Skip to content
For the complete documentation index, see llms.txt. This page is also available as Markdown.

Python SDK

Release notes for dnotifier on PyPI.

Install the latest version with pip install dnotifier. Pin a specific version for reproducible builds: pip install dnotifier==1.0.9.

Cross-SDK feature table: Changelog.


1.0.9

Checkpoints, sessions & resume

  • Pass session_id on run_agent / run_workflow to keep conversation and workflow state across turns.

  • Resume, retry, or replay a run with resume, mode, and optional from_checkpoint_id.

  • Optional in-memory checkpointer for local tests; cloud storage when you pass a session_id.

→ Checkpoints, sessions & resume · Python examples


1.0.8

Middleware

  • Agent middleware: hooks around the full run and around model / tool steps (before_agent, after_agent, before_model, after_model, wrap_model_call, before_tool, after_tool, wrap_tool_call, on_error).

  • Workflow middleware: hooks around entry (before_workflow, after_workflow, on_error, on_complete).

  • Register with middleware on define_agent or Workflow (snake_case preferred; camelCase accepted).

→ Middleware · Python examples


1.0.7

MCP server capability

  • capabilities={"mcpServer": True} on agents and workflows.

  • Listen so Claude, Cursor, and other MCP clients can discover and call your agent through DNotifier.

→ Serve agents over MCP

A2A capability

  • capabilities={"a2a": True} on agents and workflows.

  • Expose agents over the open Agent2Agent protocol (Agent Card + JSON-RPC).

→ Agent-to-agent (A2A)

Capabilities

  • Declare what an agent or workflow supports: mcpServer, a2a, tools, rag, memory, workflows, humanApproval, subagents.

→ Define an agent


1.0.5

Vector stores & embeddings

  • Choose knowledge-base storage and embedding model via server_config or per call (vector_store / embedding).

→ Vector stores and embeddings


1.0.3

Agent tools & MCP client

  • notifier.tools.register for local tools; notifier.mcp.connect for HTTP MCP servers.

  • Tool agents via define_agent(..., tools=..., instructions=..., permissions=...).

  • Standalone notifier.run_agent(agent, input=...) for tool agents.

  • permission_enforcement and on_agent_event constructor options.

  • Workflow observability records tool-agent runs in the dashboard.

→ Tools & MCP agents · Local tool example · MCP example

Multi-model providers

  • Pass provider and model on send_ai and agent defaults.

→ Multi-model providers

License

  • Package license remains proprietary (LICENSE + pyproject.toml).


1.0.2

Remote agents

  • Register a remote agent by receiver_id on a workflow and call it with ctx.run_agent the same way as a local agent.

  • Requires WebSocket transport for request/reply across servers.

→ Remote agents


1.0.0

Initial release

  • Async-first Python SDK with parity to JavaScript and Dart packages.

  • WebSocket and HTTP transports.

  • Realtime 1:1 messaging with automatic chunking and reassembly.

  • Chat history: fetch_chat_history, delete_chat_history_message.

  • AI: send_ai, sessions via session_id, AI history fetch/delete.

  • Knowledge base (RAG): document CRUD, semantic search.

  • Workflows: define_agent, Workflow, WorkflowContext, run_workflow.

  • Optional workflow observability when observability=True.

  • Session logging with logs=True.

→ Install — Python → Workflows overview

Requirements

  • Python 3.9+

  • Dependencies: httpx, websockets


Last updated