From 4de1bd6e666ba9305d259f8f043608c4f9c6aac9 Mon Sep 17 00:00:00 2001 From: Lauren Hirata Singh Date: Tue, 8 Jul 2025 12:33:14 -0400 Subject: [PATCH] docs: cleanup (#5401) * docs: cleanup * fix nav * fix * fix nits --- docs/_scripts/extract_images.py | 70 -- docs/docs/additional-resources/index.md | 11 + docs/docs/agents/agents.md | 2 +- docs/docs/cloud/concepts/threads.md | 12 - docs/docs/examples/index.md | 23 + docs/docs/guides/index.md | 41 + .../tutorials/multi_agent/agent_supervisor.md | 2 +- docs/docs/tutorials/overview.md | 21 - docs/docs/tutorials/sql-agent.ipynb | 759 ------------------ docs/mkdocs.yml | 3 + 10 files changed, 80 insertions(+), 864 deletions(-) delete mode 100644 docs/_scripts/extract_images.py create mode 100644 docs/docs/additional-resources/index.md delete mode 100644 docs/docs/cloud/concepts/threads.md create mode 100644 docs/docs/examples/index.md create mode 100644 docs/docs/guides/index.md delete mode 100644 docs/docs/tutorials/overview.md delete mode 100644 docs/docs/tutorials/sql-agent.ipynb diff --git a/docs/_scripts/extract_images.py b/docs/_scripts/extract_images.py deleted file mode 100644 index 781ba0a22..000000000 --- a/docs/_scripts/extract_images.py +++ /dev/null @@ -1,70 +0,0 @@ -#!/usr/bin/env python3 -""" -Script to extract images from the graph-api.ipynb notebook and save them to assets folder. -""" - -import json -import base64 -import os -from pathlib import Path - -def extract_images_from_notebook(notebook_path, assets_dir): - """Extract images from notebook and save them to assets directory.""" - - # Read the notebook - with open(notebook_path, 'r') as f: - notebook = json.load(f) - - # Create assets directory if it doesn't exist - os.makedirs(assets_dir, exist_ok=True) - - image_count = 0 - - # Process each cell - for cell_idx, cell in enumerate(notebook['cells']): - if cell['cell_type'] == 'code': - # Check if this cell contains draw_mermaid_png - source = ''.join(cell.get('source', [])) - if 'draw_mermaid_png' in source: - print(f"Found draw_mermaid_png in cell {cell_idx}") - - # Check for outputs with images - if 'outputs' in cell: - for output_idx, output in enumerate(cell['outputs']): - if output.get('output_type') == 'display_data': - data = output.get('data', {}) - - # Check for PNG data - if 'image/png' in data: - png_data = data['image/png'] - - # Decode base64 data - try: - image_bytes = base64.b64decode(png_data) - - # Generate filename - image_count += 1 - filename = f"graph_api_image_{image_count}.png" - filepath = os.path.join(assets_dir, filename) - - # Save the image - with open(filepath, 'wb') as img_file: - img_file.write(image_bytes) - - print(f"Saved image: {filepath}") - - except Exception as e: - print(f"Error decoding image {image_count}: {e}") - - print(f"Extracted {image_count} images to {assets_dir}") - return image_count - -if __name__ == "__main__": - notebook_path = "docs/docs/how-tos/graph-api.ipynb" - assets_dir = "docs/docs/how-tos/assets" - - if os.path.exists(notebook_path): - count = extract_images_from_notebook(notebook_path, assets_dir) - print(f"Successfully extracted {count} images") - else: - print(f"Notebook not found: {notebook_path}") \ No newline at end of file diff --git a/docs/docs/additional-resources/index.md b/docs/docs/additional-resources/index.md new file mode 100644 index 000000000..6e4e7a93b --- /dev/null +++ b/docs/docs/additional-resources/index.md @@ -0,0 +1,11 @@ +# Additional resources + +This section contains additional resources for LangGraph. + +- [Community agents](../agents/prebuilt.md): A collection of prebuilt libraries that you can use in your LangGraph applications. +- [LangGraph Academy](https://academy.langchain.com/courses/intro-to-langgraph): A collection of courses that teach you how to use LangGraph. +- [Case studies](../adopters.md): A collection of case studies that show how LangGraph is used in production. +- [FAQ](../concepts/faq.md): A collection of frequently asked questions about LangGraph. +- [llms.txt](../llms-txt-overview.md): A list of documentation files in the `llms.txt` format that allow LLMs and agents to access our documentation. +- [LangChain Forum](https://forum.langchain.com/): A place to ask questions and get help from other LangGraph users. +- [Troubleshooting](../troubleshooting/errors/index.md.md): A collection of troubleshooting guides for common issues. \ No newline at end of file diff --git a/docs/docs/agents/agents.md b/docs/docs/agents/agents.md index b00184266..04ee419ce 100644 --- a/docs/docs/agents/agents.md +++ b/docs/docs/agents/agents.md @@ -52,7 +52,7 @@ agent.invoke( ) ``` -1. Define a tool for the agent to use. Tools can be defined as vanilla Python functions. For more advanced tool usage and customization, check the [tools](./tools.md) page. +1. Define a tool for the agent to use. Tools can be defined as vanilla Python functions. For more advanced tool usage and customization, check the [tools](../how-tos/tool-calling.md) page. 2. Provide a language model for the agent to use. To learn more about configuring language models for the agents, check the [models](./models.md) page. 3. Provide a list of tools for the model to use. 4. Provide a system prompt (instructions) to the language model used by the agent. diff --git a/docs/docs/cloud/concepts/threads.md b/docs/docs/cloud/concepts/threads.md