Files
litellm/tests/logging_callback_tests/test_bedrock_knowledgebase_hook.py
T
Mateo Wang f9407bc036 chore(tests): migrate Bedrock CI to AWS account 941277531214 (#28728)
* chore(tests): migrate Bedrock CI from AWS account 888602223428 to 941277531214

The original account (888602223428) was put under a security restriction by
AWS after a root access key leaked in a PR comment. While that account works
its way through the AWS Support unlock process, Bedrock-touching CI tests have
been migrated to a fresh account (941277531214).

Changes:
  - Replace 26 hardcoded references to 888602223428 with 941277531214 across
    8 files (provisioned-model ARNs, imported-model ARNs, AgentCore runtime
    ARNs, batch execution role ARN, and example proxy config).
  - The provisioned-model and imported-model ARNs are referenced only from
    mocked unit tests — no AWS resources to recreate.
  - The batch execution IAM role has been recreated in the new account with
    the same name and equivalent permissions.
  - The two AgentCore runtimes (hosted_agent_r9jvp-3ySZuRHjLC,
    hosted_agent_13sf6-cALnp38iZD) are being recreated in the new account
    under the same names — see tools/agentcore-deploy/ in a follow-up.

CircleCI env vars AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY / AWS_REGION_NAME
were updated separately via the CircleCI API to point at the new account.

Smoke-tested locally against the new account:
  aws bedrock-runtime converse --region us-west-2 \
    --model-id us.anthropic.claude-sonnet-4-5-20250929-v1:0 \
    --messages '[{"role":"user","content":[{"text":"ping"}]}]'
  → 200, model returned 'pong'

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* chore(tests): refresh AgentCore ARN suffixes to match newly-deployed runtimes

The first migration commit replaced just the account ID, but AgentCore
auto-assigns a random 10-char suffix to every runtime on creation — we
can't reuse the original suffixes (`3ySZuRHjLC`, `cALnp38iZD`) in the
new account. Updated the AgentCore-runtime ARNs in the three files that
reference real runtime IDs (not the mock-based unit-test ARNs).

Deployed runtimes:
  arn:aws:bedrock-agentcore:us-west-2:941277531214:runtime/hosted_agent_r9jvp-Rq79QFC2fp
  arn:aws:bedrock-agentcore:us-west-2:941277531214:runtime/hosted_agent_13sf6-4046UzHSwy

Both runtimes are status=READY and pass a smoke invoke:
  $ aws bedrock-agentcore invoke-agent-runtime --agent-runtime-arn ... --payload '{"prompt":"ping"}'
  → 200, {"result": "echo: ping"}

The agent is a minimal echo (see /tmp/agentcore_deploy/agent.py for the
deploy artifacts). Tests that only verify the SDK wiring will pass; if any
test asserts on agent output content, swap the echo for the real agent.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* chore(tests): point Bedrock batch tests at new-account S3 bucket

The account migration (888602223428 -> 941277531214) was a flat
account-ID swap, which only rewrites ARNs that embed the account
number. S3 bucket names carry no account ID, so the live Bedrock
batch tests still uploaded to `litellm-proxy` — a bucket that lives
in the old account. S3 names are globally unique, and the old account
still holds that name, so it can't be recreated in the new account.

Rename to `litellm-proxy-941277531214` (account-ID suffix guarantees
global uniqueness). The bucket must be created in 941277531214 and the
batch execution role granted s3:GetObject/PutObject/ListBucket on it
before this job is run in CI.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* chore(tests): point live S3 logging test at new-account bucket

Same account-ID-free blind spot as the batch bucket: `load-testing-oct`
lives in the old account and its name can't be reused globally. The
`logging_testing` CI job is wired into the workflow and runs
test_basic_s3_logging, which uploads to this bucket with the CI env
creds, then lists and deletes objects — a live dependency.

Rename to `load-testing-oct-941277531214`. The bucket must exist in the
new account with the CI IAM principal granted
s3:PutObject/GetObject/ListBucket/DeleteObject before this job runs.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* chore(tests): repoint Bedrock guardrail IDs to new-account guardrails

The migration left guardrail IDs untouched (no account ID in them), so
all live guardrail tests failed with "guardrail identifier or version
does not exist" against 941277531214. Recreated both guardrails in the
new account and updated the hardcoded IDs:
  - wf0hkdb5x07f -> zgkmukebruil (PII mask: PHONE + CREDIT_DEBIT_CARD,
    with explicit inputAction=ANONYMIZE so masking applies to INPUT,
    which is the source litellm's moderation hook sends)
  - ff6ujrregl1q -> 4w3d1di3snt5 (blocks "coffee"; blocked message set
    to the exact string the tests assert on)

