AI Functions - ClickHouse Documentation

AI Functions are built-in functions in ClickHouse that you can use to call AI or generate embeddings to work with your data, extract information, classify data, etc…

AI functions are experimental. Set allow_experimental_ai_functions to enable them.

AI functions can return unpredictable outputs. The result will highly depend on the quality of the prompt and the model used.

All functions are sharing a common infrastructure that provides:

Quota enforcement: Per-query limits on tokens (ai_function_max_input_tokens_per_query, ai_function_max_output_tokens_per_query) and API calls (ai_function_max_api_calls_per_query).
Retry with backoff: Transient failures are retried (ai_function_max_retries) with exponential backoff (ai_function_retry_initial_delay_ms).

Configuration

AI functions resolve provider credentials and configuration from a named collection. To set a named collection to use for credentials, use the ai_function_credentials setting. Example statement to create a named collection with provider credentials:

CREATE NAMED COLLECTION my_ai_credentials AS
    provider = 'openai',
    endpoint = 'https://api.openai.com/v1/chat/completions',
    model = 'gpt-4o-mini',
    api_key = 'sk-...';

Select the collection with the ai_function_credentials setting, for the session or for a single query:

-- For the session:
SET allow_experimental_ai_functions = 1;
SET ai_function_credentials = 'my_ai_credentials';
SELECT aiClassify('I love this product!', ['positive', 'negative', 'neutral']);

-- Or for a single query:
SELECT aiClassify('I love this product!', ['positive', 'negative', 'neutral'])
SETTINGS allow_experimental_ai_functions = 1, ai_function_credentials = 'my_ai_credentials';

When ai_function_credentials is empty (the default), an exception is raised.

Named collection parameters

Parameter	Type	Default	Description
`provider`	String	—	Model provider. Supported: `'openai'`, `'anthropic'`. See note below.
`endpoint`	String	—	API endpoint URL.
`model`	String	—	Model name (e.g. `'gpt-4o-mini'`, `'text-embedding-3-small'`).
`api_key`	String	—	Authentication key for the provider. Optional: when omitted, the auth header is not sent, which allows targeting OpenAI-compatible servers that do not require authentication.
`max_tokens`	UInt64	`1024`	Maximum number of output tokens per API call.
`api_version`	String	—	API version string. Used by Anthropic (`'2023-06-01'`).

Any OpenAI-compatible API (e.g. vLLM, Ollama, LiteLLM) can be used by setting provider = 'openai' and pointing the endpoint to your service.

Query-level settings

Which named collection to use is controlled by the ai_function_credentials setting. Other AI-related settings are listed in Settings under the ai_function_ prefix.

Use in `DEFAULT` and `MATERIALIZED` columns

The ai_function_credentials setting is read when the default expression is evaluated, NOT when the column is defined. The collection name is not stored in the column definition:

CREATE TABLE t (id UInt32, doc String, vector Array(Float32) DEFAULT aiEmbed(doc)) ...;
-- The stored default is `aiEmbed(doc)`; no collection is captured.

Evaluating the expression requires three things: allow_experimental_ai_functions and ai_function_credentials must be set, and the evaluating user must hold GRANT NAMED COLLECTION on the collection (resolving the credentials runs a NAMED COLLECTION access check). Any of them missing raises an exception (SUPPORT_IS_DISABLED, an empty-credentials error, or ACCESS_DENIED). A DEFAULT column is evaluated at INSERT, so both settings must be set in the inserting session or query:

GRANT NAMED COLLECTION ON my_ai_credentials TO user;
SET allow_experimental_ai_functions = 1;
SET ai_function_credentials = 'my_ai_credentials';
INSERT INTO t (id, doc) VALUES (1, 'hello');

To make such tables insertable without setting these per session, set both in a settings profile:

<profiles>
    <default>
        <allow_experimental_ai_functions>1</allow_experimental_ai_functions>
        <ai_function_credentials>my_ai_credentials</ai_function_credentials>
    </default>
</profiles>

A MATERIALIZED column is computed at INSERT like a DEFAULT column, and is also recomputed by mutations such as ALTER TABLE ... MATERIALIZE COLUMN. Mutations run outside a user session and do not inherit a query’s SETTINGS clause, but they do inherit settings from a settings profile. Set both settings in a settings profile, and grant NAMED COLLECTION to the table owner, for mutation-driven recomputation to succeed.

