From 59825e894d678face073ae2ae9535d73615b10e6 Mon Sep 17 00:00:00 2001 From: David Montague <35119617+dmontagu@users.noreply.github.com> Date: Tue, 19 Nov 2024 09:20:27 -0700 Subject: [PATCH] Index improvements --- docs/index.md | 96 ++++++++++++++++++++++++++++++++++--------------- docs/install.md | 2 +- 2 files changed, 69 insertions(+), 29 deletions(-) diff --git a/docs/index.md b/docs/index.md index e938e01e90..ed2727d308 100644 --- a/docs/index.md +++ b/docs/index.md @@ -2,30 +2,30 @@ --8<-- "docs/.partials/index-header.html" -When I first found FastAPI, I got it immediately, I was excited to find something so genuinely innovative and yet ergonomic built on Pydantic. +When I first found FastAPI, I got it immediately. I was excited to find something so innovative and ergonomic built on Pydantic. -Virtually every Agent Framework and LLM library in Python uses Pydantic, but when we came to use Gen AI in [Pydantic Logfire](https://pydantic.dev/logfire), I couldn't find anything that gave me the same feeling. +Virtually every Agent Framework and LLM library in Python uses Pydantic, but when we began to use LLMs in [Pydantic Logfire](https://pydantic.dev/logfire), I couldn't find anything that gave me the same feeling. PydanticAI is a Python Agent Framework designed to make it less painful to build production grade applications with Generative AI. ## Why use PydanticAI -* Built by the team behind Pydantic (the validation layer of the OpenAI SDK, the Anthropic SDK, Langchain, LlamaIndex, AutoGPT, Transformers, Instructor and many more) -* Multi-model — currently with OpenAI and Gemini are support, Anthropic [coming soon](https://github.com/pydantic/pydantic-ai/issues/63), simply interface to implement other models or adapt existing ones +* Built by the team behind Pydantic (the validation layer of the OpenAI SDK, the Anthropic SDK, LangChain, LlamaIndex, AutoGPT, Transformers, Instructor and many more) +* Model-agnostic — currently both OpenAI and Gemini are supported, and Anthropic [is coming soon](https://github.com/pydantic/pydantic-ai/issues/63). And there is a simple interface to implement and customize support for other models. * Type-safe -* Built on tried and tested best practices in Python +* Control flow and composing agents is done with vanilla python, allowing you to make use of the same Python development best practices you'd use in any other (non-AI) project * [Structured response](results.md#structured-result-validation) validation with Pydantic -* [Streamed responses](results.md#streamed-results) , including validation of streamed structured responses with Pydantic -* Novel, type-safe [dependency injection system](dependencies.md) -* [Logfire integration](logfire.md) for debugging and performance monitoring +* [Streamed responses](results.md#streamed-results) , including validation of streamed _structured_ responses with Pydantic +* Novel, type-safe [dependency injection system](dependencies.md), useful for testing and eval-driven iterative development +* [Logfire integration](logfire.md) for debugging and monitoring the performance and general behavior of your LLM-powered application !!! example "In Beta" - PydanticAI is in early beta, the API is subject to change and there's a lot more to do. + PydanticAI is in early beta, the API is still subject to change and there's a lot more to do. [Feedback](https://github.com/pydantic/pydantic-ai/issues) is very welcome! ## Hello World Example -Here's a very minimal example of PydanticAI. +Here's a minimal example of PydanticAI: ```py title="hello_world.py" from pydantic_ai import Agent @@ -42,23 +42,22 @@ The first known use of "hello, world" was in a 1974 textbook about the C program """ ``` -1. Define a very simple agent, here we configure the agent to use [Gemini 1.5's Flash](api/models/gemini.md) model, you can also set the model when running the agent. -2. Static [system prompts](agents.md#system-prompts) can be registered as keyword arguments to the agent. For more complex system prompts, see the example below. -3. [Run the agent](agents.md#running-agents) synchronously, conducting a conversation with the LLM, here the exchange should be very short: PydanticAI will send the system prompt and the user query to the LLM, the model will return a text response. +1. Define a very simple agent — here we configure the agent to use [Gemini 1.5's Flash](api/models/gemini.md) model, but you can also set the model when running the agent. +2. Register a static [system prompt](agents.md#system-prompts) using a keyword argument to the agent. For more complex dynamically-generated system prompts, see the example below. +3. [Run the