Updated test_bedrock_guardrails.py, otel_test_config.yaml, and the
guardrailConfig in test_bedrock_completion.py. Verified locally: the 5
previously-failing guardrail tests now pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* test(bedrock): migrate legacy models to current inference profiles

The new CI account (941277531214) cannot invoke legacy Bedrock models
(AWS gates them: "marked by provider as Legacy... not actively using in
the last 30 days"). Migrated the live-call tests:
  - anthropic.claude-3-sonnet-20240229    -> us.anthropic.claude-sonnet-4-5-20250929-v1:0
  - anthropic.claude-3-haiku-20240307     -> us.anthropic.claude-haiku-4-5-20251001-v1:0
Current Claude models on Bedrock require the us. inference-profile prefix
(bare on-demand ids are rejected).

cohere.command-r-plus has no working replacement (all Cohere is legacy-
gated in the new account): swapped to claude-haiku-4-5 in provider-
agnostic param lists. amazon.titan-image-generator skipped (no working
replacement). Mocked/transformation/cost tests that reference the legacy
strings are intentionally left unchanged. Verified live against the new
account.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* test(bedrock): repoint SageMaker + Knowledge Base to new-account resources

These referenced account-scoped resources by hardcoded id that only
existed in the old account, so the migration's account-ID swap missed
them. Recreated in 941277531214 and repointed:
  - SageMaker endpoint jumpstart-dft-hf-textgeneration1-mp-20240815-185614
    -> litellm-ci-textgen (gpt2 on a TGI container, ml.g5.xlarge)
  - Bedrock Knowledge Base T37J8R4WTM -> LCYXFBR2TU (OpenSearch Serverless
    vector store + titan-embed-text-v2, seeded with a LiteLLM doc)
Verified live: test_sagemaker.py (12 passed) and
test_bedrock_knowledgebase_hook.py (12 passed).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* test(reasoning_effort_grid): skip bedrock claude-opus-4-7 cells (not entitled on 941277531214)

claude-opus-4-7 is listed in the new Bedrock CI account's foundation
models but invoke is denied (AccessDeniedException: "not available for
this account"). Bedrock access to the flagship Opus requires an AWS
Sales request, not the self-serve model-access toggle, so it can't be
enabled inline with the rest of the account migration.

Add an optional `skip_reason` to ModelEntry and set it on the
bedrock-claude-opus-4-7 entry; the grid test honors it via pytest.skip.
Cell count (231) and route coverage are unchanged, so the structural
asserts still pass. Restore coverage by deleting the one skip_reason
line once access is granted.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* test(bedrock): swap/skip legacy-gated models unavailable on new CI account

The migrated AWS account (941277531214) cannot access several models that
the old account could, so the remaining red CI jobs were hitting real
Bedrock "Access denied / Legacy" and "account not authorized" errors:

- image_gen: skip both Nova Canvas test classes (amazon.nova-canvas-v1:0 is
  legacy-gated), matching the existing titan skip.
- batches: skip test_async_file_and_batch (Bedrock batch inference is not
  authorized on the new account; requires an AWS support case).
- litellm_overhead: swap legacy claude-3-5-haiku for the active
  us.anthropic.claude-haiku-4-5 inference profile.
- test_completion_claude_3_function_call: swap legacy claude-3-sonnet for the
  active us.anthropic.claude-sonnet-4-5 inference profile.

https://claude.ai/code/session_01Y7zgHYu9GX29YRwV4yiWAa

* test(bedrock): fix remaining e2e legacy-model + batch failures on new CI account

- e2e_openai_endpoints: skip test_bedrock_batches_api (Bedrock batch inference
  is not authorized on account 941277531214) and migrate the missed
  s3_bucket_name in oai_misc_config.yaml to litellm-proxy-941277531214.
- build_and_test: swap legacy bedrock claude-3-sonnet for the active
  us.anthropic.claude-sonnet-4-5 inference profile in the proxy structured
  output e2e test.

https://claude.ai/code/session_01Y7zgHYu9GX29YRwV4yiWAa

* test(bedrock): make opus-4-7 + batch cells fail loudly and mock image-gen (#28791)

Replace the silent skips added for the new CI account with noisier behavior:
- reasoning-effort grid: opus-4-7 cells now fail (when AWS creds are present)
  instead of skipping, so the missing entitlement stays visible in CI; they
  still skip when AWS creds are absent (local dev)
- Bedrock batch inference tests: drop the skip so they run and fail until
  batch access is granted
- Titan + Nova Canvas image-gen tests: mock the Bedrock HTTP call so the
  transform + cost-tracking path stays under test without live model access

https://claude.ai/code/session_01MT7SWDnXUjv6e6EPG7BDjT

Co-authored-by: Claude <noreply@anthropic.com>

* test(bedrock): use pytest.xfail for known-failing opus-4-7 cells

Replace pytest.fail with pytest.xfail when a model has a fail_reason,
so known-broken cells stay visible as XFAIL without keeping CI red.

