This document may be outdated. Please refer to the latest configuration files:

​
Common Variables

​
CONSOLE_API_URL

The backend URL for the console API. This is used to construct the authorization callback. If left empty, it defaults to the same domain as the application. Example: https://api.console.dify.ai

​
CONSOLE_WEB_URL

The front-end URL of the console web interface. This is used to construct front-end addresses and for CORS configuration. If left empty, it defaults to the same domain as the application. Example: https://console.dify.ai

​
SERVICE_API_URL

The Service API URL, used to display Service API Base URL in the front-end. If left empty, it defaults to the same domain as the application. Example: https://api.dify.ai

​
APP_API_URL

The WebApp API backend URL, used to specify the backend URL for the front-end API. If left empty, it defaults to the same domain as the application. Example: https://app.dify.ai

​
APP_WEB_URL

The WebApp URL, used to display File preview or download Url to the front-end or as Multi-model inputs; If left empty, it defaults to the same domain as the application. Example: https://udify.app/

​
FILES_URL

The prefix for file preview or download URLs, used to display these URLs in the front-end and provide them as input for multi-modal models. To prevent forgery, image preview URLs are signed and expire after 5 minutes.

​
Server

​
MODE

Startup mode: This is only available when launched using docker. It is not applicable when running from source code.

  • api: Start API Server.
  • worker: Start asynchronous queue worker.

​
DEBUG

Debug mode: Disabled by default. It’s recommended to enable this setting during local development to prevent issues caused by monkey patching.

​
FLASK_DEBUG

Flask debug mode: When enabled, it outputs trace information in the API responses, facilitating easier debugging.

​
SECRET_KEY

A secret key used for securely signing session cookies and encrypting sensitive information in the database.

This variable must be set before the first launch.

Run openssl rand -base64 42 to generate a strong key for it.

​
DEPLOY_ENV

Deployment environment:

  • PRODUCTION (default): Production environment.
  • TESTING: Testing environment. There will be a distinct color label on the front-end page, indicating that this environment is a testing environment.

​
LOG_LEVEL

The log output level. Default is INFO. For production environments, it’s recommended to set this to ERROR.

​
MIGRATION_ENABLED

When set to true, database migrations are automatically executed on container startup. This is only available when launched using docker and does not apply when running from source code.

For source code launches, you need to manually run flask db upgrade in the api directory.

​
CHECK_UPDATE_URL

Controls the version checking policy. If set to false, the system will not call https://updates.dify.ai to check for updates.

Currently, the version check interface based on CloudFlare Worker is not directly accessible in China. Setting this variable to an empty value will disable this API call.

​
TEXT_GENERATION_TIMEOUT_MS

Default value: 60000 (milliseconds). Specifies the timeout for text generation and workflow processes. This setting prevents system-wide service disruptions caused by individual processes exceeding their allocated time.

​
OPENAI_API_BASE

Used to change the OpenAI base address, default is https://api.openai.com/v1.

When OpenAI cannot be accessed in China, replace it with a domestic mirror address, or when a local model provides OpenAI compatible API, it can be replaced.

Only effective when starting with docker image or docker-compose.

  • DIFY_BIND_ADDRESS: API service binding address, default: 0.0.0.0, i.e., all addresses can be accessed.

  • DIFY_PORT: API service binding port number, default to 5001.

  • SERVER_WORKER_AMOUNT: The number of API server workers, i.e., the number of gevent workers. Formula: number of cpu cores x 2 + 1

    Reference: https://docs.gunicorn.org/en/stable/design.html#how-many-workers

  • SERVER_WORKER_CLASS: Defaults to gevent. If using windows, it can be switched to sync or solo.

  • GUNICORN_TIMEOUT: Request handling timeout. Default is 200. Recommended value is 360 to support longer SSE (Server-Sent Events) connection times.

  • CELERY_WORKER_CLASS: Similar to SERVER_WORKER_CLASS. Default is gevent. If using windows, it can be switched to sync or solo.

  • CELERY_WORKER_AMOUNT: The number of Celery workers. The default is 1, and can be set as needed.

​
Database Configuration

The database uses PostgreSQL. Please use the public schema.

