Skip to content

feat(rag): add VoyageAI voyage-context-4 contextualized embedding support #6368

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
fzowl wants to merge 1 commit into
base: main
Choose a base branch
from
+125 −1
Open

feat(rag): add VoyageAI voyage-context-4 contextualized embedding support #6368

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 21 additions & 1 deletion lib/crewai/src/crewai/rag/embeddings/providers/voyageai/embedding_callable.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,9 @@
from crewai.rag.embeddings.providers.voyageai.types import VoyageAIProviderConfig


CONTEXTUALIZED_CHUNK_SIZE = 32000


class VoyageAIEmbeddingFunction(EmbeddingFunction[Documents]):
"""Embedding function for VoyageAI models."""

Expand Down Expand Up @@ -50,9 +53,26 @@ def __call__(self, input: Documents) -> Embeddings:
if isinstance(input, str):
input = [input]

model = self._config.get("model", "voyage-2")

if model.startswith("voyage-context"):
result = self._client.contextualized_embed(
inputs=input,
model=model,
input_type="document",
output_dtype=self._config.get("output_dtype"),
output_dimension=self._config.get("output_dimension"),
enable_auto_chunking=True,
chunk_size=CONTEXTUALIZED_CHUNK_SIZE,
Comment on lines +58 to +66

@coderabbitai coderabbitai Bot Jun 27, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

🧩 Analysis chain

🌐 Web query:

For voyageai Python client version 0.3.5, what is the signature of Client.contextualized_embed, and which input_typevalues does it support? In particular, isquery supported or is the endpoint document-only?

💡 Result:

The contextualized_embed method in the Voyage AI Python client supports both query and document types [1][2]. It is not restricted to documents only [1][2]. The method signature for voyageai.Client.contextualized_embed is as follows [1][2]: def contextualized_embed( self, inputs: Union[List[List[str]], List[str]], model: str, input_type: Optional[str] = None, output_dimension: Optional[int] = None, output_dtype: Optional[str] = "float", enable_auto_chunking: Optional[bool] = False, chunk_size: Optional[int] = 512, chunk_overlap: Optional[int] = 0, chunk_fn: Optional[Callable[[str], List[str]]] = None,) The input_type parameter accepts the values None, "query", and "document" [1][2]. When using "query", each inner list should typically contain a single query string [3][4]. When using "document", each inner list typically contains chunks from a document [3][4]. The input_type argument is optional, but setting it to "query" or "document" is recommended for retrieval tasks as it prompts the model appropriately [1][2].

Citations:

🏁 Script executed:

sed -n '1,160p' lib/crewai/src/crewai/rag/embeddings/providers/voyageai/embedding_callable.py
printf '\n---\n'
rg -n "voyage-context|contextualized_embed|input_type" lib/crewai/src/crewai/rag -S

Repository: crewAIInc/crewAI

Length of output: 3654

Respect input_type for voyage-context models contextualized_embed() accepts both query and document, so hardcoding "document" here ignores the configured value. Either forward self._config.get("input_type") or reject non-document configs explicitly.

🤖 Prompt for AI Agents 
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In
`@lib/crewai/src/crewai/rag/embeddings/providers/voyageai/embedding_callable.py`
around lines 58 - 66, The voyage-context path in embedding_callable’s callable
currently hardcodes input_type to document, ignoring any configured value.
Update the contextualized_embed call to use the configured input_type from
self._config when appropriate, or add explicit validation in the embedding
callable to reject non-document configurations for voyage-context models. Make
the fix in the branch that handles model.startswith("voyage-context") so the
behavior matches the selected config.