Co-authored-by: Yassin Kortam <yassin@berri.ai>

---------

Co-authored-by: Mateo <mateo@Mateos-MacBook-Pro.local>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
2026-05-25 12:03:17 -07:00

935 lines
33 KiB
Python

import io
import os
import sys
sys.path.insert(0, os.path.abspath("../.."))
import asyncio
import litellm
import litellm.vector_stores.main
import gzip
import json
import logging
import time
from typing import Optional, List
from unittest.mock import AsyncMock, patch, Mock
import pytest
import litellm
from litellm import completion
from litellm._logging import verbose_logger
from litellm.integrations.vector_store_integrations.vector_store_pre_call_hook import (
VectorStorePreCallHook,
)
from litellm.llms.custom_httpx.http_handler import HTTPHandler, AsyncHTTPHandler
from litellm.integrations.custom_logger import CustomLogger
from litellm.types.utils import (
StandardLoggingPayload,
StandardLoggingVectorStoreRequest,
)
from litellm.types.vector_stores import (
VectorStoreSearchResponse,
VectorStoreResultContent,
VectorStoreSearchResult,
)
class MockCustomLogger(CustomLogger):
def __init__(self):
self.standard_logging_payload: Optional[StandardLoggingPayload] = None
self.completion_logging_payload: Optional[StandardLoggingPayload] = None
super().__init__()
async def async_log_success_event(self, kwargs, response_obj, start_time, end_time):
payload = kwargs.get("standard_logging_object")
# Store the payload - completion calls have call_type='acompletion'
if payload and payload.get("call_type") == "acompletion":
self.completion_logging_payload = payload
self.standard_logging_payload = payload
pass
@pytest.fixture(autouse=True)
def add_aws_region_to_env(monkeypatch):
monkeypatch.setenv("AWS_REGION", "us-west-2")
@pytest.fixture
def setup_vector_store_registry():
from litellm.vector_stores.vector_store_registry import (
VectorStoreRegistry,
LiteLLM_ManagedVectorStore,
)
# Init vector store registry
litellm.vector_store_registry = VectorStoreRegistry(
vector_stores=[
LiteLLM_ManagedVectorStore(
vector_store_id="LCYXFBR2TU", custom_llm_provider="bedrock"
)
]
)
@pytest.mark.asyncio
async def test_e2e_bedrock_knowledgebase_retrieval_with_completion(
setup_vector_store_registry,
):
litellm._turn_on_debug()
client = AsyncHTTPHandler()
print("value of litellm.vector_store_registry:", litellm.vector_store_registry)
with patch.object(client, "post") as mock_post:
# Mock the response for the LLM call
mock_response = Mock()
mock_response.status_code = 200
mock_response.headers = {"Content-Type": "application/json"}
# Provide proper JSON response content
mock_response.text = json.dumps(
{
"id": "msg_01ABC123",
"type": "message",
"role": "assistant",
"content": [
{
"type": "text",
"text": "LiteLLM is a library that simplifies LLM API access.",
}
],
"model": "claude-3.5-sonnet",
"stop_reason": "end_turn",
"stop_sequence": None,
"usage": {"input_tokens": 100, "output_tokens": 50},
}
)
mock_response.json = lambda: json.loads(mock_response.text)
mock_post.return_value = mock_response
try:
response = await litellm.acompletion(
model="anthropic/claude-3.5-sonnet",
messages=[{"role": "user", "content": "what is litellm?"}],
vector_store_ids=["LCYXFBR2TU"],
client=client,
)
except Exception as e:
print(f"Error: {e}")
# Verify the LLM request was made
mock_post.assert_called_once()
# Verify the request body
print("call args:", mock_post.call_args)
request_body = mock_post.call_args.kwargs["json"]
print("Request body:", json.dumps(request_body, indent=4, default=str))
# Assert content from the knowedge base was applied to the request
# 1. we should have 2 content blocks, the first is the context from the knowledge base, the second is the user message
content = request_body["messages"][0]["content"]
assert len(content) == 2
assert content[0]["type"] == "text"
assert content[1]["type"] == "text"
# 2. the first content block should have the bedrock knowledge base prefix string
# this helps confirm that the context from the knowledge base was applied to the request
assert VectorStorePreCallHook.CONTENT_PREFIX_STRING in content[0]["text"]
@pytest.mark.asyncio
async def test_e2e_bedrock_knowledgebase_retrieval_with_llm_api_call(
setup_vector_store_registry,
):
"""
Test that the Bedrock Knowledge Base Hook works when making a real llm api call and returns citations.
"""
# Init client
litellm._turn_on_debug()
async_client = AsyncHTTPHandler()
response = await litellm.acompletion(
model="bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0",
messages=[{"role": "user", "content": "what is litellm?"}],
