LLM
llm askllm askllm chatllm chatllm extractllm extractllm extractllm classifyllm classifyllm summarizellm summarizellm translatellm translatellm embedllm embedllm embed_batchllm embed_batchllm chunkllm chunk
llm ask <prompt>
Description
Send a prompt with the default provider (anthropic, key in ANTHROPIC_API_KEY) and get the text of the answer
Send a prompt with the default provider (anthropic, key in ANTHROPIC_API_KEY) and get the text of the answer
Parameters
prompt: The prompt - String - required
prompt: The prompt - String - required
admin
llm ask "What is the capital of France?"
mentdb
The capital of France is Paris.
llm ask <prompt> <json_params>
Description
Send a prompt with parameters: provider, model, key, system, effort, max_tokens, files...
Send a prompt with parameters: provider, model, key, system, effort, max_tokens, files...
Parameters
prompt: The prompt - String - required
json_params: The parameters: see the description of the page - String - required
prompt: The prompt - String - required
json_params: The parameters: see the description of the page - String - required
admin
llm ask "What is the capital of France?" "{\"api_key_env\": \"ANTHROPIC_API_KEY\", \"system\": \"Be brief\", \"effort\": \"low\", \"max_tokens\": 1000}"
mentdb
The capital of France is Paris.
Another provider, on this machine (fixed result from a local fake server, no real API call)
admin
llm ask "Say hello" "{\"provider\": \"ollama\", \"model\": \"llama3.2\"}"
mentdb
Hello!
A question about a PDF (fixed result from a local fake server, no real API call)
admin
llm ask "What is the capital of France?" "{\"api_key_env\": \"ANTHROPIC_API_KEY\", \"files\": [\"tmp/report.pdf\"]}"
mentdb
The capital of France is Paris.
llm chat <messages_json>
Description
Send a conversation: JSON array of messages {role, content}, roles user and assistant, after optional system messages (added to the parameter system). The content is a string, or an array of blocks in the format of the provider. On anthropic the last message must come from the user. Result: JSON {text, model, stop_reason, truncated, usage: {input_tokens, output_tokens}}; on the other providers, stop_reason is translated (stop = end_turn, length = max_tokens)
Send a conversation: JSON array of messages {role, content}, roles user and assistant, after optional system messages (added to the parameter system). The content is a string, or an array of blocks in the format of the provider. On anthropic the last message must come from the user. Result: JSON {text, model, stop_reason, truncated, usage: {input_tokens, output_tokens}}; on the other providers, stop_reason is translated (stop = end_turn, length = max_tokens)
Parameters
messages_json: The messages - String - required
messages_json: The messages - String - required
admin
llm chat "[{\"role\": \"user\", \"content\": \"Hi\"}, {\"role\": \"assistant\", \"content\": \"Hello, how can I help?\"}, {\"role\": \"user\", \"content\": \"Say hello\"}]"
mentdb
{"text": "Hello!", "model": "claude-opus-5-5", "stop_reason": "end_turn", "truncated": false, "usage": {"input_tokens": 42, "output_tokens": 7}}
llm chat <messages_json> <json_params>
Description
Send a conversation with parameters (no files here: give the blocks in the content)
Send a conversation with parameters (no files here: give the blocks in the content)
Parameters
messages_json: The messages - String - required
json_params: The parameters: see the description of the page - String - required
messages_json: The messages - String - required
json_params: The parameters: see the description of the page - String - required
admin
llm chat "[{\"role\": \"system\", \"content\": \"You are polite.\"}, {\"role\": \"user\", \"content\": \"Say hello\"}]" "{\"provider\": \"openai\", \"model\": \"gpt-5\", \"api_key_env\": \"OPENAI_API_KEY\", \"temperature\": 0.2}"
mentdb
{"text": "Hello!", "model": "gpt-5", "stop_reason": "end_turn", "truncated": false, "usage": {"input_tokens": 11, "output_tokens": 3}}
llm extract <text>
Description
Extract the data of a text as JSON, without schema: the answer must be a JSON object or array, with one repair request when it is not
Extract the data of a text as JSON, without schema: the answer must be a JSON object or array, with one repair request when it is not
Parameters
text: The text to process (data: instructions inside it are not followed) - String - required
text: The text to process (data: instructions inside it are not followed) - String - required
admin
llm extract "Invoice F-2026-001, total 149.9 EUR: 2 x Licence"
