openbb-pydantic-ai 0.1.6

Pydantic AI adapter for OpenBB SDK. Enables connettion any pydantic-ai agent to the OpenBB Workspace.

OpenBB Pydantic AI Adapter

openbb-pydantic-ai lets any Pydantic AI agent run behind OpenBB Workspace by translating QueryRequest payloads into a Pydantic AI run, exposing Workspace widgets as deferred tools, and streaming native OpenBB SSE events back to the UI.

• Stateless by design: each QueryRequest already carries the full conversation history, widgets, context, and URLs, so the adapter can process requests independently.
• First-class widget tools: every widget becomes a deferred Pydantic AI tool; when the model calls one, the adapter emits copilotFunctionCall events via get_widget_data and waits for the Workspace to return data before resuming.
• Rich event stream: reasoning steps, “Thinking“ traces, tables, charts, and citations are streamed as OpenBB SSE payloads so the Workspace UI can group them into dropdowns automatically.
• Output helpers included: structured model outputs (dicts/lists) are auto-detected and turned into tables or charts, with chart parameter normalization to ensure consistent rendering.

To learn more about the underlying SDK types, see the OpenBB Custom Agent SDK repo and the Pydantic AI UI adapter docs.

Installation

The adapter is published as a lightweight package, install it wherever you build custom agents:

pip install openbb-pydantic-ai
# or with uv
uv add openbb-pydantic-ai

Quick Start (FastAPI)

from fastapi import FastAPI, Request
from pydantic_ai import Agent
from openbb_pydantic_ai import OpenBBAIAdapter, OpenBBDeps

agent = Agent(
    "openai:gpt-5",
    instructions="Be concise and helpful. Only use widget tools for data lookups.",
    deps_type=OpenBBDeps,
)
app = FastAPI()

@app.post("/query")
async def query(request: Request):
    return await OpenBBAIAdapter.dispatch_request(request, agent=agent)

How It Works

1. Request Handling

• OpenBB Workspace calls the /query endpoint with a QueryRequest body
• OpenBBAIAdapter validates it and builds the Pydantic AI message stack
• Workspace context and URLs are injected as system prompts

2. Widget Tool Conversion

• Widgets in the request become deferred tools
• Each call emits a copilotFunctionCall event (via get_widget_data)
• The adapter pauses until Workspace responds with data

3. Event Streaming

Pydantic AI events are wrapped into OpenBB SSE events:

• Text chunks → stream via copilotMessageChunk
• Reasoning steps → appear under the "Step-by-step reasoning" dropdown (including Thinking sections)
• Tables/charts → emitted as copilotMessageArtifact events with correct chart parameters for consistent rendering
• Citations → fire at the end of the run for every widget tool used

Advanced Usage

Need more control? Instantiate the adapter manually:

from openbb_pydantic_ai import OpenBBAIAdapter

run_input = OpenBBAIAdapter.build_run_input(body_bytes)
adapter = OpenBBAIAdapter(agent=agent, run_input=run_input)
async for event in adapter.run_stream():
    yield event  # Already encoded as OpenBB SSE payloads

You can also supply message_history, deferred_tool_results, or on_complete callbacks—any option supported by Agent.run_stream_events() is accepted.

Runtime deps & prompts

• OpenBBDeps bundles widgets (grouped by priority), context rows, relevant URLs, workspace state, timezone, and a serialized state dict you can pass to toolsets or output validators.
• The adapter merges dashboard context and current widget parameter values into the runtime instructions automatically; append your own instructions without re-supplying that context.

Features

Widget Toolsets

• Widgets are grouped by priority (primary, secondary, extra) and exposed through dedicated toolsets so you can gate access if needed.
• Tool names start with openbb_widget_ plus the widget identifier; any redundant openbb_ prefix from the origin is trimmed so names stay concise (e.g., openbb_widget_sandbox_financial_statements). Use build_widget_tool_name to reproduce the routing string exactly.

MCP Tools

• Any tools listed in QueryRequest.tools are exposed as a external MCP toolset, so the model can call the same names the Workspace UI presents.
• Deferred execute_agent_tool results replay on the next request just like widget results, keeping multi-turn streaming consistent.

Deferred Results & Citations

• Pending widget responses provided in the request are replayed before the run starts, making multi-turn workflows seamless.
• Every widget call records a citation via openbb_ai.helpers.cite, emitted as a copilotCitationCollection at the end of the run.

Structured Output Detection

The adapter provides built-in helpers and automatic detection for charts and tables:

• Markdown tables - Stream tabular data as Markdown; Workspace renders them as tables and lets users promote them to widgets.
• openbb_create_chart - Create chart artifacts (line, bar, scatter, pie, donut) with validation. Insert {{place_chart_here}} in the response where the chart should appear; the adapter swaps that placeholder for the rendered artifact while streaming.
• Auto-detection - Dict/list outputs shaped like {"type": "table", "data": [...]} (or just a list of dicts) automatically become tables.
• Flexible chart parameters - Chart outputs tolerate different field spellings (y_keys, yKeys, etc.) and validate required axes before emitting.
• openbb_create_table - Explicitly create a table artifact from structured data when you want predictable column ordering and metadata.

Local Development

This repo ships a UV-based workflow:

uv sync --dev         # install dependencies
uv run pytest      # run the focused test suite
uv run ty check    # type checking (ty)