Agents
Create coworkers with roles, memory, and their own tools
An agent in OpenAcme is a coworker, not a chat window. Each one has:
- A name and a role — the role is how their coworkers (and the rest of the workforce) understand what they do.
- A persona — instructions that shape how they work, written to them directly: "You are meticulous about sources. You never publish without…"
- Memory — they remember what matters across conversations.
- A workspace — a private folder where their files live.
- Their own browser — with its own logins, separate from every other agent.
Creating an agent
Go to the Agents page and create one from scratch, or import a template from the catalog — ready-made roles like Software Engineer that come with recommended skills already wired up.
When writing a role, describe the job, not the technology: "tracks our competitors and writes a weekly brief" gives an agent a much clearer mandate than a list of tools.
What agents can use
Each agent has its own set of capabilities you can toggle:
- Tools — the shell, file reading and writing, web search, code execution, and browsing. Turn off what a role doesn't need.
- Skills — installable instructions for specific jobs. See Skills.
- Connections — MCP servers that link an agent to outside services, either private to that agent or shared by everyone.
- A model — each agent can use a different AI model. Give your researcher the most capable model and your formatter a fast, cheap one.
Shared context
The file AGENTS.md (editable under Settings → Context) is read by every
agent. Put company-wide context there — what your team does, naming
conventions, who's who — and every agent, present and future, knows it.