Restricting endpoint hosts

The endpoint URL in an AI named collection is an outbound destination the server connects to under its own identity, potentially carrying (if specified) the named collection’s api_key in the request headers. By default, ClickHouse permits any host. To restrict functions to a specific set of providers, configure remote_url_allow_hosts in the server config, e.g.:

<remote_url_allow_hosts>
    <host>api.openai.com</host>
    <host>api.anthropic.com</host>
</remote_url_allow_hosts>

Note that this setting is server-wide and applies to all HTTP-using features.

Supported providers

Provider	`provider` value	Chat functions	Notes
OpenAI	`'openai'`	Yes	Default provider.
Anthropic	`'anthropic'`	Yes	Uses `/v1/messages` endpoint.

Observability

AI function activity is tracked through ClickHouse ProfileEvents:

ProfileEvent	Description
`AIAPICalls`	Number of HTTP requests made to the AI provider.
`AIInputTokens`	Total input tokens consumed.
`AIOutputTokens`	Total output tokens consumed.
`AIRowsProcessed`	Number of rows that received a result.
`AIRowsSkipped`	Number of rows skipped (quota exceeded, or error with `ai_function_throw_on_error = 0`).

Query these events:

SELECT
    ProfileEvents['AIAPICalls'] AS api_calls,
    ProfileEvents['AIInputTokens'] AS input_tokens,
    ProfileEvents['AIOutputTokens'] AS output_tokens
FROM system.query_log
WHERE query_id = 'query_id'
AND type = 'QueryFinish'
ORDER BY event_time DESC;

aiClassify

Introduced in: v26.4.0 Classifies the given text into one of the provided categories using an LLM provider. The function sends the text together with a fixed classification prompt and a JSON-schema response format constraining the model to return exactly one of the supplied labels. When the response is returned as a JSON object of the form {"category": "..."}, the label is unwrapped and the label string is returned. Provider credentials and configuration are taken from the named collection specified by the ai_function_credentials setting. Syntax

aiClassify(text, categories[, temperature])

Aliases: AIClassify Arguments

text — Text to classify. String
categories — Constant list of candidate category labels. Array(String)
temperature — Sampling temperature controlling randomness. Default: 0.0. Float64

Returned value One of the provided category labels, or the default value for the column type (empty string) if the request failed and ai_function_throw_on_error is disabled. String Examples Classify sentiment

Query

SELECT aiClassify('I love this product!', ['positive', 'negative', 'neutral']) SETTINGS ai_function_credentials = 'my_ai_credentials'

Response

positive

Classify a column

Query

SELECT body, aiClassify(body, ['bug', 'question', 'feature']) AS kind FROM issues LIMIT 5

Response

aiEmbed

Introduced in: v26.6.0 Generates an embedding vector for the given text using the configured AI provider. The function sends the text to the configured embedding endpoint and returns the resulting vector as Array(Float32). Within a single block of rows, inputs are grouped into batches of up to ai_function_embedding_max_batch_size entries per HTTP request to reduce per-call overhead. Provider credentials and configuration are taken from the named collection specified by the ai_function_credentials setting. The optional dimensions argument, when supported by the model (e.g. OpenAI’s text-embedding-3-*), requests a vector of the given size; otherwise the model’s native size is returned. Syntax

aiEmbed(text[, dimensions])

Arguments

text — Text to embed. String
dimensions — Optional target dimensionality for the output vector. 0 or omitted means the model’s native size. UInt64

Returned value The embedding vector, or an empty array if the input is NULL or empty, the request failed and ai_function_throw_on_error is disabled, or a quota was exceeded with ai_function_throw_on_quota_exceeded disabled. Array(Float32) Examples Embed a single string

Query

SELECT aiEmbed('Hello world') SETTINGS ai_function_credentials = 'my_ai_credentials'

Response

With explicit dimensions

Query

SELECT aiEmbed('Hello world', 256) SETTINGS ai_function_credentials = 'my_ai_credentials'

