Preview
When an agent must answer from thousands of documents, you cannot send them all — embeddings let you find the few passages that match the question’s meaning.
Sign in to read the full chapter.
Checking access…
Embeddings and Similarity
Code panel
Embeddings and Similarity
Contract
- Put your API key in the practice file (fine for local learning)
- Create an authenticated client
- Embed several short sentences with an embedding model
- Print cosine similarity for related vs unrelated pairs
- Print how many numbers are in one vector
embeddings_demo.py
Highlighting…
Example console
Sample run — exact text can vary
Sentences:
[0] The dog runs in the park.
[1] A puppy plays outside.
[2] The weather is sunny today.
[3] It's a beautiful warm day.
[4] Python is a programming language.
Each embedding has 3072 numbers.
Similarity scores (higher = more similar meaning):
related (dog / puppy)
[0] The dog runs in the park.
[1] A puppy plays outside.
→ 0.7841
related (sunny / warm day)
[2] The weather is sunny today.
[3] It's a beautiful warm day.
→ 0.8012
unrelated (dog / Python)
[0] The dog runs in the park.
[4] Python is a programming language.
→ 0.3124
unrelated (puppy / Python)
[1] A puppy plays outside.
[4] Python is a programming language.
→ 0.2987
Check: both 'related' lines should beat both 'unrelated' lines.