Snowflake Cortex AI

What it does

When your agent uses it

• Writing SQL that calls an LLM
• Extracting structured data from text
• Building a natural-language WHERE filter
• Setting up hybrid vector+keyword search for retrieval-augmented generation

Example prompts

• “/snowflake-cortex-ai”

Requirements

• Python 3

Workflow steps

7 steps, taken from the first numbered list in SKILL.md.

1. 1Pick the narrowest function for the task — Use AI_CLASSIFY for categorization, AI_FILTER for natural-language row filtering, AI_EXTRACT…
2. 2Check token cost before batch jobs — Run AI_COUNT_TOKENS(model, text) on a sample before running a function over a large table.
3. 3Prototype in SQL — Call the function on a small LIMIT-ed sample and inspect results before running it over a full table.
4. 4Structure prompts explicitly — Use PROMPT('template {0}', arg) for parameterized prompts and cast JSON output to VARIANT when you need…
5. 5For RAG, stand up Cortex Search — Create a CORTEX SEARCH SERVICE over the source table, query it via the Python or REST API to retrieve…
6. 6Use dedicated compute — Size Cortex Search's backing warehouse no larger than MEDIUM, and separate it from other pipeline warehouses.
7. 7Guard against failures — Use TRY_COMPLETE in place of AI_COMPLETE for batch jobs where a single failure shouldn't fail the whole run; it…

What it can do on your machine

Read from SKILL.md and the folder at commit 7682ca7. It shows what the files ask for, not the result of running them.

• Tool permissions

  Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

  From allowed-tools in the SKILL.md frontmatter.

• Runs code

  No scripts in the folder and no shell commands in SKILL.md (its code samples are sql and python).

  From the folder's file list and the shell code blocks in SKILL.md.

• Network

  No URLs in SKILL.md.

  From URLs in SKILL.md, links to its own repository left out.

• Credentials

  Names no API keys, tokens, secrets or passwords.

  From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Snowflake Cortex AI loads about 2.4k tokens when it runs. Until then it costs about 102 tokens; SKILL.md has 734 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it

~102

When it runs · the whole SKILL.md, loaded when a task matches

~2.4k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from Mindrally/skills at commit 7682ca7, republished under its Apache-2.0 licence (© Mindrally). 734 words, ~2,367 tokens.

name

snowflake-cortex-ai

description

Reference for Snowflake Cortex AI Functions (AI_COMPLETE, AI_CLASSIFY, AI_EXTRACT, AI_FILTER, etc.) and Cortex Search for building RAG applications entirely inside Snowflake. Use when writing SQL that calls an LLM, classifying or extracting structured data from text, building a natural-language WHERE filter, or setting up hybrid vector+keyword search for retrieval-augmented generation.

metadata.maintainer

Mindrally

metadata.source

https://github.com/Mindrally/skills

-- Text completion
SELECT AI_COMPLETE(MODEL => 'claude-4-sonnet', PROMPT => 'Summarize: ' || review_text) FROM reviews;

-- Document processing
SELECT AI_COMPLETE(
  MODEL => 'claude-4-sonnet',
  PROMPT => PROMPT('Extract the invoice total from {0}', TO_FILE('@docs', 'invoice.pdf'))
);

-- Structured JSON output
SELECT AI_COMPLETE(MODEL => 'claude-4-sonnet',
  PROMPT => 'Extract name, email, company as JSON: ' || raw_text)::VARIANT AS extracted
FROM contacts;

-- AI_CLASSIFY: cheaper than AI_COMPLETE for categorization tasks
SELECT AI_CLASSIFY(ticket_text, ['billing', 'technical', 'account', 'other']) AS category FROM tickets;
-- Multi-label: AI_CLASSIFY(input, categories, {'output_mode': 'multi'})

-- AI_FILTER: natural-language predicate directly in WHERE
SELECT * FROM reviews WHERE AI_FILTER(review_text, 'mentions product quality issues');

-- AI_AGG: cross-row aggregation with no context-window limit
SELECT AI_AGG(feedback_text, 'What are the top 3 themes?') FROM customer_feedback;

-- AI_EXTRACT: pull named fields out of free text
SELECT AI_EXTRACT(email_body, 'meeting date', 'attendees', 'action items') FROM emails;

SELECT review_text, AI_SENTIMENT(review_text) AS sentiment FROM product_reviews;
SELECT AI_EMBED(description) AS embedding FROM products;
SELECT AI_PARSE_DOCUMENT(TO_FILE('@docs', 'contract.pdf'), MODE => 'LAYOUT');
SELECT AI_TRANSCRIBE(TO_FILE('@media', 'recording.mp3')) AS transcript;
SELECT AI_REDACT(customer_notes) AS redacted FROM support_cases;

CREATE OR REPLACE CORTEX SEARCH SERVICE my_search
  ON transcript_text
  ATTRIBUTES region, agent_id
  WAREHOUSE = my_wh
  TARGET_LAG = '1 day'
  EMBEDDING_MODEL = 'snowflake-arctic-embed-l-v2.0'
  AS (SELECT transcript_text, region, agent_id FROM support_transcripts);

CREATE OR REPLACE CORTEX SEARCH SERVICE my_multi_search
  TEXT INDEXES transcript_text, summary
  VECTOR INDEXES transcript_text (model='snowflake-arctic-embed-l-v2.0')
  ATTRIBUTES region
  WAREHOUSE = my_wh
  TARGET_LAG = '1 hour'
  AS (SELECT transcript_text, summary, region FROM support_transcripts);

from snowflake.core import Root

root = Root(session)
service = root.databases["db"].schemas["schema"].cortex_search_services["my_search"]
resp = service.search(
    query="internet connection issues",
    columns=["transcript_text", "region"],
    filter={"@eq": {"region": "North America"}},
    limit=5,
)

POST /api/v2/databases/<db>/schemas/<schema>/cortex-search-services/<service>:query
Body: {"query": "...", "columns": [...], "filter": {...}, "limit": N}

{"@eq": {"region": "NA"}}
{"@contains": {"tags": "urgent"}}
{"@gte": {"score": 0.8}}
{"@and": [f1, f2]}
{"@or": [f1, f2]}
{"@not": f}

resp = service.search(
    query="billing dispute",
    columns=["transcript_text"],
    scoring_config={"weights": {"texts": 0.3, "vectors": 0.5, "reranker": 0.2}},
    limit=10,
)

SELECT AI_COMPLETE(
  MODEL => 'claude-4-sonnet',
  PROMPT => 'Answer using only this context: ' || context || ' Question: ' || question
);

© Mindrally, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in snowflake-cortex-ai of Mindrally/skills.

Open the folder on GitHubat commit 7682ca7