Resources

A resource is a durable store the agent can read from: a file, a bucket, a database snapshot, a git repo. You name the resource once; Levain makes it available to the agent on every run.

Why resources are a first-class concept

Agents are only as useful as the context they can reach. Hard-coding a URL or S3 path inside a prompt doesn't hold up over time: the data changes, the path moves, the credentials rotate. Named resources let you:

• Swap data sources without rewriting the agent. Point the customer-data resource at a different bucket and every new run picks it up.
• Audit access. Every read shows up in the session as an event. You know which files the agent looked at and when.
• Scope permissions. A resource can be read-only or read-write, scoped to a specific agent, and subject to the same workspace credentials the rest of your team already uses.

The shape of a resource

Every resource has two parts:

• A source — where the data lives (S3, Google Cloud Storage, a git remote, a local file store).
• A path — where the agent looks for it.

The agent reads from the path. Levain handles the rest: keeping the data fresh, the credentials secure, and the access audited.

Resource configuration is workspace-scoped and evolving. If you have a data source you'd like supported, reach out.