deleted file mode 100644 index ffd48faa8..000000000 --- a/docs/docs/cloud/concepts/threads.md +++ /dev/null @@ -1,12 +0,0 @@ -# Threads - -A thread contains the accumulated state of a sequence of [runs](../../concepts/assistants.md#execution). When a run is executed, the [state](../../concepts/low_level.md#state) of the underlying graph of the assistant will be persisted to the thread. - -A thread's current and historical state can be retrieved. To persist state, a thread must be created prior to executing a run. - -The state of a thread at a particular point in time is called a [checkpoint](../../concepts/persistence.md#checkpoints). Checkpoints are persisted and can be used to restore the state of a thread at a later time. - -## Learn more - -* For more on threads and checkpoints, see this section of the [LangGraph conceptual guide](../../concepts/persistence.md). -* The LangGraph Platform API provides several endpoints for creating and managing threads and thread state. See the [API reference](../../cloud/reference/api/api_ref.html#tag/threads) for more details. diff --git a/docs/docs/examples/index.md b/docs/docs/examples/index.md new file mode 100644 index 000000000..84c721ac3 --- /dev/null +++ b/docs/docs/examples/index.md @@ -0,0 +1,23 @@ +# Examples + +The pages in this section provide end-to-end examples for the following topics: + +## General + +- [Template Applications](../concepts/template_applications.md): Create a LangGraph application from a template. +- [Agentic RAG](./rag/langgraph_agentic_rag.md): Build a retrieval agent that can decide when to use a retriever tool. +- [Agent Supervisor](./multi_agent/agent_supervisor.md): Build a supervisor agent that can manage a team of agents. +- [SQL agent](./sql/sql-agent.md): Build a SQL agent that can execute SQL queries and return the results. +- [Prebuilt chat UI](../agents/ui.md): Use a prebuilt chat UI to interact with any LangGraph agent. +- [Graph runs in LangSmith](../how-tos/run-id-langsmith.md): Use LangSmith to track and analyze graph runs. + +## LangGraph Platform + +- [Set up custom authentication](./auth/getting_started.md): Set up custom authentication for your LangGraph application. +- [Make conversations private](./auth/resource_auth.md): Make conversations private by using resource-based authentication. +- [Connect an authentication provider](./auth/add_auth_server.md): Connect an authentication provider to your LangGraph application. +- [Rebuild graph at runtime](../cloud/deployment/graph_rebuild.md): Rebuild a graph at runtime. +- [Use RemoteGraph](../how-tos/use-remote-graph.md): Use RemoteGraph to deploy your LangGraph application to a remote server. +- [Deploy CrewAI, AutoGen, and other frameworks](../how-tos/autogen-integration.md): Deploy CrewAI, AutoGen, and other frameworks with LangGraph. +- [Integrate LangGraph into a React app](../cloud/how-tos/use_stream_react.md) +- [Implement Generative User Interfaces with LangGraph](../cloud/how-tos/generative_ui_react.md) \ No newline at end of file diff --git a/docs/docs/guides/index.md b/docs/docs/guides/index.md new file mode 100644 index 000000000..1d57f63ea --- /dev/null +++ b/docs/docs/guides/index.md @@ -0,0 +1,41 @@ +# Guides + +The pages in this section provide a conceptual overview and how-tos for the following topics: + +## LangGraph APIs + +- [Graph API](../concepts/low_level.md): Use the Graph API to define workflows using a graph paradigm. +- [Functional API](../concepts/functional_api.md): Use Functional API to build workflows using a functional paradigm without thinking about the graph structure. +- [Runtime](../concepts/pregel.md): Pregel implements LangGraph's runtime, managing the execution of LangGraph applications. + +## Core capabilities + +These capabilities are available in both LangGraph OSS and the LangGraph Platform. + +- [Streaming](../concepts/streaming.md): Stream outputs from a LangGraph graph. +- [Persistence](../concepts/persistence.md): Persist the state of a LangGraph graph. +- [Durable execution](../concepts/durable_execution.md): Save progress at key points in the graph execution. +- [Memory](../concepts/memory.md): Remember information about previous interactions. +- [Context](../agents/context.md): Pass outside data to a LangGraph graph to provide context for the graph execution. +- [Models](../agents/models.md): Integrate various LLMs into your LangGraph application. +- [Tools](../concepts/tools.md): Interface directly with external systems. +- [Human-in-the-loop](../concepts/human_in_the_loop.md): Enable human intervention at any point in a workflow. +- [Breakpoints](../concepts/breakpoints.md): Pause the execution of a LangGraph graph at a specific point. +- [Time travel](../concepts/time-travel.md): Travel back in time to a specific point in the execution of a LangGraph graph. +- [Subgraphs](../concepts/subgraphs.md): Build modular graphs. +- [Multi-agent](../concepts/multi_agent.md): Break down a complex workflow into multiple agents. +- [MCP](../concepts/mcp.md): Use MCP servers in a LangGraph graph. +- [Evaluation](../agents/evals.md): Use LangSmith to evaluate your graph's performance. + +## Platform-only capabilities + +These capabilities are only available in [LangGraph Platform](../concepts/langgraph_platform.md). + +- [Authentication and access