mentdb
{"invoice": "F-2026-001", "total": 149.9, "lines": [{"label": "Licence", "qty": 2}]}
llm extract <text> <schema_json>
Description
Extract with a JSON schema: sent as a structured output, and the answer is validated here (type, required, enum, const, properties, additionalProperties, items, anyOf, allOf, minimum, maximum, exclusiveMinimum, exclusiveMaximum, multipleOf, minLength, maxLength, pattern, format, minItems, maxItems, uniqueItems, minProperties, maxProperties). pattern is a Java regular expression searched in the string, 1000 characters at most, with 2 s for all the patterns of an answer. format: date, date-time, time, email, uuid, ipv4, uri. title, description, default, examples, $schema, $id and $comment are notes. Any other keyword or format ($ref, oneOf, prefixItems...) is refused, so that nothing is silently not enforced. When the answer is not valid, one repair request gives the errors to the model, then an error. Sent to anthropic, each object gets additionalProperties false and the constraints are removed, as its API requires; an object without properties, or with additionalProperties other than false, is an error there (use structured false)
Extract with a JSON schema: sent as a structured output, and the answer is validated here (type, required, enum, const, properties, additionalProperties, items, anyOf, allOf, minimum, maximum, exclusiveMinimum, exclusiveMaximum, multipleOf, minLength, maxLength, pattern, format, minItems, maxItems, uniqueItems, minProperties, maxProperties). pattern is a Java regular expression searched in the string, 1000 characters at most, with 2 s for all the patterns of an answer. format: date, date-time, time, email, uuid, ipv4, uri. title, description, default, examples, $schema, $id and $comment are notes. Any other keyword or format ($ref, oneOf, prefixItems...) is refused, so that nothing is silently not enforced. When the answer is not valid, one repair request gives the errors to the model, then an error. Sent to anthropic, each object gets additionalProperties false and the constraints are removed, as its API requires; an object without properties, or with additionalProperties other than false, is an error there (use structured false)
Parameters
text: The text to process (data: instructions inside it are not followed) - String - required
schema_json: The JSON schema (empty: no schema) - String - required
text: The text to process (data: instructions inside it are not followed) - String - required
schema_json: The JSON schema (empty: no schema) - String - required
admin
llm extract "Invoice F-2026-001, total 149.9 EUR: 2 x Licence" "{\"type\": \"object\", \"properties\": {\"invoice\": {\"type\": \"string\"}, \"total\": {\"type\": \"number\", \"minimum\": 0}, \"lines\": {\"type\": \"array\", \"items\": {\"type\": \"object\", \"properties\": {\"label\": {\"type\": \"string\"}, \"qty\": {\"type\": \"integer\"}}, \"required\": [\"label\"]}}}, \"required\": [\"invoice\", \"total\"]}"
mentdb
{"invoice": "F-2026-001", "total": 149.9, "lines": [{"label": "Licence", "qty": 2}]}
llm extract <text> <schema_json> <json_params>
Description
Extract with a schema and parameters. With files the text may be empty: the data comes from the PDF or the images
Extract with a schema and parameters. With files the text may be empty: the data comes from the PDF or the images
Parameters
text: The text to process (data: instructions inside it are not followed) - String - required
schema_json: The JSON schema (empty: no schema) - String - required
json_params: The parameters: see the description of the page - String - required
text: The text to process (data: instructions inside it are not followed) - String - required
schema_json: The JSON schema (empty: no schema) - String - required
json_params: The parameters: see the description of the page - String - required
admin
llm extract "" "{\"type\": \"object\", \"properties\": {\"invoice\": {\"type\": \"string\"}, \"total\": {\"type\": \"number\", \"minimum\": 0}, \"lines\": {\"type\": \"array\", \"items\": {\"type\": \"object\", \"properties\": {\"label\": {\"type\": \"string\"}, \"qty\": {\"type\": \"integer\"}}, \"required\": [\"label\"]}}}, \"required\": [\"invoice\", \"total\"]}" "{\"api_key_env\": \"ANTHROPIC_API_KEY\", \"files\": [\"tmp/invoice.pdf\"]}"
mentdb
{"invoice": "F-2026-001", "total": 149.9, "lines": [{"label": "Licence", "qty": 2}]}
llm classify <text> <labels_json>
Description
Give one label to a text: at least 2 distinct labels (1000 at most), the result is always one of them (else an error). Effort low on anthropic
