Today we're excited to announce Agent Builder — a space for teams to build and share agents.
Think about Agent Builder as a composable agent studio:
- Developers write tools, select models, and build workflows.
- Non-developer teammates (PMs, ops, support) use the Builder UI to assemble those primitives into agents (or vibe code their own), chat with them, and publish them.
- It includes RBAC, ownership, and visibility controls so it's safe to give a wider org access.
- All agents built are “just” code so you can run them anywhere.
Why we built Agent Builder
One of the main places we’ve seen agents are internal automation projects. Teams across the organization have work they want to automate, but there aren’t enough engineers to help.
We kept hearing the same question: "how do we let the rest of the team build agents themselves?"
We’ve watched many organizations at scale build internal AI tools with dedicated engineering teams: Marsh McLennan built a search application used by ~75,000 employees; Softbank shipped Satto Workspace for document automation.
Even in our small team, our internal agents summarizing feedback from call recordings, auto-posting metrics, and drafting GitHub release notes have made us far more efficient.
But we’re all engineers. And the community kept telling us: they didn’t think agent creation should just be the domain of engineers.
How it works
In your Mastra instance, add a builder key to the MastraEditor, and add in tools you want to allow.
export const mastra = new Mastra({
// ...
editor: new MastraEditor({
builder: {
enabled: true,
configuration: {
agent: {
tools: { allowed: ["crm-lookup-tool", "ticket-search-tool"] },
agents: { allowed: ["account-research-agent"] },
workflows: { allowed: ["customer-onboarding-workflow"] },
models: { allowed: [{ provider: "openai" }, { provider: "anthropic" }] },
memory: { observationalMemory: true }
}
}
}
})When a request is made to create a new agent, the configured tools, agents, and workflows are discovered from their descriptions. The resulting agent is saved to the configured storage. It stays private until it's published, when it shows up in the Agent Builder UI for other team members to find and test.
Install and configure the Slack provider, and the team can connect their own agents to Slack.
Permissioning
Agent Builder ships with two roles — admin (full access) and member (a scoped allow-list of resource:action permissions). These are granular, IAM-style — e.g. stored-agents:write, tools:execute, workflows:read with glob-style wildcards. Roles are just bundles of these permission patterns.
Agents are private until made public, and can also be in draft status before being published.
Try it out
Agent Builder is available for self-hosting, and on the Mastra platform under an enterprise edition license. You can try it out locally in Mastra Studio, but to deploy it you’ll need a license key.
Book a call and we’ll get you set up.
This version includes the natural language builder and governance layer with Composio support shipping next week.
For full setup instructions, see the Agent Builder docs.
