Agents SDK Deployment Manager

Local control-plane app for running and observing Agents SDK demo projects.

Prerequisites

• uv
• npm
• Docker, when using the default local-docker target

Run

make run

Open:

http://127.0.0.1:8732

Vite builds the React UI from frontend/ into dist/, and the Flask backend serves dist/ plus /api/*.

Screenshots

Deployments

App details

Traces

Scope

• Import a local Agents SDK project.
• Inspect entrypoints, dependencies, env vars, and sandbox usage.
• Create a local deployment record.
• Start and stop the app/orchestrator as a labeled Docker container by default.
• Generate or reuse an app-level Dockerfile for containerized deployment.
• Show deployment logs, traces, app events, and Docker container activity.
• Label sandbox containers with agents-sdk.* metadata so the manager can map a Docker container back to its deployment and run.

For Docker deployments, the default target builds an app Dockerfile and runs the orchestrator with:

docker run -p 127.0.0.1:<app-port>:<app-port> -v /var/run/docker.sock:/var/run/docker.sock ...

The Docker socket mount lets SandboxAgent create sandbox containers through Docker.

The manager sets AGENTS_SDK_MANAGER, AGENTS_SDK_PROJECT, AGENTS_SDK_DEPLOYMENT_DATA_DIR, and AGENTS_SDK_MANAGER_TRACE_ENDPOINT on the app process. Trace capture is posted back to the manager and stored in the manager-owned SQLite database.

Deployable App Contract

Apps deployed through this manager should have:

• pyproject.toml with openai-agents in the dependencies.
• main.py as the app entrypoint.
• PORT support for local startup.
• uv run python main.py starting the HTTP service when PORT is set.
• A /health readiness endpoint.

Keep CLI-only smoke behavior behind explicit a