Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/dais-polymtl/flock/llms.txt

Use this file to discover all available pages before exploring further.

Flock supports Anthropic’s Claude family of models through a dedicated provider. Claude models are available for text completion, filtering, and aggregation. Note that Anthropic does not provide an embeddings API — if you need embeddings for RAG pipelines or similarity search, use the OpenAI or Ollama provider alongside Claude.

​
Prerequisites

You need an Anthropic API key. Create one at console.anthropic.com. Make sure Flock is installed and loaded before continuing — see the Quickstart if you haven’t done that yet.

​
Configure your secret

Store your API key using CREATE SECRET with the anthropic type. You can optionally specify an API version — Flock defaults to 2023-06-01: 
CREATE SECRET (
    TYPE anthropic,
    API_KEY 'your-api-key'
);
The API_KEY field is required. API_VERSION is optional — the default 2023-06-01 works with all current Claude models.
DuckDB’s secret manager automatically redacts api_key values in output. You can inspect your secrets with FROM duckdb_secrets() without exposing the key.

​
Create a model

Register a named model using a Claude model ID and the anthropic provider. Anthropic requires a max_tokens parameter — Flock defaults to 4096 if you omit it. 
CREATE MODEL(
    'ClaudeModel',
    'claude-sonnet-4-5',
    'anthropic',
    {"tuple_format": "json", "batch_size": 32, "model_parameters": {"temperature": 0.7, "max_tokens": 1024}}
);
The four arguments are:
ArgumentDescription
'ClaudeModel'Unique name you reference in queries
'claude-sonnet-4-5'Anthropic model ID
'anthropic'Provider name
{...}Config: batch size, tuple format, and model parameters

​
Run a query

With your secret and model in place, call llm_complete: 
SELECT llm_complete(
    {'model_name': 'ClaudeModel'},
    {'prompt': 'Explain what a database is in simple terms.'}
);
To use column data as context: 
SELECT llm_complete(
    {'model_name': 'ClaudeModel'},
    {
        'prompt': 'Summarize this support ticket in one sentence: {ticket}',
        'context_columns': [{'data': ticket_text, 'name': 'ticket'}]
    }
) AS summary
FROM support_tickets;

​
Available Claude models

Flock works with all current Claude models. The adapter automatically selects the correct API method based on the model version.
Model IDDescriptionStructured output method
claude-opus-4-5Most capableoutput_format
claude-sonnet-4-5Best balance of speed and capabilityoutput_format
claude-haiku-4-5Fast and cost-effectiveoutput_format
claude-3-5-sonnet-20241022Previous Sonnet generationtool_use
claude-3-haiku-20240307Previous Haiku generationtool_use
For the full and current list, see the Anthropic models documentation.

​
Structured output

Flock uses a hybrid approach for structured JSON output depending on the Claude model generation:
  • Claude 4.x models use the native output_format API, which guarantees strict schema compliance
  • Claude 3.x models fall back to tool_use, which provides structured output on models that predate the output_format API
You do not need to configure this — Flock detects the model version and applies the right method automatically. For Claude 4.x models, you can pass a custom JSON schema: 
SELECT llm_complete(
    {
        'model_name': 'ClaudeModel',
        'model_parameters': '{
            "output_format": {
                "type": "json_schema",
                "schema": {
                    "type": "object",
                    "properties": {
                        "sentiment": {"type": "string"},
                        "confidence": {"type": "number"}
                    },
                    "required": ["sentiment", "confidence"],
                    "additionalProperties": false
                }
            }
        }'
    },
    {'prompt': 'Analyze the sentiment of this text: I love this product!'}
) AS analysis;
Custom schemas must include "additionalProperties": false on all objects — this is required by the Anthropic output_format API.

​
Model parameters

You can customize Claude’s behavior with additional parameters in the model config: 
CREATE MODEL(
    'AnalystClaude',
    'claude-sonnet-4-5',
    'anthropic',
    {
        "model_parameters": {
            "temperature": 0.5,
            "max_tokens": 2048,
            "system": "You are an expert data analyst. Always provide structured, actionable insights."
        }
    }
);
Supported parameters include:
ParameterDescription
temperatureControls randomness, 0.0 to 1.0
max_tokensMaximum response length (required by Anthropic, default: 4096)
systemSystem prompt for context and instructions
top_pNucleus sampling threshold
top_kLimits token selection to top K options
Flock forwards all parameters to the Anthropic Messages API unchanged.
Anthropic does not provide an embeddings API. Calling llm_embedding with an Anthropic model returns an error. Use the OpenAI or Ollama provider for embeddings instead.

​
Next steps

Structured output

Full guide to enforcing JSON schemas on LLM responses

Scalar functions

Reference for llm_complete, llm_filter, and aggregate functions

Build docs developers (and LLMs) love

Get started for freeTalk to us