)
return cast(
Embeddings,
[embedding for r in result.results for embedding in r.embeddings],
)

result = self._client.embed(
texts=input,
model=self._config.get("model", "voyage-2"),
model=model,
input_type=self._config.get("input_type"),
truncation=self._config.get("truncation", True),
output_dtype=self._config.get("output_dtype"),
Expand Down
104 changes: 104 additions & 0 deletions lib/crewai/tests/rag/embeddings/test_voyageai_embedding_callable.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,104 @@
"""Tests for the VoyageAI embedding function."""

from unittest.mock import MagicMock, patch

import numpy as np

from crewai.rag.embeddings.providers.voyageai.embedding_callable import (
CONTEXTUALIZED_CHUNK_SIZE,
VoyageAIEmbeddingFunction,
)


class TestVoyageAIEmbeddingFunction:
"""Test the VoyageAI embedding function call routing."""

def test_standard_model_uses_embed(self):
"""Standard models should call the regular embed endpoint."""
with patch("voyageai.Client") as mock_client_class:
mock_client = MagicMock()
mock_client_class.return_value = mock_client
mock_client.embed.return_value = MagicMock(embeddings=[[0.1, 0.2]])

fn = VoyageAIEmbeddingFunction(api_key="voyage-key", model="voyage-2")
result = fn(["aa", "bb"])

mock_client.embed.assert_called_once()
mock_client.contextualized_embed.assert_not_called()
assert np.allclose(result, [[0.1, 0.2]])

def test_contextualized_model_uses_contextualized_embed(self):
"""voyage-context-4 should call the contextualized embeddings endpoint."""
with patch("voyageai.Client") as mock_client_class:
mock_client = MagicMock()
mock_client_class.return_value = mock_client
mock_client.contextualized_embed.return_value = MagicMock(
results=[
MagicMock(embeddings=[[0.1, 0.2]]),
MagicMock(embeddings=[[0.3, 0.4]]),
]
)

fn = VoyageAIEmbeddingFunction(
api_key="voyage-key", model="voyage-context-4"
)
result = fn(["aa", "bb"])

mock_client.embed.assert_not_called()
mock_client.contextualized_embed.assert_called_once()
assert np.allclose(result, [[0.1, 0.2], [0.3, 0.4]])

def test_contextualized_call_sets_chunk_size_to_max(self):
"""chunk_size must be set to 32000 on every contextualized call."""
with patch("voyageai.Client") as mock_client_class:
mock_client = MagicMock()
mock_client_class.return_value = mock_client
mock_client.contextualized_embed.return_value = MagicMock(
results=[MagicMock(embeddings=[[0.1, 0.2]])]
)

fn = VoyageAIEmbeddingFunction(
api_key="voyage-key", model="voyage-context-4"
)
fn(["aa"])

_, kwargs = mock_client.contextualized_embed.call_args
assert kwargs["chunk_size"] == CONTEXTUALIZED_CHUNK_SIZE
assert CONTEXTUALIZED_CHUNK_SIZE == 32000

def test_contextualized_input_is_flat_list(self):
"""Input must be passed as a flat List[str], not wrapped in an extra list."""
with patch("voyageai.Client") as mock_client_class:
mock_client = MagicMock()
mock_client_class.return_value = mock_client
mock_client.contextualized_embed.return_value = MagicMock(
results=[
MagicMock(embeddings=[[0.1, 0.2]]),
MagicMock(embeddings=[[0.3, 0.4]]),
]
)

fn = VoyageAIEmbeddingFunction(
api_key="voyage-key", model="voyage-context-4"
)
fn(["aa", "bb"])

_, kwargs = mock_client.contextualized_embed.call_args
assert kwargs["inputs"] == ["aa", "bb"]

def test_contextualized_string_input_normalized_to_flat_list(self):
"""A single string input is normalized to a flat list of one string."""
with patch("voyageai.Client") as mock_client_class:
mock_client = MagicMock()
mock_client_class.return_value = mock_client
mock_client.contextualized_embed.return_value = MagicMock(
results=[MagicMock(embeddings=[[0.1, 0.2]])]
)

fn = VoyageAIEmbeddingFunction(
api_key="voyage-key", model="voyage-context-4"
)
fn("aa")

_, kwargs = mock_client.contextualized_embed.call_args
assert kwargs["inputs"] == ["aa"]
Loading