[BUG] CrewAI ChromaDB Embedding Dimension Mismatch Issue

[amdjedbens]

Description

When using CrewAI with knowledge sources, I'm encountering an embedding dimension mismatch error if I've previously used a different embedding model in the same project. This appears to happen because CrewAI uses ChromaDB as its default vector database, and ChromaDB enforces consistent embedding dimensions across operations.

[ERROR]: Embedding dimension mismatch. This usually happens when mixing different embedding models.
Try resetting the collection using `crewai reset-memories -a`

ValueError: Invalid Knowledge Configuration: Embedding dimension mismatch. Make sure you're using the same embedding model across all operations with this collection.
Try resetting the collection using `crewai reset-memories -a`

The issue shows up as a dimension mismatch error (e.g., 768 vs 1536) between current embeddings and previously stored embeddings.

Steps to Reproduce

1. Create a CrewAI project with agents that use knowledge sources
2. Run the project with one embedding model (e.g., OpenAI's model with 1536 dimensions)
3. Change the embedding model to a different one (e.g., Ollama's nomic-embed-text with 768 dimensions)
4. Run the project again without clearing previous embeddings

Expected behavior

The project should either:

• Detect the embedding model change and automatically reset collections
• Convert embeddings to be compatible
• Provide a clearer error message with automated recovery

Current Behavior

The project fails with a cryptic ChromaDB error about dimension mismatch that is confusing since there's no clear indication that CrewAI is using ChromaDB under the hood.

I've tried running the suggested command crewai reset-memories -a but didn't work as well

Help Needed

Has anyone encountered this issue and found a reliable solution? I need a way to either:

1. Properly reset the ChromaDB collections
2. Configure CrewAI to use a different vector database
3. Ensure consistent embedding dimensions across runs

Screenshots/Code snippets

# My embedding configuration
embedder_config = {
    "provider": "ollama",
    "config": {
        "model": "nomic-embed-text",
        "api_url": "http://localhost:11434",
    },
}

# Knowledge source initialization
data_knowledge_source = JSONKnowledgeSource(
    file_paths=["data_source.json"],
    embedder=embedder_config,
    collection_name=f"collection_{timestamp}"
)

# Create generic agents
data_analyst = Agent(
    role="Data Analyst",
    goal="Analyze data and extract insights",
    backstory="Experienced data analyst with expertise in pattern recognition",
    tools=[data_tool],
    knowledge_sources=[data_knowledge_source],
    verbose=False
)

report_writer = Agent(
    role="Report Writer",
    goal="Create comprehensive reports from data analysis",
    backstory="Expert in creating clear, actionable reports",
    tools=[data_tool],
    knowledge_sources=[data_knowledge_source],
    verbose=False
)

# Crew setup with knowledge sources
crew = Crew(
    agents=[data_analyst, report_writer],
    tasks=[analyze_task, report_task],
    process=Process.sequential,
    verbose=True,
    embedder=embedder_config,
    memory=True,
    short_term_memory=ShortTermMemory(
        storage=RAGStorage(
            embedder_config=embedder_config,
            type="short_term",
            path="db/memory.json"
        ),
    ),
    knowledge_sources=[data_knowledge_source],
)

# Reset attempt that doesn't work
crew.reset_memories(command_type="all")

### Operating System

macOS Sonoma

### Python Version

3.12

### crewAI Version

0.108.0

### crewAI Tools Version

0.38.1

### Virtual Environment

Venv

### Evidence

[2025-03-25 14:30:17][ERROR]: Embedding dimension mismatch. This usually happens when mixing different embedding models. Try resetting the collection using crewai reset-memories -a ╭─────────────────────────────────────────────────────────────────────────────── Crew Failure ───────────────────────────────────────────────────────────────────────────────╮ │ │ │ Crew Execution Failed │ │ Name: crew │ │ ID: d1616bf2-90bc-44c4-a32d-4042b318482b │ │ │ │ │ ╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

Traceback (most recent call last): File "/Users/.pyenv/versions/3.12.3/lib/python3.12/site-packages/crewai/knowledge/storage/knowledge_storage.py", line 161, in save self.collection.upsert( File "/Users/.pyenv/versions/3.12.3/lib/python3.12/site-packages/chromadb/api/models/Collection.py", line 343, in upsert self._client._upsert( File "/Users/.pyenv/versions/3.12.3/lib/python3.12/site-packages/chromadb/telemetry/opentelemetry/init.py", line 150, in wrapper return f(*args, **kwargs) ^^^^^^^^^^^^^^^^^^ File "/Users/.pyenv/versions/3.12.3/lib/python3.12/site-packages/chromadb/api/segment.py", line 103, in wrapper return self._rate_limit_enforcer.rate_limit(func)(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/Users/.pyenv/versions/3.12.3/lib/python3.12/site-packages/chromadb/rate_limit/simple_rate_limit/init.py", line 23, in wrapper return func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/Users/.pyenv/versions/3.12.3/lib/python3.12/site-packages/chromadb/api/segment.py", line 536, in _upsert self._validate_embedding_record_set(coll, records_to_submit) File "/Users/.pyenv/versions/3.12.3/lib/python3.12/site-packages/chromadb/telemetry/opentelemetry/init.py", line 150, in wrapper return f(*args, **kwargs) ^^^^^^^^^^^^^^^^^^ File "/Users/.pyenv/versions/3.12.3/lib/python3.12/site-packages/chromadb/api/segment.py", line 864, in _validate_embedding_record_set self._validate_dimension( File "/Users/.pyenv/versions/3.12.3/lib/python3.12/site-packages/chromadb/api/segment.py", line 881, in _validate_dimension raise InvalidDimensionException( chromadb.errors.InvalidDimensionException: Embedding dimension 768 does not match collection dimensionality 1536

### Possible Solution

None

### Additional context

This issue typically happens when:
1. Switching between embedding providers (OpenAI to local models or vice versa)
2. Changing embedding models within the same provider
3. Testing different configurations with the same codebase

[Vidit-Ostwal]

Hi @amdjedbens, the crewai reset-memories -a command will get fixed soon.

checkout this PR : #2312 The changes are very less, you can try to pull these changes, while this PR gets merged. Let me know if this works.

[lucasgomide]

@Vidit-Ostwal so can we close this PR

[Vidit-Ostwal]

Hi @lucasgomide, yes I think we can close this PR.

[lucasgomide]

@Vidit-Ostwal can you mention on #2312 that it solves this issue for tracking purposes?

[GabrielBoninUnity]

Following up on this issue: The CLI command crewai reset-memories -a is still not working for me, even after updating to the latest (PR #2312).

Since I'm using macOS, I had to take the following steps to resolve the problem:

If you're changing the embedder model and encounter the error [ERROR]: Embedding dimension mismatch, you'll need to reset the memory. However, because the CLI tool command crewai reset-memories -a isn't functional for me, I had to manually delete the SQLite database that's used for storing memories.

Location of the database file:

/Users/your-user-name/Library/Application Support/name-of-your-project You'll need to navigate to this location manually and delete the file yourself.

I hope this workaround helps anyone facing the same problem!

[Vidit-Ostwal]

@amdjedbens, Can you update the crewai version and check whether this is working or not?

[github-actions[bot]]

This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[github-actions[bot]]

This issue was closed because it has been stalled for 5 days with no activity.