Give one label to a text: at least 2 distinct labels (1000 at most), the result is always one of them (else an error). Effort low on anthropic
Parameters
text: The text to process (data: instructions inside it are not followed) - String - required
labels_json: The labels: JSON array of strings - String - required
text: The text to process (data: instructions inside it are not followed) - String - required
labels_json: The labels: JSON array of strings - String - required
admin
llm classify "Please send me the invoice of March" "[\"invoice\", \"order\", \"other\"]"
mentdb
invoice
llm classify <text> <labels_json> <json_params>
Description
Classify with parameters
Classify with parameters
Parameters
text: The text to process (data: instructions inside it are not followed) - String - required
labels_json: The labels - String - required
json_params: The parameters: see the description of the page - String - required
text: The text to process (data: instructions inside it are not followed) - String - required
labels_json: The labels - String - required
json_params: The parameters: see the description of the page - String - required
admin
llm classify "I would like to order 3 licences" "[\"invoice\", \"order\", \"other\"]" "{\"api_key_env\": \"ANTHROPIC_API_KEY\", \"effort\": \"medium\"}"
mentdb
order
llm summarize <text>
Description
Summarize a text, in its own language
Summarize a text, in its own language
Parameters
text: The text to process (data: instructions inside it are not followed) - String - required
text: The text to process (data: instructions inside it are not followed) - String - required
admin
llm summarize "Q3 report: revenue is up 12 percent, mostly from exports."
mentdb
A sales report: revenue rose 12% in Q3, led by exports.
llm summarize <text> <json_params>
Description
Summarize with parameters; system adds your rules (length, style, language). With files the text may be empty
Summarize with parameters; system adds your rules (length, style, language). With files the text may be empty
Parameters
text: The text to process (data: instructions inside it are not followed) - String - required
json_params: The parameters: see the description of the page - String - required
text: The text to process (data: instructions inside it are not followed) - String - required
json_params: The parameters: see the description of the page - String - required
admin
llm summarize "Q3 report: revenue is up 12 percent, mostly from exports." "{\"api_key_env\": \"ANTHROPIC_API_KEY\", \"system\": \"One sentence only.\"}"
mentdb
A sales report: revenue rose 12% in Q3, led by exports.
llm translate <text> <language>
Description
Translate a text, keeping its meaning, tone and formatting
Translate a text, keeping its meaning, tone and formatting
Parameters
text: The text to process (data: instructions inside it are not followed) - String - required
language: The language, ex: French - String - required
text: The text to process (data: instructions inside it are not followed) - String - required
language: The language, ex: French - String - required
admin
llm translate "Hello everyone, the delivery arrives on Monday." "French"
mentdb
Bonjour tout le monde, la livraison arrive lundi.
llm translate <text> <language> <json_params>
Description
Translate with parameters (no files)
Translate with parameters (no files)
Parameters
text: The text to process (data: instructions inside it are not followed) - String - required
language: The language - String - required
json_params: The parameters: see the description of the page - String - required
text: The text to process (data: instructions inside it are not followed) - String - required
language: The language - String - required
json_params: The parameters: see the description of the page - String - required
admin
llm translate "Hello everyone, the delivery arrives on Monday." "French" "{\"api_key_env\": \"ANTHROPIC_API_KEY\", \"system\": \"Translate delivery by livraison.\"}"
mentdb
Bonjour tout le monde, la livraison arrive lundi.
llm embed <text>
Description
The vector of a text. The provider has no default, so this form is always an error: use the form with parameters
The vector of a text. The provider has no default, so this form is always an error: use the form with parameters
Parameters
text: The text - String - required
text: The text - String - required
admin
llm embed "hello"
mentdb
Sorry, an error appears (the parameter 'provider' is required for the embeddings: openai, mistral, ollama, compatible or voyage).