vector_store_ids=["LCYXFBR2TU"],
client=async_client,
)
print("OPENAI RESPONSE:", json.dumps(dict(response), indent=4, default=str))
assert response is not None
# Check that search_results are present in provider_specific_fields
assert hasattr(response.choices[0].message, "provider_specific_fields")
provider_fields = response.choices[0].message.provider_specific_fields
assert provider_fields is not None
assert "search_results" in provider_fields
search_results = provider_fields["search_results"]
assert search_results is not None
assert len(search_results) > 0
# Check search result structure (OpenAI-compatible format)
first_search_result = search_results[0]
assert "object" in first_search_result
assert first_search_result["object"] == "vector_store.search_results.page"
assert "data" in first_search_result
assert len(first_search_result["data"]) > 0
# Check individual result structure
first_result = first_search_result["data"][0]
assert "score" in first_result
assert "content" in first_result
print(f"Search results returned: {len(search_results)}")
print(f"First search result has {len(first_search_result['data'])} items")
@pytest.mark.asyncio
async def test_e2e_bedrock_knowledgebase_retrieval_with_llm_api_call_streaming(
setup_vector_store_registry,
):
"""
Test that the Bedrock Knowledge Base Hook works with streaming and returns search_results in chunks.
"""
# Init client
# litellm._turn_on_debug()
async_client = AsyncHTTPHandler()
response = await litellm.acompletion(
model=f"anthropic/{os.environ.get('CI_CD_DEFAULT_ANTHROPIC_MODEL', 'claude-haiku-4-5-20251001')}",
messages=[{"role": "user", "content": "what is litellm?"}],
vector_store_ids=["LCYXFBR2TU"],
stream=True,
client=async_client,
)
# Collect chunks
chunks = []
search_results_found = False
async for chunk in response:
chunks.append(chunk)
print(f"Chunk: {chunk}")
# Check if this chunk has search_results in provider_specific_fields
if hasattr(chunk, "choices") and chunk.choices:
for choice in chunk.choices:
if hasattr(choice, "delta") and choice.delta:
provider_fields = getattr(
choice.delta, "provider_specific_fields", None
)
if provider_fields and "search_results" in provider_fields:
search_results = provider_fields["search_results"]
print(
f"Found search_results in streaming chunk: {len(search_results)} results"
)
# Verify structure
assert search_results is not None
assert len(search_results) > 0
first_search_result = search_results[0]
assert "object" in first_search_result
assert (
first_search_result["object"]
== "vector_store.search_results.page"
)
assert "data" in first_search_result
assert len(first_search_result["data"]) > 0
search_results_found = True
print(f"Total chunks received: {len(chunks)}")
assert len(chunks) > 0
assert search_results_found, "search_results should be present in streaming chunks"
@pytest.mark.asyncio
async def test_e2e_bedrock_knowledgebase_retrieval_with_llm_api_call_with_tools(
setup_vector_store_registry,
):
"""
Test that the Bedrock Knowledge Base Hook works when making a real llm api call
"""
# Init client
litellm._turn_on_debug()
response = await litellm.acompletion(
model=f"anthropic/{os.environ.get('CI_CD_DEFAULT_ANTHROPIC_MODEL', 'claude-haiku-4-5-20251001')}",
messages=[{"role": "user", "content": "what is litellm?"}],
max_tokens=10,
tools=[{"type": "file_search", "vector_store_ids": ["LCYXFBR2TU"]}],
)
assert response is not None
@pytest.mark.asyncio
async def test_e2e_bedrock_knowledgebase_retrieval_with_llm_api_call_with_tools_and_filters(
setup_vector_store_registry,
):
"""
Test that filters from file_search tools are properly passed through to vector store search.
This test verifies the entire flow: tool parsing -> filter extraction -> vector store API call.
In this case we filter for a non-existent user_id, which should return no results.
"""
litellm._turn_on_debug()
response = await litellm.acompletion(
model=f"anthropic/{os.environ.get('CI_CD_DEFAULT_ANTHROPIC_MODEL', 'claude-haiku-4-5-20251001')}",
messages=[{"role": "user", "content": "what is litellm?"}],
max_tokens=10,
tools=[
{
"type": "file_search",
"vector_store_ids": ["LCYXFBR2TU"],
"filters": {
"key": "user_id",
"value": "fake-user-id",
"operator": "eq",
},
}
],
)
# Verify response is not None
assert response is not None
# Verify search results were added to the response (this proves the search was called)
assert hasattr(response.choices[0].message, "provider_specific_fields")
provider_fields = response.choices[0].message.provider_specific_fields
assert provider_fields is not None
assert (
"search_results" in provider_fields