control](../concepts/auth.md): Authenticate and authorize users to access a Langraph graph. +- [Assistants](../concepts/assistants.md): Build assistants that can be used to interact with a LangGraph graph. +- [Double-texting](../concepts/double_texting.md): Handle double-texting (consecutive messages before a first response is returned) in a LangGraph graph. +- [Webhooks](../cloud/concepts/webhooks.md): Send webhooks to a LangGraph graph. +- [Cron jobs](../cloud/concepts/cron_jobs.md): Schedule jobs to run at a specific time. +- [Server customization](../how-tos/http/custom_lifespan.md): Customize the server that runs a LangGraph graph. +- [Data management](../cloud/concepts/data_storage_and_privacy.md): Manage data in a LangGraph graph. +- [Deployment](../concepts/deployment_options.md): Deploy a LangGraph graph to a server. \ No newline at end of file diff --git a/docs/docs/tutorials/multi_agent/agent_supervisor.md b/docs/docs/tutorials/multi_agent/agent_supervisor.md index ba8f84ff6..71335c8db 100644 --- a/docs/docs/tutorials/multi_agent/agent_supervisor.md +++ b/docs/docs/tutorials/multi_agent/agent_supervisor.md @@ -206,7 +206,7 @@ Name: tavily_search ### Math agent -For math agent tools we will use [vanilla Python functions](../../agents/tools.md#define-simple-tools): +For math agent tools we will use [vanilla Python functions](../../how-tos/tool-calling.md#define-a-tool): ```python def add(a: float, b: float): diff --git a/docs/docs/tutorials/overview.md b/docs/docs/tutorials/overview.md deleted file mode 100644 index 6eacbd4e4..000000000 --- a/docs/docs/tutorials/overview.md +++ /dev/null @@ -1,21 +0,0 @@ -# Examples - -The pages in this section provide end-to-end examples for the following topics: - -## General - -- [Agentic RAG](./rag/langgraph_adaptive_rag.ipynb) -- [Agent Supervisor](./multi_agent/agent_supervisor.md) -- [SQL agent](./sql/sql-agent.md) -- [Graph runs in LangSmith](../how-tos/run-id-langsmith.md) - -## LangGraph Platform - -- [Set up custom authentication](./auth/getting_started.md) -- [Make conversations private](./auth/resource_auth.md) -- [Connect an authentication provider](./auth/add_auth_server.md) -- [Rebuild graph at runtime](../cloud/deployment/graph_rebuild.md) -- [Use RemoteGraph](../how-tos/use-remote-graph.md) -- [Deploy CrewAI, AutoGen, and other frameworks](../how-tos/autogen-integration.md) -- [Integrate LangGraph into a React app](../cloud/how-tos/use_stream_react.md) -- [Implement Generative User Interfaces with LangGraph](../cloud/how-tos/generative_ui_react.md) \ No newline at end of file diff --git a/docs/docs/tutorials/sql-agent.ipynb b/docs/docs/tutorials/sql-agent.ipynb deleted file mode 100644 index a06626f40..000000000 --- a/docs/docs/tutorials/sql-agent.ipynb +++ /dev/null @@ -1,759 +0,0 @@ -{ - "cells": [ - { - "cell_type": "markdown", - "id": "cb71c4a7-9f99-4f08-be34-b3ba81ac075f", - "metadata": {}, - "source": [ - "# Build a SQL agent\n", - "\n", - "In this tutorial, we will walk through how to build an agent that can answer questions about a SQL database.\n", - "\n", - "At a high level, the agent will:\n", - "\n", - "1. Fetch the available tables from the database\n", - "2. Decide which tables are relevant to the question\n", - "3. Fetch the schemas for the relevant tables\n", - "4. Generate a query based on the question and information from the schemas\n", - "5. Double-check the query for common mistakes using an LLM\n", - "6. Execute the query and return the results\n", - "7. Correct mistakes surfaced by the database engine until the query is successful\n", - "8. Formulate a response based on the results\n", - "\n", - "!!! warning \"Security note\"\n", - "\n", - " Building Q&A systems of SQL databases requires executing model-generated SQL queries. There are inherent risks in doing this. Make sure that your database connection permissions are always scoped as narrowly as possible for your agent's needs. This will mitigate though not eliminate the risks of building a model-driven system.\n", - "\n", - "## 1. Setup\n", - "\n", - "Let's first install some dependencies. This tutorial uses SQL database and tool abstractions from [langchain-community](https://python.langchain.com/docs/concepts/architecture/#langchain-community). We will also require a LangChain [chat model](https://python.langchain.com/docs/concepts/chat_models/)." - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "1b613286-41c3-4ca5-9514-36aac0121a5a", - "metadata": {}, - "outputs": [], - "source": [ - "%%capture --no-stderr\n", - "%pip install -U langgraph langchain_community \"langchain[openai]\"" - ] - }, - { - "cell_type": "markdown", - "id": "f5258ad0-39e1-418e-bddb-56e9aab1b43a", - "metadata": {}, - "source": [ - "
\n", - "

Set up LangSmith for LangGraph development

\n", - "

\n", - " Sign up for LangSmith to quickly spot issues and improve the performance of your LangGraph projects. LangSmith lets you use trace data to debug, test, and monitor your LLM apps built with LangGraph — read more about how to get started here. \n", - "

\n", - "
