pydantic-ai-toolkits 0.0.2

Modular, independent tool kits (filesystem, SQL, pandas, memory, RAG, ...) for pydantic-ai agents.

pydantic-ai-toolkits

If you've used pydantic-ai, you already know the feeling: it's the first agent framework that feels like a regular Python library. Typed RunContext, a clean FunctionToolset protocol, model providers swapped behind one string. After a decade of frameworks that pretended to be Pythonic, this one actually is.

And then you sit down to wire up your agent, and realize: pydantic-ai will happily call any tool you give it — but the tools themselves are still on you. Want the agent to read a file? You write the sandbox. Run a SQL query? You write the read-only guard and the schema introspection. Search local documents? Text splitter, vector index, cosine math, persistence — all you.

After the third project where you wrote those by hand, the shape stops being interesting. This is them, written once:

pip install pydantic-ai-toolkits

from pydantic_ai import Agent
from pydantic_ai_toolkits import (
    FilesystemToolkit, SQLToolkit, PandasToolkit, MemoryToolkit, RAGToolkit,
)

agent = Agent(
    "openai:gpt-4o-mini",
    toolsets=[
        FilesystemToolkit(root="./workspace", read_only=False),
        SQLToolkit(dsn="postgresql://user:pwd@localhost/app"),
        PandasToolkit(),
        MemoryToolkit(storage_path="./memory.json"),
        RAGToolkit(embedder=my_embedder),
    ],
    system_prompt="You are a data assistant.",
)

print(agent.run_sync("Read README.md from the workspace and summarise it.").output)

That's the whole story. Five toolkits, one toolsets=[...], no new framework on top of pydantic-ai — each toolkit is a thin FunctionToolset subclass, exactly what pydantic-ai expects.

---

Install

pip install pydantic-ai-toolkits

The base install gives you FilesystemToolkit and MemoryToolkit (stdlib only). The rest are opt-in so you only pull in what you use:

pip install "pydantic-ai-toolkits[sql]"      # + SQLAlchemy
pip install "pydantic-ai-toolkits[pandas]"   # + pandas + pyarrow
pip install "pydantic-ai-toolkits[rag]"      # + numpy
pip install "pydantic-ai-toolkits[all]"      # everything

Extras are independent — picking up one doesn't pull in the others. Details: docs/INSTALL.md.

---

Toolkits

Toolkit	What an agent can do with it	Docs
FilesystemToolkit	List, read, write, append, delete, mkdir, stat, glob — under one sandbox root, with path-escape rejection and an optional read-only mode.	docs/FILESYSTEM.md
SQLToolkit	List tables/views, describe schemas, run parameterised reads, optional execute for writes. Single-statement read-only by default.	docs/SQL.md
PandasToolkit	Manage a named dataframe registry; load CSV/Parquet; head / describe / schema / query / aggregate / value_counts.	docs/PANDAS.md
MemoryToolkit	Append/read/search messages; key-value scratchpad facts; optional atomic JSON persistence and per-namespace isolation.	docs/MEMORY.md
RAGToolkit	Recursive character text splitter + in-memory numpy vector index with cosine search and per-document delete.	docs/RAG.md

Tiny snippets to taste each one:

# Filesystem — sandbox a workspace, then let the agent edit files
FilesystemToolkit(root="./workspace", read_only=False)

# SQL — read-only Postgres
SQLToolkit(dsn="postgresql://user:pwd@localhost/app")

# Pandas — start with an empty registry, agent loads CSVs as needed
PandasToolkit()

# Memory — persisted scratchpad, 200-message cap
MemoryToolkit(storage_path="./memory.json", max_messages=200)

# RAG — bring your own embedder
RAGToolkit(embedder=lambda texts: [embed(t) for t in texts])

Runnable end-to-end scripts live in examples/ (see docs/EXAMPLES.md).

---

What's not in here (and where to find it)

Before reaching for a toolkit here, check whether pydantic-ai already ships the capability you need — most of the time it does:

Need	Use this
Web search	pydantic_ai.common_tools.{duckduckgo, exa, tavily} or native_tools.WebSearchTool
Fetch a page and convert to Markdown	pydantic_ai.common_tools.web_fetch.web_fetch_tool
Provider-side code execution / image gen	pydantic_ai.native_tools.{CodeExecutionTool, ImageGenerationTool, FileSearchTool}
Provider-managed long-term memory	pydantic_ai.native_tools.MemoryTool
Third-party MCP server (fs, postgres, …)	pydantic_ai.mcp.MCPServerStdio / MCPServerHTTP

This package fills the gaps that aren't on that list — local sandboxed filesystem access, generic SQL via SQLAlchemy, in-memory dataframe ops, self-hosted conversation memory, and local RAG without an external vector DB.

---

Write your own toolkit

A toolkit is a BaseToolkit subclass whose public methods carry @tool:

from pydantic_ai_toolkits import BaseToolkit, tool


class WeatherToolkit(BaseToolkit):
    """Look up current weather for a configurable provider."""

    def __init__(self, api_key: str, units: str = "metric") -> None:
        self.api_key = api_key
        self.units = units
        super().__init__()          # MUST be last — scans @tool methods

    @tool
    def current_temperature(self, city: str) -> float:
        """Return the current temperature for `city` in the configured units."""
        ...

Full rules, schema-mapping table, and the contribution checklist: docs/WRITING.md, AGENTS.md.

---

Install from git (latest unreleased)

pip install git+https://github.com/wachawo/pydantic-ai-toolkits.git

Install from source (local development)

git clone [e-mail]:wachawo/pydantic-ai-toolkits.git
cd pydantic-ai-toolkits
python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[all]"
pip install -r requirements-dev.txt
pytest --cov          # 80% coverage gate

---

Documentation

Rendered with MkDocs at docs/:

• Overview
• Install
• Toolkits: Filesystem, SQL, Pandas, Memory, RAG
• Write your own
• Examples
• Building the docs site
• Changelog

If something's off — a missing convenience method, an awkward signature, a default that doesn't match your use case — the API is intentionally small. Open an issue on GitHub and say what you'd want instead.

---

License

MIT.