deploy

Valid sandbox provider identifiers.

Valid sandbox scope values.

Execute the deepagents deploy command.

Execute the deepagents dev command.

Execute the deepagents init command.

Register the top-level init, dev, and deploy subparsers.

The three commands used to live under deepagents deploy {init,dev} but are now flat: deepagents init, deepagents dev, and deepagents deploy. This function registers all three on the root subparsers object.

Bundle a deepagents project for deployment.

Reads the canonical project layout:

<project>/
    deepagents.toml  # required — agent + sandbox config
    AGENTS.md        # required — system prompt + seeded memory
    .env             # optional — environment variables
    mcp.json         # optional — HTTP/SSE MCP servers
    skills/          # optional — auto-seeded into skills namespace

...and writes everything langgraph deploy needs to a build directory.

Deploy configuration parsing and validation.

Reads deepagents.toml and produces a validated DeployConfig.

The new minimal surface has exactly two sections:

• [agent]: name + model
• [sandbox]: sandbox provider settings

AGENTS.md is always seeded into a shared memory namespace so the agent can read it at runtime, but writes/edits to that path are blocked by a read-only middleware in the generated graph.

Skills (skills/) and MCP servers (mcp.json) are auto-detected from the project layout. The agent's system prompt is read from AGENTS.md at bundle time — there is no system_prompt key.

String templates for generated deployment artifacts.

These templates are rendered by the bundler with values from ~deepagents_cli.deploy.config.DeployConfig.

The generated deploy_graph.py uses a CompositeBackend with two read-only StoreBackend routes (memories and skills) and the configured sandbox as the default writable backend.

There is no hub path and no custom Python tools.

CLI commands for deepagents init, dev, and deploy.

Registered with the CLI via setup_deploy_parsers in main.py.