[Bug]: PyInstaller support for CLI app fails due to dynamic imports and EmbeddingFunction validation

[memelab74]

What happened?

Summary

I'm building a CLI-based chatbot prototype using ChromaDB with SentenceTransformers and a persistent local database. To make it shareable with non-technical users, I attempted to package it using PyInstaller — but ran into significant issues due to dynamic imports and embedding function validation in ChromaDB (v0.4.16+).

This issue documents those findings and may be helpful to others attempting similar packaging strategies. Yes, AI helped me write this, and I hope it's helpful to you.

---

Environment

• OS: macOS (ARM64, Python 3.12.9 via pyenv)
• ChromaDB: 0.4.16
• Backend: PersistentClient(path="./chroma_db") (SQLite)
• Embedding: Custom EmbeddingFunction using sentence-transformers
• Packager: PyInstaller 6.5.0
• App: CLI chatbot that queries multiple collections

---

Problem 1: EmbeddingFunction interface mismatch

Error:

ValueError: Expected EmbeddingFunction.__call__ to have the following signature: odict_keys(['self', 'input']), got odict_keys(['self', 'args', 'kwargs'])

🔧 Fix: Wrapped model like this:

class ChromaEmbeddingFunction(EmbeddingFunction):
    def __init__(self, model):
        self.model = model

    def __call__(self, input):
        if isinstance(input, str):
            input = [input]
        return self.model.encode(input).tolist()

---

Problem 2: Dynamic import errors in PyInstaller build

Each PyInstaller run failed with a new error. Examples:

• ModuleNotFoundError: No module named 'chromadb.telemetry.product.posthog'
• ModuleNotFoundError: No module named 'chromadb.segment.impl.metadata.sqlite'
• ModuleNotFoundError: No module named 'chromadb.execution.executor.local'
• ...and several others

🔧 Fix: Required a large list of hiddenimports in the .spec file:

hiddenimports = [
    'chromadb.telemetry.product.posthog',
    'chromadb.api.segment',
    'chromadb.db.impl',
    'chromadb.db.impl.sqlite',
    'chromadb.migrations',
    'chromadb.migrations.embeddings_queue',
    'chromadb.segment.impl.manager',
    'chromadb.segment.impl.manager.local',
    'chromadb.segment.impl.metadata',
    'chromadb.segment.impl.metadata.sqlite',
    'chromadb.segment.impl.vector',
    'chromadb.execution.executor.local',
    'chromadb.quota.simple_quota_enforcer',
    'chromadb.rate_limit.simple_rate_limit',
    'analytics',  # dependency for posthog
]

---

Suggestions

• 📘 Documentation: It would be helpful to document Chroma’s reliance on dynamic imports and what embedding/segment modules are required for local use.
• ⚙️ Static mode?: Consider offering a “static embedding + sqlite” mode that avoids lazy loading or plugin discovery — ideal for CLI and embedded apps.

---

Final Notes

If helpful, I'm happy to share:

• A working CLI demo script
• The full .spec file
• Repro steps and final packaged CLI tool

Thanks for the great work — ChromaDB is fantastic for local, lightweight AI apps. Just wanted to surface some of the rough edges when packaging for offline users.

Versions

• OS: macOS (ARM64, Python 3.12.9 via pyenv)
• ChromaDB: 0.4.16
• Backend: PersistentClient(path="./chroma_db") (SQLite)
• Embedding: Custom EmbeddingFunction using sentence-transformers
• Packager: PyInstaller 6.5.0
• App: CLI chatbot that queries multiple collections

Relevant log output

[jeffchuber]

Chroma is now on 1.0.x - can you give that a shot?