), "search_results not in provider_specific_fields"
search_results = provider_fields["search_results"]
assert (
search_results is not None and len(search_results) > 0
), "No search results found"
# The search was performed - this confirms filters were passed through
# The logs above show: litellm.asearch(... filters={'key': 'user_id', 'value': 'fake-user-id', 'operator': 'eq'})
# And the Bedrock API request contains: {'filter': {'equals': {'key': 'user_id', 'value': 'fake-user-id'}}}
print("✅ Filters were successfully passed through to vector store search")
print(f" Search was performed and {len(search_results)} result(s) returned")
@pytest.mark.asyncio
async def test_bedrock_kb_request_body_has_transformed_filters(
setup_vector_store_registry,
):
"""
Validate that the Bedrock Knowledge Base request body contains the transformed filters.
"""
captured_request_body: dict = {}
async def fake_async_vector_store_search_handler(
vector_store_id,
query,
vector_store_search_optional_params,
vector_store_provider_config,
custom_llm_provider,
litellm_params,
logging_obj,
extra_headers=None,
extra_body=None,
timeout=None,
client=None,
_is_async=False,
):
litellm_params_dict = (
litellm_params.model_dump(exclude_none=False)
if hasattr(litellm_params, "model_dump")
else dict(litellm_params)
)
api_base = vector_store_provider_config.get_complete_url(
api_base=litellm_params_dict.get("api_base"),
litellm_params=litellm_params_dict,
)
url, request_body = (
vector_store_provider_config.transform_search_vector_store_request(
vector_store_id=vector_store_id,
query=query,
vector_store_search_optional_params=vector_store_search_optional_params,
api_base=api_base,
litellm_logging_obj=logging_obj,
litellm_params=litellm_params_dict,
extra_body=None,
)
)
captured_request_body["url"] = url
captured_request_body["body"] = request_body
return VectorStoreSearchResponse(
object="vector_store.search_results.page",
search_query=query if isinstance(query, str) else " ".join(query),
data=[
VectorStoreSearchResult(
score=0.9,
content=[
VectorStoreResultContent(
text="LiteLLM is a library", type="text"
)
],
)
],
)
with patch.object(
litellm.vector_stores.main.base_llm_http_handler,
"async_vector_store_search_handler",
new=AsyncMock(side_effect=fake_async_vector_store_search_handler),
):
response = await litellm.acompletion(
model=f"anthropic/{os.environ.get('CI_CD_DEFAULT_ANTHROPIC_MODEL', 'claude-haiku-4-5-20251001')}",
messages=[{"role": "user", "content": "what is litellm?"}],
max_tokens=10,
tools=[
{
"type": "file_search",
"vector_store_ids": ["LCYXFBR2TU"],
"filters": {
"key": "user_id",
"value": "fake-user-id",
"operator": "eq",
},
}
],
)
assert response is not None
print(
"captured_request_body:",
json.dumps(captured_request_body, indent=4, default=str),
)
assert "body" in captured_request_body, "Bedrock KB request body was not captured"
vector_search = captured_request_body["body"]["retrievalConfiguration"][
"vectorSearchConfiguration"
]
aws_filter = vector_search["filter"]
assert "equals" in aws_filter, f"Expected 'equals' in AWS format, got: {aws_filter}"
assert aws_filter["equals"]["key"] == "user_id"
assert aws_filter["equals"]["value"] == "fake-user-id"
print("✅ Filters transformed correctly: OpenAI format -> AWS Bedrock format")
@pytest.mark.asyncio
async def test_openai_with_knowledge_base_mock_openai(setup_vector_store_registry):
"""
Tests that knowledge base content is correctly passed to the OpenAI API call
"""
litellm.set_verbose = True
from openai import AsyncOpenAI
client = AsyncOpenAI(api_key="fake-api-key")
# Variable to capture the request
captured_request = {}
with patch.object(
client.chat.completions.with_raw_response, "create"
) as mock_client:
# Create async mock that returns proper structure
async def mock_create(**kwargs):
mock_response = Mock()
mock_response.choices = [
Mock(
message=Mock(content="Mock response from OpenAI", role="assistant")
)
]
mock_response.usage = Mock(
prompt_tokens=100, completion_tokens=50, total_tokens=150
)
mock_response.id = "chatcmpl-123"
mock_response.object = "chat.completion"
mock_response.created = 1234567890
mock_response.model = "gpt-5.5"
# Store the request for verification
captured_request.update(kwargs)
# Return wrapper with parse method
wrapper = Mock()
wrapper.parse.return_value = mock_response
return wrapper
mock_client.side_effect = mock_create
try:
await litellm.acompletion(
model="gpt-5.5",
messages=[{"role": "user", "content": "what is litellm?"}],
vector_store_ids=["LCYXFBR2TU"],
client=client,
)
except Exception as e:
print(f"Error: {e}")
