Examples and integrations
Start here when choosing libraries or investigating a Junjo SDK integration. The examples own their runnable configuration and dependencies; the linked guides explain the supported boundaries.
Recommended starting point for native Junjo Agents with OpenAI: junjo_openai_sdk. Its order-support Agent uses a direct Node tool and a conditional Workflow tool on one typed application Store. Run eligible and ineligible orders, then follow the recorded facts, policy decision, and model answer in Studio. The README shows factory-created and caller-supplied Stores, explicit isolated mappings, and the Agent’s separate private runtime state.
| Example | Execution runtime | Provider SDK | Instrumentation | Demonstrates | Setup requirements |
|---|---|---|---|---|---|
| junjo_openai_sdk | Junjo Agent and Workflow | OpenAI Python SDK | Junjo + OpenInference OpenAI | Direct Node and conditional Workflow tools; shared application state; both Store creation options | Python/uv, OpenAI key/model, Studio endpoint/key; mocked tests need no credentials |
| Getting started | Junjo Workflow | None | Native Junjo; no exporter configured | Typed Store actions and conditional graph edges | Python/uv; no provider or Studio required |
| AI Chat | Junjo Agents and Workflows in FastAPI | Google GenAI or xAI | Junjo + OpenInference Google GenAI or xAI SDK tracing; FastAPI OTel | A complete chat application, specialist Agents, provider adapters, and evaluation targets | Provider key; Docker Compose or Python/Node; Studio credentials for evidence/evaluation |
| base_openai_agents | External OpenAI Agents SDK + nested Junjo executions | Scripted model by default | Junjo OpenAI Agents tracing bridge; FastAPI OTel | External runtime tools, mixed traces, and evaluation | Python/uv; no provider key for scripted runs; Studio credentials for evidence/evaluation |
| Standalone evaluation | Junjo evaluation tooling and targets | None; scripted driver | Native Junjo | Independent application declarations, Studio datasets, and execution evidence | Built SDK wheel and Python environment; Studio telemetry key and developer token for runs |
Follow each example’s README for its exact dependency and environment setup. Creating a Store in a factory and sharing it with selected executions are independent choices; see Store composition.
For native Junjo Agents, read Agent definitions, Store composition, and model drivers together. Junjo runs the Agent loop; the chosen provider SDK sends model requests. Add its instrumentor through OpenInference and OpenTelemetry to capture provider-call evidence in Studio.
For the external OpenAI Agents SDK, read its dedicated integration guide. For deployment and credentials, start with Studio deployment. Configure the application and diagnostic entrypoints through the same telemetry bootstrap, then verify one representative trace before relying on the setup.