  • DB_USERNAME: username
  • DB_PASSWORD: password
  • DB_HOST: database host
  • DB_PORT: database port number, default is 5432
  • DB_DATABASE: database name
  • SQLALCHEMY_POOL_SIZE: The size of the database connection pool. The default is 30 connections, which can be appropriately increased.
  • SQLALCHEMY_POOL_RECYCLE: Database connection pool recycling time, the default is 3600 seconds.
  • SQLALCHEMY_ECHO: Whether to print SQL, default is false.

​
Redis Configuration

This Redis configuration is used for caching and for pub/sub during conversation.

  • REDIS_HOST: Redis host
  • REDIS_PORT: Redis port, default is 6379
  • REDIS_DB: Redis Database, default is 0. Please use a different Database from Session Redis and Celery Broker.
  • REDIS_USERNAME: Redis username, default is empty
  • REDIS_PASSWORD: Redis password, default is empty. It is strongly recommended to set a password.
  • REDIS_USE_SSL: Whether to use SSL protocol for connection, default is false
  • REDIS_USE_SENTINEL: Use Redis Sentinel to connect to Redis servers
  • REDIS_SENTINELS: Sentinel nodes, format: <sentinel1_ip>:<sentinel1_port>,<sentinel2_ip>:<sentinel2_port>,<sentinel3_ip>:<sentinel3_port>
  • REDIS_SENTINEL_SERVICE_NAME: Sentinel service name, same as Master Name
  • REDIS_SENTINEL_USERNAME: Username for Sentinel
  • REDIS_SENTINEL_PASSWORD: Password for Sentinel
  • REDIS_SENTINEL_SOCKET_TIMEOUT: Sentinel timeout, default value: 0.1, unit: seconds

​
Celery Configuration

  • CELERY_BROKER_URL

    Format as follows(direct connection mode):

    redis://<redis_username>:<redis_password>@<redis_host>:<redis_port>/<redis_database>

    Example: redis://:difyai123456@redis:6379/1

    Sentinel mode:

    sentinel://<sentinel_username>:<sentinel_password>@<sentinel_host>:<sentinel_port>/<redis_database>

    Example: sentinel://localhost:26379/1;sentinel://localhost:26380/1;sentinel://localhost:26381/1

  • BROKER_USE_SSL: If set to true, use SSL protocol for connection, default is false

  • CELERY_USE_SENTINEL: If set to true, Sentinel mode will be enabled, default is false

  • CELERY_SENTINEL_MASTER_NAME: The service name of Sentinel, i.e., Master Name

  • CELERY_SENTINEL_SOCKET_TIMEOUT: Timeout for connecting to Sentinel, default value: 0.1, unit: seconds

​
CORS Configuration

Used to set the front-end cross-domain access policy.

  • CONSOLE_CORS_ALLOW_ORIGINS: Console CORS cross-domain policy, default is *, that is, all domains can access.
  • WEB_API_CORS_ALLOW_ORIGINS: WebAPP CORS cross-domain policy, default is *, that is, all domains can access.

​
File Storage Configuration

Used to store uploaded data set files, team/tenant encryption keys, and other files.

  • STORAGE_TYPE

    Type of storage facility

    • local (default): Local file storage, if this option is selected, the following STORAGE_LOCAL_PATH configuration needs to be set.

    • s3: S3 object storage, if this option is selected, the following S3_ prefixed configurations need to be set.

    • azure-blob: Azure Blob object storage, if this option is selected, the following AZURE_BLOB_ prefixed configurations need to be set.

    • aliyun-oss: Alibaba Cloud OSS object storage, if this option is selected, the following ALIYUN_OSS_ prefixed configurations need to be set.

    • huawei-obs: Huawei OBS object storage, if this option is selected, the following HUAWEI_OBS_ prefixed configurations need to be set.

    • volcengine-tos: Volcengine TOS object storage, if this option is selected, the following VOLCENGINE_TOS_ prefixed configurations need to be set.

  • STORAGE_LOCAL_PATH

    Default is storage, that is, it is stored in the storage directory of the current directory.

    If you are deploying with docker or docker-compose, be sure to mount the /app/api/storage directory in both containers to the same local directory, otherwise, you may encounter file not found errors.

