v0.4.0

Consuming artifacts

Deliver artifacts into a harness, narrow what a workspace receives, browse the catalog at runtime, and build custom consumers.

One catalog serves every harness. These pages cover the paths artifacts take out of the catalog and into a runtime: filesystem materialization through podium sync, runtime discovery through the MCP meta-tools, programmatic access through the SDKs, and git-repo distribution through a marketplace target.

PageWhat it covers
Configure your harnessCross-harness delivery. The harness roster, the adapter each one uses, where every artifact type lands on disk, and the MCP and podium sync setup per harness.
Selective materializationSyncing a subset of the catalog into a workspace with include, exclude, and type filters, and defining named profiles to switch between scopes.
Browsing the catalogProgressive discovery. How an agent traverses domains and finds artifacts at runtime through load_domain, search_domains, search_artifacts, and load_artifact, materializing artifacts and their bundled files only when it loads them.
Custom consumers via the SDKBuilding programmatic consumers (LangChain, Bedrock, OpenAI Assistants, custom orchestrators, and eval harnesses) with podium-py or podium-ts.
Handling artifact responsesWhat a consumer does with the manifest and materialized files returned by load_artifact: which bytes arrive in the response and which are written to disk, routing by hints, safety and trust constraints, capability checks, dependency walks, and external resources.
Marketplace publishingRendering the catalog into harness-native git-repo distributions with a kind: marketplace sync target, and pushing them to a git remote.

Start with Configure your harness to get artifacts into a runtime, then narrow the set with Selective materialization.