# Verify the API was called
mock_client.assert_called_once()
request_body = captured_request
# Verify the request contains messages with knowledge base context
assert "messages" in request_body
messages = request_body["messages"]
# We expect at least 2 messages:
# 1. User message with the knowledge base context
# 2. User message with the question
assert len(messages) >= 2
print("request messages:", json.dumps(messages, indent=4, default=str))
# assert message[0] is the user message with the knowledge base context
assert messages[0]["role"] == "user"
assert VectorStorePreCallHook.CONTENT_PREFIX_STRING in messages[0]["content"]
@pytest.mark.asyncio
async def test_openai_with_vector_store_ids_in_tool_call_mock_openai(
setup_vector_store_registry,
):
"""
Tests that vector store ids can be passed as tools
This is the OpenAI format
"""
litellm.set_verbose = True
from openai import AsyncOpenAI
client = AsyncOpenAI(api_key="fake-api-key")
# Variable to capture the request
captured_request = {}
with patch.object(
client.chat.completions.with_raw_response, "create"
) as mock_client:
# Create async mock that returns proper structure
async def mock_create(**kwargs):
mock_response = Mock()
mock_response.choices = [
Mock(
message=Mock(content="Mock response from OpenAI", role="assistant")
)
]
mock_response.usage = Mock(
prompt_tokens=100, completion_tokens=50, total_tokens=150
)
mock_response.id = "chatcmpl-123"
mock_response.object = "chat.completion"
mock_response.created = 1234567890
mock_response.model = "gpt-5.5"
# Store the request for verification
captured_request.update(kwargs)
# Return wrapper with parse method
wrapper = Mock()
wrapper.parse.return_value = mock_response
return wrapper
mock_client.side_effect = mock_create
try:
await litellm.acompletion(
model="gpt-5.5",
messages=[{"role": "user", "content": "what is litellm?"}],
tools=[{"type": "file_search", "vector_store_ids": ["LCYXFBR2TU"]}],
client=client,
)
except Exception as e:
print(f"Error: {e}")
# Verify the API was called
mock_client.assert_called_once()
request_body = captured_request
print("request body:", json.dumps(request_body, indent=4, default=str))
# Verify the request contains messages with knowledge base context
assert "messages" in request_body
messages = request_body["messages"]
# We expect at least 2 messages:
# 1. User message with the knowledge base context
# 2. User message with the question
assert len(messages) >= 2
print("request messages:", json.dumps(messages, indent=4, default=str))
# assert message[0] is the user message with the knowledge base context
assert messages[0]["role"] == "user"
assert VectorStorePreCallHook.CONTENT_PREFIX_STRING in messages[0]["content"]
# assert that the tool call was not sent to the upstream llm API if it's a litellm vector store
assert "tools" not in request_body
@pytest.mark.asyncio
async def test_openai_with_mixed_tool_call_mock_openai(setup_vector_store_registry):
"""Ensure unrecognized vector store tools are forwarded to the provider"""
from openai import AsyncOpenAI
client = AsyncOpenAI(api_key="fake-api-key")
# Variable to capture the request
captured_request = {}
with patch.object(
client.chat.completions.with_raw_response, "create"
) as mock_client:
# Create async mock that returns proper structure
async def mock_create(**kwargs):
mock_response = Mock()
mock_response.choices = [
Mock(
message=Mock(content="Mock response from OpenAI", role="assistant")
)
]
mock_response.usage = Mock(
prompt_tokens=100, completion_tokens=50, total_tokens=150
)
mock_response.id = "chatcmpl-123"
mock_response.object = "chat.completion"
mock_response.created = 1234567890
mock_response.model = "gpt-5.5"
# Store the request for verification
captured_request.update(kwargs)
# Return wrapper with parse method
wrapper = Mock()
wrapper.parse.return_value = mock_response
return wrapper
mock_client.side_effect = mock_create
try:
await litellm.acompletion(
model="gpt-5.5",
messages=[{"role": "user", "content": "what is litellm?"}],
tools=[
{"type": "file_search", "vector_store_ids": ["LCYXFBR2TU"]},
{"type": "file_search", "vector_store_ids": ["unknownVS"]},
],
client=client,
)
except Exception as e:
print(f"Error: {e}")
mock_client.assert_called_once()
request_body = captured_request
assert "messages" in request_body
messages = request_body["messages"]
assert len(messages) >= 2
assert messages[0]["role"] == "user"
assert VectorStorePreCallHook.CONTENT_PREFIX_STRING in messages[0]["content"]
assert "tools" in request_body
tools = request_body["tools"]
assert len(tools) == 1
assert tools[0]["vector_store_ids"] == ["unknownVS"]