  • S3_ENDPOINT: S3 endpoint address

  • S3_BUCKET_NAME: S3 bucket name

  • S3_ACCESS_KEY: S3 Access Key

  • S3_SECRET_KEY: S3 Secret Key

  • S3_REGION: S3 region information, such as: us-east-1

  • AZURE_BLOB_ACCOUNT_NAME: your-account-name eg, ‘difyai’

  • AZURE_BLOB_ACCOUNT_KEY: your-account-key eg, ‘difyai’

  • AZURE_BLOB_CONTAINER_NAME: your-container-name eg, ‘difyai-container’

  • AZURE_BLOB_ACCOUNT_URL: ‘https://your_account_name.blob.core.windows.net’

  • ALIYUN_OSS_BUCKET_NAME: your-bucket-name eg, ‘difyai’

  • ALIYUN_OSS_ACCESS_KEY: your-access-key eg, ‘difyai’

  • ALIYUN_OSS_SECRET_KEY: your-secret-key eg, ‘difyai’

  • ALIYUN_OSS_ENDPOINT: https://oss-ap-southeast-1-internal.aliyuncs.com # reference: https://www.alibabacloud.com/help/en/oss/user-guide/regions-and-endpoints

  • ALIYUN_OSS_REGION: ap-southeast-1 # reference: https://www.alibabacloud.com/help/en/oss/user-guide/regions-and-endpoints

  • ALIYUN_OSS_AUTH_VERSION: v4

  • ALIYUN_OSS_PATH: your-path # Don’t start with ’/’. OSS doesn’t support leading slash in object names. reference: https://www.alibabacloud.com/help/en/oss/support/0016-00000005

  • HUAWEI_OBS_BUCKET_NAME: your-bucket-name eg, ‘difyai’

  • HUAWEI_OBS_SECRET_KEY: your-secret-key eg, ‘difyai’

  • HUAWEI_OBS_ACCESS_KEY: your-access-key eg, ‘difyai’

  • HUAWEI_OBS_SERVER: your-server-url # reference: https://support.huaweicloud.com/sdk-python-devg-obs/obs_22_0500.html

  • VOLCENGINE_TOS_BUCKET_NAME: your-bucket-name eg, ‘difyai’

  • VOLCENGINE_TOS_SECRET_KEY: your-secret-key eg, ‘difyai’

  • VOLCENGINE_TOS_ACCESS_KEY: your-access-key eg, ‘difyai’

  • VOLCENGINE_TOS_REGION: your-region eg, ‘cn-guangzhou’ # reference: https://www.volcengine.com/docs/6349/107356

  • VOLCENGINE_TOS_ENDPOINT: your-endpoint eg, ‘tos-cn-guangzhou.volces.com’ # reference: https://www.volcengine.com/docs/6349/107356

​
Vector Database Configuration

  • VECTOR_STORE

    Available enumeration types include：

    • weaviate
    • qdrant
    • milvus
    • zilliz (share the same configuration as milvus)
    • myscale
    • pinecone (not yet open)
    • analyticdb
    • couchbase
    • oceanbase
    • tablestore
    • lindorm
    • tencent
    • opengauss

  • WEAVIATE_ENDPOINT

    Weaviate endpoint address, such as: http://weaviate:8080.

  • WEAVIATE_API_KEY

    The api-key credential used to connect to Weaviate.

  • WEAVIATE_BATCH_SIZE

    The number of index Objects created in batches in Weaviate, default is 100.

    Refer to this document: https://weaviate.io/developers/weaviate/manage-data/import#how-to-set-batch-parameters

  • WEAVIATE_GRPC_ENABLED

    Whether to use the gRPC method to interact with Weaviate, performance will greatly increase when enabled, may not be usable locally, default is true.

  • QDRANT_URL

    Qdrant endpoint address, such as: https://your-qdrant-cluster-url.qdrant.tech/

  • QDRANT_API_KEY

    The api-key credential used to connect to Qdrant.

  • PINECONE_API_KEY

    The api-key credential used to connect to Pinecone.

  • PINECONE_ENVIRONMENT

    The environment where Pinecone is located, such as: us-east4-gcp

  • MILVUS_URI

    Milvus uri configuration. e.g. http://host.docker.internal:19530. For Zilliz Cloud, adjust the uri and token to the Public Endpoint and API Key.

  • MILVUS_TOKEN

    Milvus token configuration, default is empty.

  • MILVUS_USER

    Milvus user configuration, default is empty.

  • MILVUS_PASSWORD

    Milvus password configuration, default is empty.

  • MYSCALE_HOST

    MyScale host configuration.

  • MYSCALE_PORT

    MyScale port configuration.

  • MYSCALE_USER

    MyScale user configuration, default is default.

  • MYSCALE_PASSWORD

    MyScale password configuration, default is empty.