agent](agents.md#running-agents) synchronously, conducting a conversation with the LLM. Here the exchange should be very short: PydanticAI will send the system prompt and the user query to the LLM, the model will return a text response. -4. _(This example is complete, it can be run "as is")_ +_(This example is complete, it can be run "as is")_ -Not very interesting yet, but we can easily add retrievers, dynamic system prompts and structured responses to build more powerful agents. +Not very interesting yet, but we can easily add "retrievers", dynamic system prompts, and structured responses to build more powerful agents. ## Retrievers & Dependency Injection Example -Small but complete example of using PydanticAI to build a support agent for a bank. +Here is a concise example using PydanticAI to build a support agent for a bank: ```py title="bank_support.py" from dataclasses import dataclass from pydantic import BaseModel, Field - from pydantic_ai import Agent, CallContext from bank_database import DatabaseConn @@ -72,7 +71,7 @@ class SupportDependencies: # (3)! class SupportResult(BaseModel): # (13)! support_advice: str = Field(description='Advice returned to the customer') - block_card: bool = Field(description='Whether to block their') + block_card: bool = Field(description="Whether to block the customer's card") risk: int = Field(description='Risk level of query', ge=0, le=10) @@ -124,27 +123,68 @@ async def main(): """ ``` -1. An [agent](agents.md) that acts as first-tier support in a bank, agents are generic in the type of dependencies they take and the type of result they return, in this case support agent has type `#!python Agent[SupportDependencies, SupportResult]`. +1. This [agent](agents.md) will act as first-tier support in a bank. Agents are generic in the type of dependencies they accept and the type of result they return. In this case, the support agent has type `#!python Agent[SupportDependencies, SupportResult]`. 2. Here we configure the agent to use [OpenAI's GPT-4o model](api/models/openai.md), you can also set the model when running the agent. -3. The `SupportDependencies` dataclass is used to pass data, connections and logic into the model that will be needed when running [system prompts](agents.md#system-prompts) and [retrievers](agents.md#retrievers). PydanticAI's system of dependency injection provides a powerful, type safe way to customise the behaviour of your agents, including when unit tests and evals. -4. Static [system prompts](agents.md#system-prompts) can be registered as [keyword arguments][pydantic_ai.Agent.__init__] to the agent. -5. dynamic [system prompts](agents.md#system-prompts) can be registered with the [`@agent.system_prompt`][pydantic_ai.Agent.system_prompt] decorator and benefit from dependency injection. Dependencies are carried via the [`CallContext`][pydantic_ai.dependencies.CallContext] argument, this is parameterised with the `deps_type` from above, if the type annotation here is wrong, static type checkers will catch it. -6. [Retrievers](agents.md#retrievers) let you register "tools" which the LLM may call while responding to a user. Again dependencies are carried via [`CallContext`][pydantic_ai.dependencies.CallContext], any other arguments become the tool schema passed to the LLM, Pydantic is used to validate these arguments, errors are passed back to the LLM so it can retry. -7. The docstring is also passed to the LLM as a description of the tool. Parameter descriptions are [extracted](agents.md#retrievers-tools-and-schema) from the docstring and added to the tool schema sent to the LLM. -8. [Run the agent](agents.md#running-agents) asynchronously, conducting a conversation with the LLM until a final response is reached. Even in this fair simply case, the agent will exchange multiple messages with the LLM as retrievers are called to each a result. +3. The `SupportDependencies` dataclass is used to pass data, connections, and logic into the model that will be needed when running [system prompt](agents.md#system-prompts) and [retriever](agents.md#retrievers) functions. PydanticAI's system of dependency injection provides a type-safe way to customise the behaviour of your agents, and can be especially useful when running unit tests and evals. +4. Static [system prompts](agents.md#system-prompts) can be registered with the [`system_prompt` keyword argument][pydantic_ai.Agent.