" - ] - }, - { - "cell_type": "markdown", - "id": "62886241-caa7-40bd-8867-096a97acaf40", - "metadata": {}, - "source": [ - "### Select a LLM\n", - "\n", - "First we [initialize our LLM](https://python.langchain.com/docs/how_to/chat_models_universal_init/). Any model supporting [tool-calling](https://python.langchain.com/docs/integrations/chat/#featured-providers) should work. We use OpenAI below." - ] - }, - { - "cell_type": "code", - "execution_count": 1, - "id": "2b981efc-bcc9-49f4-aa66-4c15a76210a4", - "metadata": {}, - "outputs": [], - "source": [ - "from langchain.chat_models import init_chat_model\n", - "\n", - "llm = init_chat_model(\"openai:gpt-4.1\")" - ] - }, - { - "cell_type": "markdown", - "id": "804ca604-7ac4-4229-a3e9-3a9825eeefa4", - "metadata": {}, - "source": [ - "### Configure the database\n", - "\n", - "We will be creating a SQLite database for this tutorial. SQLite is a lightweight database that is easy to set up and use. We will be loading the `chinook` database, which is a sample database that represents a digital media store.\n", - "Find more information about the database [here](https://www.sqlitetutorial.net/sqlite-sample-database/).\n", - "\n", - "For convenience, we have hosted the database (`Chinook.db`) on a public GCS bucket." - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "b58ba6b7-5f80-4970-bc29-961de0e24fbd", - "metadata": {}, - "outputs": [], - "source": [ - "import requests\n", - "\n", - "url = \"https://storage.googleapis.com/benchmarks-artifacts/chinook/Chinook.db\"\n", - "\n", - "response = requests.get(url)\n", - "\n", - "if response.status_code == 200:\n", - " # Open a local file in binary write mode\n", - " with open(\"Chinook.db\", \"wb\") as file:\n", - " # Write the content of the response (the file) to the local file\n", - " file.write(response.content)\n", - " print(\"File downloaded and saved as Chinook.db\")\n", - "else:\n", - " print(f\"Failed to download the file. Status code: {response.status_code}\")" - ] - }, - { - "cell_type": "markdown", - "id": "5ae005fb-e7b0-43f1-95b2-565d19c82610", - "metadata": {}, - "source": [ - "We will use a handy SQL database wrapper available in the `langchain_community` package to interact with the database. The wrapper provides a simple interface to execute SQL queries and fetch results:" - ] - }, - { - "cell_type": "code", - "execution_count": 2, - "id": "18822c9b-6df2-4a5c-a12d-046d2b2131c9", - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Dialect: sqlite\n", - "Available tables: ['Album', 'Artist', 'Customer', 'Employee', 'Genre', 'Invoice', 'InvoiceLine', 'MediaType', 'Playlist', 'PlaylistTrack', 'Track']\n", - "Sample output: [(1, 'AC/DC'), (2, 'Accept'), (3, 'Aerosmith'), (4, 'Alanis Morissette'), (5, 'Alice In Chains')]\n" - ] - } - ], - "source": [ - "from langchain_community.utilities import SQLDatabase\n", - "\n", - "db = SQLDatabase.from_uri(\"sqlite:///Chinook.db\")\n", - "\n", - "print(f\"Dialect: {db.dialect}\")\n", - "print(f\"Available tables: {db.get_usable_table_names()}\")\n", - "print(f'Sample output: {db.run(\"SELECT * FROM Artist LIMIT 5;\")}')" - ] - }, - { - "cell_type": "markdown", - "id": "3bf47b6c-30dd-405f-8a03-21327af3c764", - "metadata": {}, - "source": [ - "### Tools for database interactions\n", - "\n", - "`langchain-community` implements some built-in tools for interacting with our `SQLDatabase`, including tools for listing tables, reading table schemas, and checking and running queries:" - ] - }, - { - "cell_type": "code", - "execution_count": 3, - "id": "aab9e00a-aa28-494a-a95a-a6638ea03bd1", - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "sql_db_query: Input to this tool is a detailed and correct SQL query, output is a result from the database. If the query is not correct, an error message will be returned. If an error is returned, rewrite the query, check the query, and try again. If you encounter an issue with Unknown column 'xxxx' in 'field list', use sql_db_schema to query the correct table fields.\n", - "\n", - "sql_db_schema: Input to this tool is a comma-separated list of tables, output is the schema and sample rows for those tables. Be sure that the tables actually exist by calling sql_db_list_tables first! Example Input: table1, table2, table3\n", - "\n", - "sql_db_list_tables: Input is an empty string, output is a comma-separated list of tables in the database.\n", - "\n", - "sql_db_query_checker: Use this tool to double check if your query is correct before executing it. Always use this tool before executing a query with sql_db_query!\n", - "\n" - ] - } - ], - "source": [ - "from langchain_community.agent_toolkits import SQLDatabaseToolkit\n", - "\n", - "toolkit = SQLDatabaseToolkit(db=db, llm=llm)\n", - "\n", - "tools = toolkit.get_tools()\n", - "\n", - "for tool in tools:\n", - " print(f\"{tool.name}: {tool.description}\\n\")" - ] - }, - { - "cell_type": "markdown", - "id": "6baf93d3-afd5-484e-9bf7-3ff20401e06f", - "metadata": {}, - "source": [ - "## 2. Using a prebuilt agent\n", - "\n", - "Given these tools, we can initialize a pre-built agent in a single line. To customize our agents behavior, we write a descriptive system prompt." - ] - }, - { - "cell_type": "code", - "execution_count": 4, - "id": "f6b9f10f-c6cb-43bc-b671-79aadc71280b", - "metadata": {}, - "outputs": [], - "source": [ - "from langgraph.prebuilt import create_react_agent\n", - "\n", - "system_prompt = \"\"\"\n", - "You are an agent designed to interact with a SQL database.