  • MYSCALE_DATABASE

    MyScale database configuration, default is default.

  • MYSCALE_FTS_PARAMS

    MyScale text-search params, check MyScale docs for multi-language support, default is empty.

  • ANALYTICDB_KEY_ID

    The access key ID used for Aliyun OpenAPI authentication. Read the Analyticdb documentation to create your AccessKey.

  • ANALYTICDB_KEY_SECRET

    The access key secret used for Aliyun OpenAPI authentication.

  • ANALYTICDB_INSTANCE_ID

    The unique identifier for your AnalyticDB instance, such as : gp-xxxxxx. Read the Analyticdb documentation to create your instance.

  • ANALYTICDB_REGION_ID

    The region identifier where the AnalyticDB instance is located, such as: cn-hangzhou.

  • ANALYTICDB_ACCOUNT

    The account name used to connect to the AnalyticDB instance. Read the Analyticdb documentation to create an account.

  • ANALYTICDB_PASSWORD

    The password for the account used to connect to the AnalyticDB instance.

  • ANALYTICDB_NAMESPACE

    The namespace(schema) within the AnalyticDB instance that you wish to interact with, such as dify. If this namespace does not exist, it will be created automatically.

  • ANALYTICDB_NAMESPACE_PASSWORD

    The password for the namespace(schema). If the namespace does not exist, it will be created with this password.

  • COUCHBASE_CONNECTION_STRING

    The connection string for the Couchbase cluster.

  • COUCHBASE_USER

    The username for the database user.

  • COUCHBASE_PASSWORD

    The password for the database user.

  • COUCHBASE_BUCKET_NAME

    The name of the bucket to use.

  • COUCHBASE_SCOPE_NAME

    The name of the scope to use.

  • OCEANBASE_VECTOR_HOST

    The hostname or IP address of OceanBase vector database.

  • OCEANBASE_VECTOR_PORT

    The port of OceanBase vector database.

  • OCEANBASE_VECTOR_USER

    The username of OceanBase vector database.

  • OCEANBASE_VECTOR_PASSWORD

    The password of OceanBase vector database user.

  • OCEANBASE_VECTOR_DATABASE

    The database name of OceanBase vector database.

  • OCEANBASE_CLUSTER_NAME

    The cluster name of OceanBase vector database, only available for Docker deployment.

  • OCEANBASE_MEMORY_LIMIT

    The memory limit of OceanBase vector database, only available for Docker deployment.

  • TABLESTORE_ENDPOINT

    The endpoint address of the TableStore server (e.g. ‘https://instance-name.cn-hangzhou.ots.aliyuncs.com’)

  • TABLESTORE_INSTANCE_NAME

    The instance name to access TableStore server (e.g. ‘instance-name’)

  • TABLESTORE_ACCESS_KEY_ID

    The accessKey id for the instance name

  • TABLESTORE_ACCESS_KEY_SECRET

    The accessKey secret for the instance name

  • LINDORM_URL

    The URL of LINDORM search engine，you can get it from the console

  • LINDORM_USERNAME

    The username of lindorm search engine

  • LINDORM_PASSWORD

    The password of lindorm search engine

  • TENCENT_VECTOR_DB_URL

    The access address for Tencent Cloud VectorDB can be obtained from the console.

  • TENCENT_VECTOR_DB_API_KEY

    The API key (password) for the VectorDB server is used for identity authentication. Key Management.

  • TENCENT_VECTOR_DB_USERNAME

    The vector database account, default ‘root’. Account Management.

  • TENCENT_VECTOR_DB_TIMEOUT

    Set the default request timeout duration.

  • TENCENT_VECTOR_DB_DATABASE

    Set up a Database for storing data. Create Database.

  • TENCENT_VECTOR_DB_SHARD

    Specify the number of shards.

  • TENCENT_VECTOR_DB_REPLICAS

    Specify the number of replicas.

  • TENCENT_VECTOR_DB_ENABLE_HYBRID_SEARCH

    Specify whether to enable HybridSearch. Sparse Vector Documentation.

  • OPENGAUSS_HOST

    The hostname or IP address of the openGauss vector database.

  • OPENGAUSS_PORT

    The port of the openGauss vector database.

  • OPENGAUSS_USER

    The username of the openGauss vector database.

  • OPENGAUSS_PASSWORD

    The password of the openGauss vector database.

  • OPENGAUSS_DATABASE

    The database name of the openGauss vector database.