__init__] to the agent. +5. Dynamic [system prompts](agents.md#system-prompts) can be registered with the [`@agent.system_prompt`][pydantic_ai.Agent.system_prompt] decorator, and can make use of dependency injection. Dependencies are carried via the [`CallContext`][pydantic_ai.dependencies.CallContext] argument, which is parameterized with the `deps_type` from above. If the type annotation here is wrong, static type checkers will catch it. +6. [Retrievers](agents.md#retrievers) let you register "tools" which the LLM may call while responding to a user. Again, dependencies are carried via [`CallContext`][pydantic_ai.dependencies.CallContext], and any other arguments become the tool schema passed to the LLM. Pydantic is used to validate these arguments, and errors are passed back to the LLM so it can retry. +7. The docstring of a retriever also passed to the LLM as a description of the tool. Parameter descriptions are [extracted](agents.md#retrievers-tools-and-schema) from the docstring and added to the tool schema sent to the LLM. +8. [Run the agent](agents.md#running-agents) asynchronously, conducting a conversation with the LLM until a final response is reached. Even in this fairly simple case, the agent will exchange multiple messages with the LLM as retrievers are called to retrieve a result. 9. The response from the agent will, be guaranteed to be a `SupportResult`, if validation fails [reflection](agents.md#reflection-and-self-correction) will mean the agent is prompted to try again. 10. The result will be validated with Pydantic to guarantee it is a `SupportResult`, since the agent is generic, it'll also be typed as a `SupportResult` to aid with static type checking. 11. In a real use case, you'd add many more retrievers and a longer system prompt to the agent to extend the context it's equipped with and support it can provide. 12. This is a simple sketch of a database connection, used to keep the example short and readable. In reality, you'd be connecting to an external database (e.g. PostgreSQL) to get information about customers. 13. This [Pydantic](https://docs.pydantic.dev) model is used to constrain the structured data returned by the agent. From this simple definition, Pydantic builds teh JSON Schema that tells the LLM how to return the data, and performs validation to guarantee the data is correct at the end of the conversation. +To help make things more clear, here is a diagram of what is happening in the `#!python await support_agent.run('What is my balance?', deps=deps)` call within `main`: +```mermaid +sequenceDiagram + participant DatabaseConn + participant Agent + participant LLM + + Note over Agent: Dynamic system prompt
add_customer_name() + Agent ->> DatabaseConn: Retrieve customer name + activate DatabaseConn + DatabaseConn -->> Agent: "John" + deactivate DatabaseConn + + Note over Agent: User query + + Agent ->> LLM: Request
System: "You are a support agent..."
System: "The customer's name is John"
User: "What is my balance?" + activate LLM + Note over LLM: LLM decides to use a retriever + LLM ->> Agent: Call retriever
customer_balance() + deactivate LLM + activate Agent + Note over Agent: Retrieve account balance + + Agent ->> DatabaseConn: Retrieve balance
Include pending + activate DatabaseConn + DatabaseConn -->> Agent: "$123.45" + deactivate DatabaseConn + + Agent -->> LLM: ToolReturn
"$123.45" + deactivate Agent + activate LLM + Note over LLM: LLM processes response + + LLM ->> Agent: StructuredResponse
SupportResult + deactivate LLM + activate Agent + Note over Agent: Support session complete + deactivate Agent +``` + + !!! tip "Complete `bank_support.py` example" - This example is incomplete for the sake of brevity (the definition of `DatabaseConn` is missing); you can find a complete `bank_support.py` example [here](examples/bank-support.md). + The code included here is incomplete for the sake of brevity (the definition of `DatabaseConn` is missing); you can find the complete `bank_support.py` example [here](examples/bank-support.md). ## Next Steps -To try PydanticAI yourself, follow instructions [in examples](examples/index.md). +To try PydanticAI yourself, follow the instructions [in the examples](examples/index.md). -Read the conceptual [documentation](agents.md) to learn more about building applications with PydanticAI. +Read the [conceptual documentation](agents.md) to learn more about building applications with PydanticAI. Read the [API Reference](api/agent.md) to understand PydanticAI's interface. diff --git a/docs/install.md b/docs/install.md index 1db34c429d..cad4782036 100644 --- a/docs/install.md +++ b/docs/install.md @@ -34,4 +34,4 @@ To install extra dependencies required to run examples, install the `examples` o pip/uv-add 'pydantic-ai[examples]' ``` -For next steps, follow instructions [in examples](examples/index.md). +For next steps, follow the instructions [in the examples](examples/index.md).