Response

Embed a column of texts

Query

SELECT aiEmbed(title, 256) FROM articles LIMIT 10

Response

aiExtract

Introduced in: v26.4.0 Extracts structured information from unstructured text using an LLM provider. The second argument may be either a free-form natural-language instruction (e.g. 'the main complaint') or a JSON-encoded schema of the form '{"field_a": "description of field a", "field_b": "description of field b"}'. In instruction mode, the function returns the extracted value as a plain string, or an empty string if nothing was found. In schema mode, the function returns a JSON object string whose keys match the requested schema; missing fields are null. Provider credentials and configuration are taken from the named collection specified by the ai_function_credentials setting. Syntax

aiExtract(text, instruction_or_schema[, temperature])

Aliases: AIExtract Arguments

text — Text to extract information from. String
instruction_or_schema — Free-form extraction instruction, or a constant JSON object describing the fields to extract. const String
temperature — Sampling temperature controlling randomness. Default: 0.0. const Float64

Returned value A single extracted value (instruction mode) or a JSON object string (schema mode). Returns the default value for the column type (empty string) if the request failed and ai_function_throw_on_error is disabled. String Examples Free-form instruction

Query

SELECT aiExtract('The package arrived late and was damaged.', 'the main complaint') SETTINGS ai_function_credentials = 'my_ai_credentials'

Response

late and damaged package

Schema extraction

Query

SELECT aiExtract(review, '{"sentiment": "positive, negative or neutral", "topic": "main topic of the review"}') FROM reviews LIMIT 5

Response

aiGenerate

Introduced in: v26.4.0 Generates free-form text content from a prompt using an LLM provider. The function sends the prompt to the configured AI provider and returns the generated text. An optional system prompt can be provided to guide the model’s behavior (e.g. tone, format, role). If no system prompt is given, the default system prompt is: You are a helpful assistant. Provide a clear and concise response. Provider credentials and configuration are taken from the named collection specified by the ai_function_credentials setting. Syntax

aiGenerate(prompt[, system_prompt[, temperature]])

Aliases: AIGenerate Arguments

prompt — The user prompt or question to send to the model. String
system_prompt — Optional constant system-level instruction that guides the model’s behavior (e.g. persona, output format), sent along with each prompt. String
temperature — Sampling temperature controlling randomness. Default: 0.7. Float64

Returned value The generated text response, or the default value for the column type (empty string) if the request failed and ai_function_throw_on_error is disabled. String Examples Simple question

Query

SELECT aiGenerate('What is 2 + 2? Reply with just the number.') SETTINGS ai_function_credentials = 'my_ai_credentials'

Response

With system prompt

Query

SELECT aiGenerate('Explain ClickHouse', 'You are a database expert. Be concise.') SETTINGS ai_function_credentials = 'my_ai_credentials'

Response

Summarize column values

Query

SELECT article_title, aiGenerate(concat('Summarize in one sentence: ', article_body)) AS summary FROM articles LIMIT 5

Response

aiTranslate

Introduced in: v26.4.0 Translates the given text into the specified target language using an LLM provider. Additional style or dialect instructions may be passed as a third argument (e.g. 'keep technical terms untranslated'). Provider credentials and configuration are taken from the named collection specified by the ai_function_credentials setting. Syntax

aiTranslate(text, target_language[, instructions[, temperature]])

Aliases: AITranslate Arguments

text — Text to translate. String
target_language — Target language name or BCP-47 code (e.g. 'French', 'es-MX'). String
instructions — Optional constant additional instructions for the translator. String
temperature — Sampling temperature controlling randomness. Default: 0.3. Float64

Returned value The translated text, or the default value for the column type (empty string) if the request failed and ai_function_throw_on_error is disabled. String Examples Translate to French

Query

SELECT aiTranslate('Hello, world!', 'French') SETTINGS ai_function_credentials = 'my_ai_credentials'

Response

Bonjour le monde!

Translate to Japanese with style instructions

Query

SELECT aiTranslate(body, 'Japanese', 'Use polite form (desu/masu)') FROM articles LIMIT 5

Response

​Configuration

​Named collection parameters

​Query-level settings

​Use in DEFAULT and MATERIALIZED columns

​Restricting endpoint hosts

​Supported providers

​Observability

​aiClassify

​aiEmbed

​aiExtract

​aiGenerate

​aiTranslate

Configuration

Named collection parameters

Query-level settings

Use in `DEFAULT` and `MATERIALIZED` columns

Restricting endpoint hosts

Supported providers

Observability

aiClassify

aiEmbed

aiExtract

aiGenerate

aiTranslate