\n", - "Given an input question, create a syntactically correct {dialect} query to run,\n", - "then look at the results of the query and return the answer. Unless the user\n", - "specifies a specific number of examples they wish to obtain, always limit your\n", - "query to at most {top_k} results.\n", - "\n", - "You can order the results by a relevant column to return the most interesting\n", - "examples in the database. Never query for all the columns from a specific table,\n", - "only ask for the relevant columns given the question.\n", - "\n", - "You MUST double check your query before executing it. If you get an error while\n", - "executing a query, rewrite the query and try again.\n", - "\n", - "DO NOT make any DML statements (INSERT, UPDATE, DELETE, DROP etc.) to the\n", - "database.\n", - "\n", - "To start you should ALWAYS look at the tables in the database to see what you\n", - "can query. Do NOT skip this step.\n", - "\n", - "Then you should query the schema of the most relevant tables.\n", - "\"\"\".format(\n", - " dialect=db.dialect,\n", - " top_k=5,\n", - ")\n", - "\n", - "agent = create_react_agent(\n", - " llm,\n", - " tools,\n", - " prompt=system_prompt,\n", - ")" - ] - }, - { - "cell_type": "markdown", - "id": "78b1c7cf-9a0f-48c3-baec-be8101bc3c77", - "metadata": {}, - "source": [ - "!!! note\n", - "\n", - " This system prompt includes a number of instructions, such as always running specific tools before or after others. In the [next section](#customizing-the-agent), we will enforce these behaviors through the graph's structure, providing us a greater degree of control and allowing us to simplify the prompt.\n", - "\n", - "\n", - "Let's run this agent on a sample query and observe its behavior:" - ] - }, - { - "cell_type": "code", - "execution_count": 5, - "id": "9317c234-6251-4723-a335-7fee6d11ac2b", - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "================================\u001b[1m Human Message \u001b[0m=================================\n", - "\n", - "Which genre on average has the longest tracks?\n", - "==================================\u001b[1m Ai Message \u001b[0m==================================\n", - "Tool Calls:\n", - " sql_db_list_tables (call_d8lCgywSroCgpVl558nmXKwA)\n", - " Call ID: call_d8lCgywSroCgpVl558nmXKwA\n", - " Args:\n", - "=================================\u001b[1m Tool Message \u001b[0m=================================\n", - "Name: sql_db_list_tables\n", - "\n", - "Album, Artist, Customer, Employee, Genre, Invoice, InvoiceLine, MediaType, Playlist, PlaylistTrack, Track\n", - "==================================\u001b[1m Ai Message \u001b[0m==================================\n", - "Tool Calls:\n", - " sql_db_schema (call_nNf6IIUcwMYLIkE0l6uWkZHe)\n", - " Call ID: call_nNf6IIUcwMYLIkE0l6uWkZHe\n", - " Args:\n", - " table_names: Genre, Track\n", - "=================================\u001b[1m Tool Message \u001b[0m=================================\n", - "Name: sql_db_schema\n", - "\n", - "\n", - "CREATE TABLE \"Genre\" (\n", - "\t\"GenreId\" INTEGER NOT NULL, \n", - "\t\"Name\" NVARCHAR(120), \n", - "\tPRIMARY KEY (\"GenreId\")\n", - ")\n", - "\n", - "/*\n", - "3 rows from Genre table:\n", - "GenreId\tName\n", - "1\tRock\n", - "2\tJazz\n", - "3\tMetal\n", - "*/\n", - "\n", - "\n", - "CREATE TABLE \"Track\" (\n", - "\t\"TrackId\" INTEGER NOT NULL, \n", - "\t\"Name\" NVARCHAR(200) NOT NULL, \n", - "\t\"AlbumId\" INTEGER, \n", - "\t\"MediaTypeId\" INTEGER NOT NULL, \n", - "\t\"GenreId\" INTEGER, \n", - "\t\"Composer\" NVARCHAR(220), \n", - "\t\"Milliseconds\" INTEGER NOT NULL, \n", - "\t\"Bytes\" INTEGER, \n", - "\t\"UnitPrice\" NUMERIC(10, 2) NOT NULL, \n", - "\tPRIMARY KEY (\"TrackId\"), \n", - "\tFOREIGN KEY(\"MediaTypeId\") REFERENCES \"MediaType\" (\"MediaTypeId\"), \n", - "\tFOREIGN KEY(\"GenreId\") REFERENCES \"Genre\" (\"GenreId\"), \n", - "\tFOREIGN KEY(\"AlbumId\") REFERENCES \"Album\" (\"AlbumId\")\n", - ")\n", - "\n", - "/*\n", - "3 rows from Track table:\n", - "TrackId\tName\tAlbumId\tMediaTypeId\tGenreId\tComposer\tMilliseconds\tBytes\tUnitPrice\n", - "1\tFor Those About To Rock (We Salute You)\t1\t1\t1\tAngus Young, Malcolm Young, Brian Johnson\t343719\t11170334\t0.99\n", - "2\tBalls to the Wall\t2\t2\t1\tNone\t342562\t5510424\t0.99\n", - "3\tFast As a Shark\t3\t2\t1\tF. Baltes, S. Kaufman, U. Dirkscneider & W. Hoffman\t230619\t3990994\t0.99\n", - "*/\n", - "==================================\u001b[1m Ai Message \u001b[0m==================================\n", - "Tool Calls:\n", - " sql_db_query_checker (call_urTRmtiGtTxkwHtscec7Fd2K)\n", - " Call ID: call_urTRmtiGtTxkwHtscec7Fd2K\n", - " Args:\n", - " query: SELECT Genre.Name, AVG(Track.Milliseconds) AS AvgMilliseconds\n", - "FROM Track\n", - "JOIN Genre ON Track.GenreId = Genre.GenreId\n", - "GROUP BY Genre.Name\n", - "ORDER BY AvgMilliseconds DESC\n", - "LIMIT 1;\n", - "=================================\u001b[1m Tool Message \u001b[0m=================================\n", - "Name: sql_db_query_checker\n", - "\n", - "```sql\n", - "SELECT Genre.Name, AVG(Track.Milliseconds) AS AvgMilliseconds\n", - "FROM Track\n", - "JOIN Genre ON Track.GenreId = Genre.GenreId\n", - "GROUP BY Genre.Name\n", - "ORDER BY AvgMilliseconds DESC\n", - "LIMIT 1;\n", - "```\n", - "==================================\u001b[1m Ai Message \u001b[0m==================================\n", - "Tool Calls:\n", - " sql_db_query (call_RNMqyUEMv0rvy0UxSwrXY2AV)\n", - " Call ID: call_RNMqyUEMv0rvy0UxSwrXY2AV\n", - " Args:\n", - " query: SELECT Genre.Name, AVG(Track.Milliseconds) AS AvgMilliseconds\n", - "FROM Track\n", - "JOIN Genre ON Track.GenreId = Genre.GenreId\n", - "GROUP BY Genre.Name\n", - "ORDER BY AvgMilliseconds DESC\n", - "LIMIT 1;\n", - "=================================\u001b[1m Tool Message \u001b[0m=================================\n", - "Name: sql_db_query\n", - "\n", - "[('Sci Fi & Fantasy', 2911783.0384615385)]\n", - "==================================\u001b[1m Ai Message \u001b[0m==================================\n", - "\n", - "The genre with the longest average track length is \"Sci Fi & Fantasy,\" with an average duration of about 2,911,783 milliseconds (approximately 48.5 minutes) per track.