothing to do pauses after 20 minutes and uses no CPU or memory. The next task wakes it up with everything in place.

Agents left alone for a long time are cleaned up step by step: archived after 3 days paused, deleted 30 days later. An admin can change these limits.

Undo a bad run

Weeks of work shouldn't be lost to one bad task. Agenhood saves a snapshot of the workspace after each task, and you can roll back to any of them from the console. Rolling back never erases history, so you can always change your mind.

Back up a persistent AI agent workspace to Git

Connect a private Git repo and every snapshot is pushed there too. If the server fails, the agent's work is safe in your repo.

Put it to work

Long-running agents shine on recurring work, where tonight's job builds on last night's. See scheduled AI agents. Everything runs on one server you control: see the self-hosted AI agent platform.

What it doesn't do

  • Unused agents expire. With the default limits, an agent left alone for about a month loses its files. Raise the limits or use the Git backup for anything you need to keep.
  • No snapshots for agents working in a Git repo. The repo's own history does that job.
  • Rollback restores files only. It can't undo things outside the workspace, like a message the agent sent.

Questions people ask

What is a long-running AI agent?

An agent that stays around after a task ends. It keeps its files, setup and history, so the next task starts where the last one stopped instead of from zero.

Is this the same as AI agent memory?

Not quite. Agenhood doesn't add a vector memory store. The agent keeps its files and setup on disk, and it can write notes to a file and read them again next time.

What happens to a stateful AI agent when it pauses?

It stops using CPU and memory, but keeps its files and settings. The next task wakes it up.

How long does a persistent agent sandbox last?

By default, an idle agent pauses after 20 minutes, is archived after 3 days paused and is deleted 30 days later. An admin can change these limits, and a Git backup keeps a copy off the server.

Can I undo what an agent changed?

Yes. Agenhood saves a snapshot after each task and you can roll back to any of them. For agents working in a Git repo, the repo's own history does this job.

Does the agent continue earlier conversations?

It always keeps its files. To also continue a conversation, send related tasks as part of the same session.

Run it locally in one command

You need Docker and Python 3.12 or newer. make dev builds the agent image, starts the stack, runs migrations, creates a login and prints the console URL.

Terminal
git clone https://github.com/appssemble/agenhood.git
cd agenhood
make dev