DataRobot Agent Framework — from dr CLI to a deployed agent · sourced from public github.com/datarobot-oss and github.com/datarobot-community repos · flow is left → right

A worked example for the architecture-diagram skill: every claim on this canvas traces to a file in a public DataRobot repo (README, source, copier.yml, GitHub Actions). A developer drives the dr CLI to ① scaffold an app — either dr templates setup (git-clones an already-composed template) or dr component add, which shells out to uvx copier copy against a component registry. Components ② render into the project template datarobot-agent-application. The generated agent/ ③ builds on the datarobot-genai runtime library, ④ routes LLM calls through get_llm() (DataRobot's LLM Gateway by default, or a deployment/NIM/external model via an LLM Blueprint), and ⑤ pulls tools from a co-built MCP server over the AG-UI protocol. task deploy ⑥ runs Pulumi (datarobot-pulumi-utils over the pulumi-datarobot provider), which ⑦ provisions the agent as a custom model (target_type=AgenticWorkflow) + serverless deployment, and ⑧ end users chat through the deployed FastAPI application. Click a step, an arrow label, or any box for details with file citations, or hit ▶ Walk the flow (bottom-right). Every box drawer links its GitHub repo — full list in Code repositories below.