\n" - ] - } - ], - "source": [ - "question = \"Which genre on average has the longest tracks?\"\n", - "\n", - "for step in agent.stream(\n", - " {\"messages\": [{\"role\": \"user\", \"content\": question}]},\n", - " stream_mode=\"values\",\n", - "):\n", - " step[\"messages\"][-1].pretty_print()" - ] - }, - { - "cell_type": "markdown", - "id": "6b2afc54-0a26-40f8-bca2-01cbb2fc3443", - "metadata": {}, - "source": [ - "This worked well enough: the agent correctly listed the tables, obtained the schemas, wrote a query, checked the query, and ran it to inform its final response.\n", - "\n", - "!!! tip\n", - "\n", - " You can inspect all aspects of the above run, including steps taken, tools invoked, what prompts were seen by the LLM, and more in the [LangSmith trace](https://smith.langchain.com/public/bd594960-73e3-474b-b6f2-db039d7c713a/r)." - ] - }, - { - "cell_type": "markdown", - "id": "2ed79712-ca6b-47dc-9bc9-7cb6fc50a8aa", - "metadata": {}, - "source": [ - "## 3. Customizing the agent\n", - "\n", - "The prebuilt agent lets us get started quickly, but at each step the agent has access to the full set of tools. Above, we relied on the system prompt to constrain its behavior— for example, we instructed the agent to always start with the \"list tables\" tool, and to always run a query-checker tool before executing the query.\n", - "\n", - "We can enforce a higher degree of control in LangGraph by customizing the agent. Below, we implement a simple ReAct-agent setup, with dedicated nodes for specific tool-calls. We will use the same [state](../../concepts/low_level/#state) as the pre-built agent.\n", - "\n", - "We construct dedicated nodes for the following steps:\n", - "\n", - "- Listing DB tables\n", - "- Calling the \"get schema\" tool\n", - "- Generating a query\n", - "- Checking the query\n", - "\n", - "Putting these steps in dedicated nodes lets us (1) force tool-calls when needed, and (2) customize the prompts associated with each step." - ] - }, - { - "cell_type": "code", - "execution_count": 6, - "id": "992f6ce6-f932-4f44-994c-c0ba5032dbd1", - "metadata": {}, - "outputs": [], - "source": [ - "from typing import Literal\n", - "from langchain_core.messages import AIMessage\n", - "from langchain_core.runnables import RunnableConfig\n", - "from langgraph.graph import END, START, MessagesState, StateGraph\n", - "from langgraph.prebuilt import ToolNode\n", - "\n", - "\n", - "get_schema_tool = next(tool for tool in tools if tool.name == \"sql_db_schema\")\n", - "get_schema_node = ToolNode([get_schema_tool], name=\"get_schema\")\n", - "\n", - "run_query_tool = next(tool for tool in tools if tool.name == \"sql_db_query\")\n", - "run_query_node = ToolNode([run_query_tool], name=\"run_query\")\n", - "\n", - "\n", - "# Example: create a predetermined tool call\n", - "def list_tables(state: MessagesState):\n", - " tool_call = {\n", - " \"name\": \"sql_db_list_tables\",\n", - " \"args\": {},\n", - " \"id\": \"abc123\",\n", - " \"type\": \"tool_call\",\n", - " }\n", - " tool_call_message = AIMessage(content=\"\", tool_calls=[tool_call])\n", - "\n", - " list_tables_tool = next(tool for tool in tools if tool.name == \"sql_db_list_tables\")\n", - " tool_message = list_tables_tool.invoke(tool_call)\n", - " response = AIMessage(f\"Available tables: {tool_message.content}\")\n", - "\n", - " return {\"messages\": [tool_call_message, tool_message, response]}\n", - "\n", - "\n", - "# Example: force a model to create a tool call\n", - "def call_get_schema(state: MessagesState):\n", - " # Note that LangChain enforces that all models accept `tool_choice=\"any\"`\n", - " # as well as `tool_choice=`.\n", - " llm_with_tools = llm.bind_tools([get_schema_tool], tool_choice=\"any\")\n", - " response = llm_with_tools.invoke(state[\"messages\"])\n", - "\n", - " return {\"messages\": [response]}\n", - "\n", - "\n", - "generate_query_system_prompt = \"\"\"\n", - "You are an agent designed to interact with a SQL database.\n", - "Given an input question, create a syntactically correct {dialect} query to run,\n", - "then look at the results of the query and return the answer. Unless the user\n", - "specifies a specific number of examples they wish to obtain, always limit your\n", - "query to at most {top_k} results.\n", - "\n", - "You can order the results by a relevant column to return the most interesting\n", - "examples in the database. Never query for all the columns from a specific table,\n", - "only ask for the relevant columns given the question.\n", - "\n", - "DO NOT make any DML statements (INSERT, UPDATE, DELETE, DROP etc.) to the database.