Artifacts

Artifacts are pre-defined tool outputs that Chariot stores and highlights so that they are easy to find. They're useful for review, observability, and surfacing key results to operators or end users. Artifacts are enabled per tool on the orchestrator and on each service agent.

Example

Both mcp_tools and chariot_tools are marked as artifacts using artifact_tools. This configuration demonstrates how:

{
  "name": "Artifacts Demo",
  "orchestrator": {
    "name": "Artifacts Demo",
    "model_settings": {
      "ai_provider": "openai",
      "name": "gpt-5-mini",
      "api_key": {
        "owner_type": "project",
        "owner_id": "<your-project-id>",
        "path": "secret/Custom/<openai-api-key-secret-id>/value"
      }
    },
    "system_prompt": "You help users compile findings and publish reports. Use the available tools to gather data.",
    "mcp_tools": {
      "export_server": {
        "mcp_config": {
          "transport": "streamable-http",
          "url": "https://your-mcp-server.example/mcp"
        },
        "artifact_tools": {
          "include": ["^export_report$"]
        }
      }
    },
    "chariot_tools": {
      "packages": ["datasets"],
      "artifact_tools": {
        "include": ["^publish_"]
      }
    }
  }
}

During a session, tools marked as artifacts will have their return payloads categorized as such, which makes it easy to filter on and retrieve those details, rather than needing to search through all tool activity history.

Field reference

• artifact_tools (optional): On mcp_tools and chariot_tools, a filter object (include or exclude) selecting which tools produce artifact outputs.

note

Only tool outputs can be artifacts—not tool inputs or agent reasoning.