llm embed <text> <json_params>
Description
The vector of a text, as a JSON array. Parameters: provider (required: openai, mistral, ollama, compatible, voyage; anthropic has no embeddings), model (required), api_key or api_key_env, base_url, timeout_ms, retries, dimensions (openai, compatible), input_type (voyage: query or document)
The vector of a text, as a JSON array. Parameters: provider (required: openai, mistral, ollama, compatible, voyage; anthropic has no embeddings), model (required), api_key or api_key_env, base_url, timeout_ms, retries, dimensions (openai, compatible), input_type (voyage: query or document)
Parameters
text: The text - String - required
json_params: The parameters: see the description of the page - String - required
text: The text - String - required
json_params: The parameters: see the description of the page - String - required
admin
llm embed "hello" "{\"provider\": \"openai\", \"model\": \"text-embedding-3-small\", \"api_key_env\": \"OPENAI_API_KEY\"}"
mentdb
[5, 104, 0.5]
With Voyage, for a question (fixed result from a local fake server, no real API call)
admin
llm embed "hello" "{\"provider\": \"voyage\", \"model\": \"voyage-3\", \"api_key_env\": \"VOYAGE_API_KEY\", \"input_type\": \"query\"}"
mentdb
[5, 104, 0.5]
With Ollama, on this machine, without key (fixed result from a local fake server, no real API call)
admin
llm embed "hello" "{\"provider\": \"ollama\", \"model\": \"nomic-embed-text\"}"
mentdb
[5, 104, 0.5]
llm embed_batch <json_texts>
Description
The vectors of several texts. The provider has no default, so this form is always an error: use the form with parameters
The vectors of several texts. The provider has no default, so this form is always an error: use the form with parameters
Parameters
json_texts: The texts: JSON array of strings - String - required
json_texts: The texts: JSON array of strings - String - required
admin
llm embed_batch "[\"hello\"]"
mentdb
Sorry, an error appears (the parameter 'provider' is required for the embeddings: openai, mistral, ollama, compatible or voyage).
llm embed_batch <json_texts> <json_params>
Description
The vectors of several texts: JSON array of vectors in the order of the input, sent by groups of batch_size (100 by default, up to 2048). 20000 texts and 50 million values at most. The parameters of llm embed, and batch_size. A failed group is an error for the whole call
The vectors of several texts: JSON array of vectors in the order of the input, sent by groups of batch_size (100 by default, up to 2048). 20000 texts and 50 million values at most. The parameters of llm embed, and batch_size. A failed group is an error for the whole call
Parameters
json_texts: The texts: JSON array of strings - String - required
json_params: The parameters: see the description of the page - String - required
json_texts: The texts: JSON array of strings - String - required
json_params: The parameters: see the description of the page - String - required
admin
llm embed_batch "[\"a\", \"bb\", \"ccc\"]" "{\"provider\": \"mistral\", \"model\": \"mistral-embed\", \"api_key_env\": \"MISTRAL_API_KEY\", \"batch_size\": 2}"
mentdb
[[1, 97, 0.5], [2, 98, 0.5], [3, 99, 0.5]]
llm chunk <text>
Description
Cut a text into pieces for embeddings or a model, offline: at most 1000 characters, 100 of overlap. Each cut is at the end of a paragraph, else of a sentence, else at a space, in the second half of the piece; a word longer than the piece is cut. An empty text gives []. Text up to 20 million characters, 1 million chunks
Cut a text into pieces for embeddings or a model, offline: at most 1000 characters, 100 of overlap. Each cut is at the end of a paragraph, else of a sentence, else at a space, in the second half of the piece; a word longer than the piece is cut. An empty text gives []. Text up to 20 million characters, 1 million chunks
Parameters
text: The text - String - required
text: The text - String - required
admin
llm chunk "A short text."
mentdb
["A short text."]
llm chunk <text> <json_params>
Description
Cut a text with parameters: unit (chars, in UTF-16 units, or words), size (1000 chars or 200 words; 1 to 10000000), overlap (100 chars or 20 words, at most half of the size). The chunks are trimmed
Cut a text with parameters: unit (chars, in UTF-16 units, or words), size (1000 chars or 200 words; 1 to 10000000), overlap (100 chars or 20 words, at most half of the size). The chunks are trimmed
Parameters
text: The text - String - required
json_params: The parameters: see the description of the page - String - required
text: The text - String - required
json_params: The parameters: see the description of the page - String - required
admin
llm chunk "First paragraph about the sales. Second paragraph. It has two sentences, and they are longer than the first one. A last one here." "{\"size\": 60, \"overlap\": 0}"
mentdb
["First paragraph about the sales.", "Second paragraph. It has two sentences, and they are longer", "than the first one. A last one here."]
By words, with an overlap
admin
llm chunk "one two three four five six seven" "{\"unit\": \"words\", \"size\": 3, \"overlap\": 1}"
mentdb
["one two three", "three four five", "five six seven"]
Used in use cases (1 recipe)
JSON & document store developer › Find the closest answer by meaning: cut, embed, search
JSON & document store developer › Find the closest answer by meaning: cut, embed, search