Open source, from IBM Research
Welcome to DAM¶
An open platform for long-lived AI agents. Any harness, any model, on Kubernetes.
DAM in 3 minutes
Follow one agent from start to finish: pick a harness and model, connect your tools, bind it to Slack, put it on a schedule, and gate sensitive actions behind human approval.
What is DAM?¶
DAM is an open platform for running long-lived AI agents, built at IBM Research. An agent runs in its own microVM with a persistent workspace, injected credentials, and its own network boundary — so it can work against real systems, on a schedule, and in Slack, and keep going after you close your laptop. Each agent is built from the same shared building blocks: connections, skills, schedules, channels, artifacts, and knowledge bases. See The building blocks to learn more.
Two ways to run DAM¶
-
IBMers: use the IBM hosted deployment
Sign in with your IBM login at ibm.biz/dam-agents. There's no cluster to manage, and it includes IBM internal model endpoints and integrations. These docs describe this deployment.
-
Open source: deploy DAM yourself
Run DAM on your own Kubernetes cluster. Start with the deployment guide in the dam-agents/dam repository.
Running your own deployment?
The building blocks, features, and CLI work the same way. A few things on these pages are specific to the IBM hosted deployment and won't match yours: the sign-in, the VPN note, the IBM ETE LiteLLM and CurveBender providers, the server URL, and the IBM Slack workspace. Where a page covers one of these, it says so.
What problem does it solve, and for whom?¶
It's for engineers and researchers who run agent workloads — and the teams and leaders who want the good ones to spread. In most organizations, people have built genuinely useful agents and workflows, but most live on one laptop and each new team rebuilds the same infrastructure — running agents, managing credentials, scheduling, storing results — before it can even start. So the work rarely compounds. DAM turns that infrastructure into shared building blocks no one has to rebuild: a practice proven in one place can be picked up everywhere, and new ones — from any team, or the wider industry — get captured once and made available to all.
Start building¶
Pick your entry point.
-
Quickstart
Launch your first Claude Code agent in minutes — in the web UI or the CLI.
-
Take the tour
A guided walkthrough that builds one agent up capability by capability, so you see the whole platform first.