\n", - "\"\"\".format(\n", - " dialect=db.dialect,\n", - " top_k=5,\n", - ")\n", - "\n", - "\n", - "def generate_query(state: MessagesState):\n", - " system_message = {\n", - " \"role\": \"system\",\n", - " \"content\": generate_query_system_prompt,\n", - " }\n", - " # We do not force a tool call here, to allow the model to\n", - " # respond naturally when it obtains the solution.\n", - " llm_with_tools = llm.bind_tools([run_query_tool])\n", - " response = llm_with_tools.invoke([system_message] + state[\"messages\"])\n", - "\n", - " return {\"messages\": [response]}\n", - "\n", - "\n", - "check_query_system_prompt = \"\"\"\n", - "You are a SQL expert with a strong attention to detail.\n", - "Double check the {dialect} query for common mistakes, including:\n", - "- Using NOT IN with NULL values\n", - "- Using UNION when UNION ALL should have been used\n", - "- Using BETWEEN for exclusive ranges\n", - "- Data type mismatch in predicates\n", - "- Properly quoting identifiers\n", - "- Using the correct number of arguments for functions\n", - "- Casting to the correct data type\n", - "- Using the proper columns for joins\n", - "\n", - "If there are any of the above mistakes, rewrite the query. If there are no mistakes,\n", - "just reproduce the original query.\n", - "\n", - "You will call the appropriate tool to execute the query after running this check.\n", - "\"\"\".format(dialect=db.dialect)\n", - "\n", - "\n", - "def check_query(state: MessagesState):\n", - " system_message = {\n", - " \"role\": \"system\",\n", - " \"content\": check_query_system_prompt,\n", - " }\n", - "\n", - " # Generate an artificial user message to check\n", - " tool_call = state[\"messages\"][-1].tool_calls[0]\n", - " user_message = {\"role\": \"user\", \"content\": tool_call[\"args\"][\"query\"]}\n", - " llm_with_tools = llm.bind_tools([run_query_tool], tool_choice=\"any\")\n", - " response = llm_with_tools.invoke([system_message, user_message])\n", - " response.id = state[\"messages\"][-1].id\n", - "\n", - " return {\"messages\": [response]}" - ] - }, - { - "cell_type": "markdown", - "id": "cd23a432-09c1-428c-88a7-6b4d1fe55577", - "metadata": {}, - "source": [ - "Finally, we assemble these steps into a workflow using the Graph API. We define a [conditional edge](../../concepts/low_level/#conditional-edges) at the query generation step that will route to the query checker if a query is generated, or end if there are no tool calls present, such that the LLM has delivered a response to the query." - ] - }, - { - "cell_type": "code", - "execution_count": 7, - "id": "8b3ce09d-0eec-456d-8447-d04f79ccfc23", - "metadata": {}, - "outputs": [], - "source": [ - "def should_continue(state: MessagesState) -> Literal[END, \"check_query\"]:\n", - " messages = state[\"messages\"]\n", - " last_message = messages[-1]\n", - " if not last_message.tool_calls:\n", - " return END\n", - " else:\n", - " return \"check_query\"\n", - "\n", - "\n", - "builder = StateGraph(MessagesState)\n", - "builder.add_node(list_tables)\n", - "builder.add_node(call_get_schema)\n", - "builder.add_node(get_schema_node, \"get_schema\")\n", - "builder.add_node(generate_query)\n", - "builder.add_node(check_query)\n", - "builder.add_node(run_query_node, \"run_query\")\n", - "\n", - "builder.add_edge(START, \"list_tables\")\n", - "builder.add_edge(\"list_tables\", \"call_get_schema\")\n", - "builder.add_edge(\"call_get_schema\", \"get_schema\")\n", - "builder.add_edge(\"get_schema\", \"generate_query\")\n", - "builder.add_conditional_edges(\n", - " \"generate_query\",\n", - " should_continue,\n", - ")\n", - "builder.add_edge(\"check_query\", \"run_query\")\n", - "builder.add_edge(\"run_query\", \"generate_query\")\n", - "\n", - "agent = builder.compile()" - ] - }, - { - "cell_type": "markdown", - "id": "c377e9dd-92a7-45d4-91bf-408a6f2af8ca", - "metadata": {}, - "source": [ - "We visualize the application below:" - ] - }, - { - "cell_type": "code", - "execution_count": 8, - "id": "f728ed5f-7bcd-489e-9174-4e4725962335", - "metadata": {}, - "outputs": [ - { - "data": { - "image/png": "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", - "text/plain": [ - "" - ] - }, - "metadata": {}, - "output_type": "display_data" - } - ], - "source": [ - "from IPython.display import Image, display\n", - "from langchain_core.runnables.graph import CurveStyle, MermaidDrawMethod, NodeStyles\n", - "\n", - "display(Image(agent.get_graph().draw_mermaid_png()))" - ] - }, - { - "cell_type": "markdown", - "id": "977173cf-8f8d-416e-a122-aebff115f29a", - "metadata": {}, - "source": [ - "We can now invoke the graph exactly as before:" - ] - }, - { - "cell_type": "code", - "execution_count": 9, - "id": "6c800a77-2359-4011-ab0c-dad22ed333b1", - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "================================\u001b[1m Human Message \u001b[0m=================================\n", - "\n", - "Which genre on average has the longest tracks?