  • OPENGAUSS_MIN_CONNECTION

    Min connection of the openGauss vector database.

  • OPENGAUSS_MAX_CONNECTION

    Max connection of the openGauss vector database.

  • OPENGAUSS_ENABLE_PQ

    Enabling PQ Acceleration for the openGauss vector database.

​
Knowledge Configuration

  • UPLOAD_FILE_SIZE_LIMIT: Upload file size limit, default 15M.

  • UPLOAD_FILE_BATCH_LIMIT: The maximum number of files that can be uploaded at a time, default 5.

  • ETL_TYPE

    Available enumeration types include:

    • dify: Dify’s proprietary file extraction scheme
    • Unstructured: Unstructured.io file extraction scheme

  • UNSTRUCTURED_API_URL

    Unstructured API path, needs to be configured when ETL_TYPE is Unstructured.

    For example: http://unstructured:8000/general/v0/general

  • TOP_K_MAX_VALUE

    The maximum top-k value of RAG, default 10.

​
Multi-modal Configuration

  • MULTIMODAL_SEND_IMAGE_FORMAT

    The format of the image sent when the multi-modal model is input, the default is base64, optional url. The delay of the call in url mode will be lower than that in base64 mode. It is generally recommended to use the more compatible base64 mode. If configured as url, you need to configure FILES_URL as an externally accessible address so that the multi-modal model can access the image.

  • UPLOAD_IMAGE_FILE_SIZE_LIMIT: Upload image file size limit, default 10M.

​
Sentry Configuration

Used for application monitoring and error log tracking.

  • SENTRY_DSN: Sentry DSN address, default is empty, when empty, all monitoring information is not reported to Sentry.
  • SENTRY_TRACES_SAMPLE_RATE: The reporting ratio of Sentry events, if it is 0.01, it is 1%.
  • SENTRY_PROFILES_SAMPLE_RATE: The reporting ratio of Sentry profiles, if it is 0.01, it is 1%.

​
Notion Integration Configuration

Notion integration configuration variables can be obtained by applying for Notion integration: https://www.notion.so/my-integrations

  • NOTION_INTEGRATION_TYPE: Configure as “public” or “internal”. Since Notion’s OAuth redirect URL only supports HTTPS, if deploying locally, please use Notion’s internal integration.
  • NOTION_CLIENT_SECRET: Notion OAuth client secret (used for public integration type)
  • NOTION_CLIENT_ID: OAuth client ID (used for public integration type)
  • NOTION_INTERNAL_SECRET: Notion internal integration secret. If the value of NOTION_INTEGRATION_TYPE is “internal”, you need to configure this variable.

  • MAIL_TYPE

    • resend
      • MAIL_DEFAULT_SEND_FROM: The sender’s email name, such as: no-reply no-reply@dify.ai, not mandatory.
      • RESEND_API_KEY: API-Key for the Resend email provider, can be obtained from API-Key.
    • smtp
      • SMTP_SERVER: SMTP server address
      • SMTP_PORT: SMTP server port number
      • SMTP_USERNAME: SMTP username
      • SMTP_PASSWORD: SMTP password
      • SMTP_USE_TLS: Whether to use TLS, default is false
      • MAIL_DEFAULT_SEND_FROM: The sender’s email name, such as: no-reply no-reply@dify.ai, not mandatory.

​
ModelProvider & Tool Position Configuration

Used to specify the model providers and tools that can be used in the app. These settings allow you to customize which tools and model providers are available, as well as their order and inclusion/exclusion in the app’s interface.

For a list of available tools and model providers, please refer to the provided links.

  • POSITION_TOOL_PINS

    Pin specific tools to the top of the list, ensuring they appear first in the interface. (Use comma-separated values with no spaces between items.)

    Example: POSITION_TOOL_PINS=bing,google

  • POSITION_TOOL_INCLUDES

    Specify the tools to be included in the app. Only the tools listed here will be available for use. If not set, all tools will be included unless specified in POSITION_TOOL_EXCLUDES. (Use comma-separated values with no spaces between items.)

    Example: POSITION_TOOL_INCLUDES=bing,google

  • POSITION_TOOL_EXCLUDES

    Exclude specific tools from being displayed or used in the app. Tools listed here will be omitted from the available options, except for pinned tools. (Use comma-separated values with no spaces between items.)

