This package has had its final release, 0.28.3. Development continues as django-spec-agents (documentation), which merges this package with djangorestframework-pydantic-ai. Its import package is django_spec_agents. See Moving to django-spec-agents.

This release is code-identical to 0.28.2. A known defect stays in it and is fixed only in django-spec-agents: a destructive tool whose registered name differs from its function's name, through @tool(name=) or a ToolSpec registered directly, is offered to the model under the function's name, so the tool guard does not ask before it runs. If you stay on 0.28.x, put the function's name in require_approval; the registered name does nothing there.

django-pydantic-agent

The settings-agnostic agent-host substrate shared by Django's Pydantic-AI transports.

It takes an AgentConfig in and returns a built pydantic_ai.Agent — with composed toolsets, audit, and user resolution — and reads no Django settings of its own. Each transport owns its settings namespace, builds an AgentConfig, and hands it down:

• django-ag-ui — AG-UI over SSE for a browser client.
• django-a2a — agent-to-agent (planned).

pip install django-pydantic-agent

Most projects install a transport and get this as a dependency. Reach for it directly when you are building a transport, sharing agent construction between two of them, or embedding an agent with no HTTP surface at all.

from django_pydantic_agent import AgentConfig, ToolRegistry, build_agent, tool

registry = ToolRegistry()


@tool(registry)
def list_orders(status: str) -> list[dict]:
    """List orders with the given status."""
    ...


agent = build_agent(registry, AgentConfig(model="anthropic:claude-sonnet-4-5"))

What it provides

• Agent construction — AgentConfig → build_agent, model and provider resolution, and an escape hatch to replace construction entirely.
• A tool registry — @tool with destructive / category / confirm / summary metadata, and JSON Schema derived from the signature.
• Storage contracts — ConversationStore / AttachmentStore protocols, reference implementations, and Django models in contrib.store.
• Policy — an audit trail over every tool call, and a server-side approval gate for destructive tools. Both off by default.
• Integrations — optional bridges exposing djangorestframework-services specs or a drf-mcp-server registry as agent tools, in process.

Optional extras: [drf-mcp] (bridge a drf-mcp-server registry as a toolset), [spec-tools] (drf-services specs as agent tools), [harness] (pydantic-ai-harness capabilities), and the provider extras [anthropic] / [openai] / [google].

Documentation

https://artui.github.io/django-pydantic-agent/

• Concepts — how a config becomes an agent, and what belongs here versus in a transport.
• Tools — registering tools and what the metadata does.
• Storage — the store contracts, owner scoping, and the reference models.
• Policy — audit and the destructive-tool gate.
• Integrations — DRF specs and MCP tools as agent tools.
• Reference — the full public API.