\n", - "==================================\u001b[1m Ai Message \u001b[0m==================================\n", - "\n", - "Available tables: Album, Artist, Customer, Employee, Genre, Invoice, InvoiceLine, MediaType, Playlist, PlaylistTrack, Track\n", - "==================================\u001b[1m Ai Message \u001b[0m==================================\n", - "Tool Calls:\n", - " sql_db_schema (call_qxKtYiHgf93AiTDin9ez5wFp)\n", - " Call ID: call_qxKtYiHgf93AiTDin9ez5wFp\n", - " Args:\n", - " table_names: Genre,Track\n", - "=================================\u001b[1m Tool Message \u001b[0m=================================\n", - "Name: sql_db_schema\n", - "\n", - "\n", - "CREATE TABLE \"Genre\" (\n", - "\t\"GenreId\" INTEGER NOT NULL, \n", - "\t\"Name\" NVARCHAR(120), \n", - "\tPRIMARY KEY (\"GenreId\")\n", - ")\n", - "\n", - "/*\n", - "3 rows from Genre table:\n", - "GenreId\tName\n", - "1\tRock\n", - "2\tJazz\n", - "3\tMetal\n", - "*/\n", - "\n", - "\n", - "CREATE TABLE \"Track\" (\n", - "\t\"TrackId\" INTEGER NOT NULL, \n", - "\t\"Name\" NVARCHAR(200) NOT NULL, \n", - "\t\"AlbumId\" INTEGER, \n", - "\t\"MediaTypeId\" INTEGER NOT NULL, \n", - "\t\"GenreId\" INTEGER, \n", - "\t\"Composer\" NVARCHAR(220), \n", - "\t\"Milliseconds\" INTEGER NOT NULL, \n", - "\t\"Bytes\" INTEGER, \n", - "\t\"UnitPrice\" NUMERIC(10, 2) NOT NULL, \n", - "\tPRIMARY KEY (\"TrackId\"), \n", - "\tFOREIGN KEY(\"MediaTypeId\") REFERENCES \"MediaType\" (\"MediaTypeId\"), \n", - "\tFOREIGN KEY(\"GenreId\") REFERENCES \"Genre\" (\"GenreId\"), \n", - "\tFOREIGN KEY(\"AlbumId\") REFERENCES \"Album\" (\"AlbumId\")\n", - ")\n", - "\n", - "/*\n", - "3 rows from Track table:\n", - "TrackId\tName\tAlbumId\tMediaTypeId\tGenreId\tComposer\tMilliseconds\tBytes\tUnitPrice\n", - "1\tFor Those About To Rock (We Salute You)\t1\t1\t1\tAngus Young, Malcolm Young, Brian Johnson\t343719\t11170334\t0.99\n", - "2\tBalls to the Wall\t2\t2\t1\tNone\t342562\t5510424\t0.99\n", - "3\tFast As a Shark\t3\t2\t1\tF. Baltes, S. Kaufman, U. Dirkscneider & W. Hoffman\t230619\t3990994\t0.99\n", - "*/\n", - "==================================\u001b[1m Ai Message \u001b[0m==================================\n", - "Tool Calls:\n", - " sql_db_query (call_RPN3GABMfb6DTaFTLlwnZxVN)\n", - " Call ID: call_RPN3GABMfb6DTaFTLlwnZxVN\n", - " Args:\n", - " query: SELECT Genre.Name, AVG(Track.Milliseconds) AS AvgTrackLength\n", - "FROM Track\n", - "JOIN Genre ON Track.GenreId = Genre.GenreId\n", - "GROUP BY Genre.GenreId\n", - "ORDER BY AvgTrackLength DESC\n", - "LIMIT 1;\n", - "==================================\u001b[1m Ai Message \u001b[0m==================================\n", - "Tool Calls:\n", - " sql_db_query (call_PR4s8ymiF3ZQLaoZADXtdqcl)\n", - " Call ID: call_PR4s8ymiF3ZQLaoZADXtdqcl\n", - " Args:\n", - " query: SELECT Genre.Name, AVG(Track.Milliseconds) AS AvgTrackLength\n", - "FROM Track\n", - "JOIN Genre ON Track.GenreId = Genre.GenreId\n", - "GROUP BY Genre.GenreId\n", - "ORDER BY AvgTrackLength DESC\n", - "LIMIT 1;\n", - "=================================\u001b[1m Tool Message \u001b[0m=================================\n", - "Name: sql_db_query\n", - "\n", - "[('Sci Fi & Fantasy', 2911783.0384615385)]\n", - "==================================\u001b[1m Ai Message \u001b[0m==================================\n", - "\n", - "The genre with the longest tracks on average is \"Sci Fi & Fantasy,\" with an average track length of approximately 2,911,783 milliseconds.\n" - ] - } - ], - "source": [ - "question = \"Which genre on average has the longest tracks?\"\n", - "\n", - "for step in agent.stream(\n", - " {\"messages\": [{\"role\": \"user\", \"content\": question}]},\n", - " stream_mode=\"values\",\n", - "):\n", - " step[\"messages\"][-1].pretty_print()" - ] - }, - { - "cell_type": "markdown", - "id": "fc3d8130-15c3-4c22-bb17-eb82ad41586e", - "metadata": {}, - "source": [ - "!!! tip\n", - "\n", - " See [LangSmith trace](https://smith.langchain.com/public/94b8c9ac-12f7-4692-8706-836a1f30f1ea/r) for the above run." - ] - }, - { - "cell_type": "markdown", - "id": "838b3acf-3186-4aa1-b211-cf136e95c39d", - "metadata": {}, - "source": [ - "## Next steps\n", - "\n", - "Check out [this guide](https://docs.smith.langchain.com/evaluation/how_to_guides/langgraph) for evaluating LangGraph applications, including SQL agents like this one, using LangSmith." - ] - } - ], - "metadata": { - "kernelspec": { - "display_name": "Python 3 (ipykernel)", - "language": "python", - "name": "python3" - }, - "language_info": { - "codemirror_mode": { - "name": "ipython", - "version": 3 - }, - "file_extension": ".py", - "mimetype": "text/x-python", - "name": "python", - "nbconvert_exporter": "python", - "pygments_lexer": "ipython3", - "version": "3.10.4" - } - }, - "nbformat": 4, - "nbformat_minor": 5 -} diff --git a/docs/mkdocs.yml b/docs/mkdocs.yml index 020c6e28a..a66966905 100644 --- a/docs/mkdocs.yml +++ b/docs/mkdocs.yml @@ -109,6 +109,7 @@ nav: - Agent architectures: concepts/agentic_concepts.md - Guides: + - guides/index.md - LangGraph APIs: - Graph API: - Overview: concepts/low_level.md @@ -268,6 +269,7 @@ nav: - Environment variables: cloud/reference/env_var.md - Examples: + - examples/index.md - Template applications: concepts/template_applications.md # TODO: make tutorial - Agentic RAG: tutorials/rag/langgraph_agentic_rag.md - Agent Supervisor: tutorials/multi_agent/agent_supervisor.md @@ -286,6 +288,7 @@ nav: - Implement generative UI with LangGraph: cloud/how-tos/generative_ui_react.md - Additional resources: + - additional-resources/index.md - agents/prebuilt.md # NOTE: prebuilt.md is auto-generated by `make build-prebuilt` - LangGraph Academy course: https://academy.langchain.com/courses/intro-to-langgraph - Case studies: adopters.md