    Example: POSITION_TOOL_EXCLUDES=yahoo,wolframalpha

  • POSITION_PROVIDER_PINS

    Pin specific model providers to the top of the list, ensuring they appear first in the interface. (Use comma-separated values with no spaces between items.)

    Example: POSITION_PROVIDER_PINS=openai,openllm

  • POSITION_PROVIDER_INCLUDES

    Specify the model providers to be included in the app. Only the providers listed here will be available for use. If not set, all providers will be included unless specified in POSITION_PROVIDER_EXCLUDES. (Use comma-separated values with no spaces between items.)

    Example: POSITION_PROVIDER_INCLUDES=cohere,upstage

  • POSITION_PROVIDER_EXCLUDES

    Exclude specific model providers from being displayed or used in the app. Providers listed here will be omitted from the available options, except for pinned providers. (Use comma-separated values with no spaces between items.)

    Example: POSITION_PROVIDER_EXCLUDES=openrouter,ollama

​
Others

  • INVITE_EXPIRY_HOURS: Member invitation link valid time (hours), Default: 72.
  • HTTP_REQUEST_NODE_MAX_TEXT_SIZE: The maximum text size of the HTTP request node in the workflow, default 1MB。
  • HTTP_REQUEST_NODE_MAX_BINARY_SIZE: The maximum binary size of HTTP request nodes in the workflow, default 10MB。

​
Web Frontend

​
SENTRY_DSN

Sentry DSN address, default is empty, when empty, all monitoring information is not reported to Sentry.

​
Deprecated

​
CONSOLE_URL

⚠️ Modified in 0.3.8, will be deprecated in 0.4.9, replaced by: CONSOLE_API_URL and CONSOLE_WEB_URL.

Console URL, used to concatenate the authorization callback, console front-end address, and CORS configuration use. If empty, it is the same domain. Example: https://console.dify.ai.

​
API_URL

⚠️ Modified in 0.3.8, will be deprecated in 0.4.9, replaced by SERVICE_API_URL.

API URL, used to display Service API Base URL to the front-end. If empty, it is the same domain. Example: https://api.dify.ai

​
APP_URL

⚠️ Modified in 0.3.8, will be deprecated in 0.4.9, replaced by APP_API_URL and APP_WEB_URL.

WebApp Url, used to display WebAPP API Base Url to the front-end. If empty, it is the same domain. Example: https://udify.app/

​
Session Configuration

⚠️ This configuration is no longer valid since v0.3.24.

Only used by the API service for interface identity verification.

  • SESSION_TYPE: Session component type

    • redis (default): If you choose this, you need to set the environment variables starting with SESSION_REDIS_ below.
    • sqlalchemy: If you choose this, the current database connection will be used and the sessions table will be used to read and write session records.

  • SESSION_REDIS_HOST: Redis host

  • SESSION_REDIS_PORT: Redis port, default is 6379

  • SESSION_REDIS_DB: Redis Database, default is 0. Please use a different Database from Redis and Celery Broker.

  • SESSION_REDIS_USERNAME: Redis username, default is empty

  • SESSION_REDIS_PASSWORD: Redis password, default is empty. It is strongly recommended to set a password.

  • SESSION_REDIS_USE_SSL: Whether to use SSL protocol for connection, default is false

⚠️ This configuration is no longer valid since v0.3.24.

Used to set the browser policy for session cookies used for identity verification.

  • COOKIE_HTTPONLY: Cookie HttpOnly configuration, default is true.
  • COOKIE_SAMESITE: Cookie SameSite configuration, default is Lax.
  • COOKIE_SECURE: Cookie Secure configuration, default is false.

​
Chunk Length Configuration

​
INDEXING_MAX_SEGMENTATION_TOKENS_LENGTH

Configuration for document chunk length. It is used to control the size of text segments when processing long documents. Default: 500. Maximum: 4000.

Larger Chunks

  • Retain more context within each chunk, ideal for tasks requiring a broader understanding of the text.
  • Reduce the total number of chunks, lowering processing time and storage overhead.

Smaller Chunks

  • Provide finer granularity, improving accuracy for tasks like extraction or summarization.
  • Reduce the risk of exceeding model token limits, making it safer for models with stricter constraints.

Configuration Recommendations

  • Choose larger chunks for context-heavy tasks like sentiment analysis or document summarization.
  • Choose smaller chunks for fine-grained tasks such as keyword extraction or paragraph-level processing.

Was this page helpful?