# @pytest.mark.asyncio
# async def test_logging_with_knowledge_base_hook(setup_vector_store_registry):
# """
# Test that the knowledge base request was logged in standard logging payload
# """
# test_custom_logger = MockCustomLogger()
# litellm.set_verbose = True
# await litellm.acompletion(
# model="gpt-5.5",
# messages=[{"role": "user", "content": "what is litellm?"}],
# vector_store_ids = [
# "LCYXFBR2TU"
# ],
# )
# # sleep for 1 second to allow the logging callback to run
# await asyncio.sleep(1)
# # assert that the knowledge base request was logged in the standard logging payload
# standard_logging_payload: Optional[StandardLoggingPayload] = test_custom_logger.standard_logging_payload
# assert standard_logging_payload is not None
# metadata = standard_logging_payload["metadata"]
# standard_logging_vector_store_request_metadata: Optional[List[StandardLoggingVectorStoreRequest]] = metadata["vector_store_request_metadata"]
# print("standard_logging_vector_store_request_metadata:", json.dumps(standard_logging_vector_store_request_metadata, indent=4, default=str))
# # 1 vector store request was made, expect 1 vector store request metadata object
# assert len(standard_logging_vector_store_request_metadata) == 1
# # expect the vector store request metadata object to have the correct values
# vector_store_request_metadata = standard_logging_vector_store_request_metadata[0]
# assert vector_store_request_metadata.get("vector_store_id") == "LCYXFBR2TU"
# assert vector_store_request_metadata.get("query") == "what is litellm?"
# assert vector_store_request_metadata.get("custom_llm_provider") == "bedrock"
# vector_store_search_response: VectorStoreSearchResponse = vector_store_request_metadata.get("vector_store_search_response")
# assert vector_store_search_response is not None
# assert vector_store_search_response.get("search_query") == "what is litellm?"
# assert len(vector_store_search_response.get("data", [])) >=0
# for item in vector_store_search_response.get("data", []):
# assert item.get("score") is not None
# assert item.get("content") is not None
# assert len(item.get("content", [])) >= 0
# for content_item in item.get("content", []):
# text_content = content_item.get("text")
# assert text_content is not None
# assert len(text_content) > 0
@pytest.mark.asyncio
async def test_e2e_bedrock_knowledgebase_retrieval_without_vector_store_registry(
setup_vector_store_registry,
):
litellm._turn_on_debug()
client = AsyncHTTPHandler()
litellm.vector_store_registry = None
with patch.object(client, "post") as mock_post:
# Mock the response for the LLM call
mock_response = Mock()
mock_response.status_code = 200
mock_response.headers = {"Content-Type": "application/json"}
# Provide proper JSON response content
mock_response.text = json.dumps(
{
"id": "msg_01ABC123",
"type": "message",
"role": "assistant",
"content": [
{
"type": "text",
"text": "LiteLLM is a library that simplifies LLM API access.",
}
],
"model": "claude-3.5-sonnet",
"stop_reason": "end_turn",
"stop_sequence": None,
"usage": {"input_tokens": 100, "output_tokens": 50},
}
)
mock_response.json = lambda: json.loads(mock_response.text)
mock_post.return_value = mock_response
try:
response = await litellm.acompletion(
model="anthropic/claude-3.5-sonnet",
messages=[{"role": "user", "content": "what is litellm?"}],
vector_store_ids=["LCYXFBR2TU"],
client=client,
)
except Exception as e:
print(f"Error: {e}")
# Verify the LLM request was made
mock_post.assert_called_once()
# Verify the request body
print("call args:", mock_post.call_args)
request_body = mock_post.call_args.kwargs["json"]
print("Request body:", json.dumps(request_body, indent=4, default=str))
# Assert content from the knowedge base was applied to the request
# 1. we should have 1 content block, the first is the user message
# There should only be one since there is no initialized vector store registry
content = request_body["messages"][0]["content"]
assert len(content) == 1
assert content[0]["type"] == "text"
@pytest.mark.asyncio
async def test_e2e_bedrock_knowledgebase_retrieval_with_vector_store_not_in_registry(
setup_vector_store_registry,
):
"""
No vector store request is made for vector store ids that are not in the registry
In this test newUnknownVectorStoreId is not in the registry, so no vector store request is made
"""
litellm._turn_on_debug()
client = AsyncHTTPHandler()
if litellm.vector_store_registry is not None:
print("Registry iniitalized:", litellm.vector_store_registry.vector_stores)
else:
print("Registry is None")
with patch.object(client, "post") as mock_post:
# Mock the response for the LLM call
mock_response = Mock()
mock_response.status_code = 200
mock_response.headers = {"Content-Type": "application/json"}
# Provide proper JSON response content
mock_response.text = json.dumps(
{
"id": "msg_01ABC123",
"type": "message",
"role": "assistant",
"content": [
{
"type": "text",
"text": "LiteLLM is a library that simplifies LLM API access.",
}
],
"model": "claude-3.5-sonnet",
"stop_reason": "end_turn",
"stop_sequence": None,
"usage": {"input_tokens": 100, "output_tokens": 50},
}
)
mock_response.json = lambda: json.loads(mock_response.text)
mock_post.return_value = mock_response
try:
response = await litellm.acompletion(
model="anthropic/claude-3.5-sonnet",
messages=[{"role": "user", "content": "what is litellm?"}],
vector_store_ids=["newUnknownVectorStoreId"],
client=client,
)
except Exception as e:
print(f"Error: {e}")
# Verify the LLM request was made
mock_post.assert_called_once()
# Verify the request body
print("call args:", mock_post.call_args)
request_body = mock_post.call_args.kwargs["json"]
print("Request body:", json.dumps(request_body, indent=4, default=str))
# Assert content from the knowedge base was applied to the request
# 1. we should have 1 content block, the first is the user message
# There should only be one since there is no initialized vector store registry
content = request_body["messages"][0]["content"]
assert len(content) == 1
assert content[0]["type"] == "text"
@pytest.mark.asyncio
async def test_provider_specific_fields_in_proxy_http_response(
setup_vector_store_registry,
):
"""
Test that provider_specific_fields (like search_results) are included
in the proxy HTTP JSON response, not just in Python SDK objects.
This test catches serialization bugs where exclude=True would strip
provider_specific_fields from the HTTP response.
"""
from fastapi.testclient import TestClient
from litellm.proxy.proxy_server import app, initialize
from litellm.proxy.utils import ProxyLogging
import litellm.proxy.proxy_server as proxy_server
from unittest.mock import patch as mock_patch
# Initialize proxy
await initialize(
model="gpt-5-mini",
alias=None,
api_base=None,
debug=False,
temperature=None,
max_tokens=None,
request_timeout=600,
max_budget=None,
telemetry=False,
drop_params=True,
add_function_to_prompt=False,
headers=None,
save=False,
use_queue=False,
config=None,
)
# Create test client
client = TestClient(app)
# Create mock response with provider_specific_fields
mock_response = litellm.ModelResponse(
id="test-123",
model="gpt-5-mini",
created=1234567890,
object="chat.completion",
)
# Create message with provider_specific_fields
mock_message = litellm.Message(
content="LiteLLM is a tool that simplifies working with multiple LLMs.",
role="assistant",
provider_specific_fields={
"search_results": [
{
"object": "vector_store.search_results.page",
"search_query": "what is litellm?",
"data": [
{
"score": 0.95,
"content": [{"text": "Test content", "type": "text"}],
"file_id": "test-file",
"filename": "test.txt",
}
],
}
]
},
)
mock_choice = litellm.Choices(finish_reason="stop", index=0, message=mock_message)
mock_response.choices = [mock_choice]
mock_response.usage = litellm.Usage(
prompt_tokens=10, completion_tokens=20, total_tokens=30
)
# Patch the completion call at the proxy level
with mock_patch("litellm.acompletion", new=AsyncMock(return_value=mock_response)):
# Make HTTP request to proxy
response = client.post(
"/v1/chat/completions",
json={
"model": "gpt-5-mini",
"messages": [{"role": "user", "content": "What is litellm?"}],
},
)
# Check HTTP response
assert response.status_code == 200
result = response.json()
print("HTTP Response JSON:", json.dumps(result, indent=2))
# THE KEY ASSERTIONS - These would FAIL with exclude=True!
assert "choices" in result
assert len(result["choices"]) > 0
choice = result["choices"][0]
assert "message" in choice
message = choice["message"]
# Verify provider_specific_fields is in the JSON response
assert (
"provider_specific_fields" in message
), "provider_specific_fields missing from HTTP JSON response! This means exclude=True is preventing serialization."
assert "search_results" in message["provider_specific_fields"]
search_results = message["provider_specific_fields"]["search_results"]
assert len(search_results) > 0
# Verify search result structure
first_result = search_results[0]
assert first_result["object"] == "vector_store.search_results.page"
assert "data" in first_result
assert len(first_result["data"]) > 0
print("✅ provider_specific_fields successfully serialized in HTTP response")