The Complete Production SOP: How to Deploy and Run ChromaDB Embedded Vector Store & Persistent Collections in Autonomous AI Agents

---
name: "chromadb-persistent-embeddings"
description: "Embed and query documents locally or in client-server mode using ChromaDB."
metadata:
  author: "Chroma"
  version: "1.0.0"
  runtime_compatibility:
    - "codex"
    - "claude-code"
    - "cursor"
    - "windsurf"
triggers:
  - "init chromadb persistent client"
  - "query chromadb collection"
  - "chromadb where metadata filter"
permissions:
  fs_read: true
  fs_write: "workspace"
  shell_exec: false
  net_access: false
---

# ChromaDB Embedded Vector Store & Persistent Collections

Manages local vector collections in ChromaDB: configuring persistent DuckDB/SQLite storage, embedding functions (OpenAI, HuggingFace, ONNX), metadata filtering, and semantic document querying.

## When to Invoke
Invoke this skill whenever you need to:
- init chromadb persistent client
- query chromadb collection
- chromadb where metadata filter

## Execution SOP & Rules
1. Create persistent client: client = chromadb.PersistentClient(path='./chroma_data').
2. Get or create collection: collection = client.get_or_create_collection('knowledge_base').
3. Add records: collection.add(documents=['...'], metadatas=[{'topic': 'ai'}], ids=['id1']).
4. Query documents: results = collection.query(query_texts=['What is RAG?'], n_results=3).

## Verification Checklist
- Run local linting, syntax validation, and type-checks before reporting completion.
- Ensure all created or modified files adhere to workspace boundaries and avoid uncommitted secrets.
- Verify tests or build commands exit cleanly with return code 0.

import chromadb
from chromadb.config import Settings
from typing import List, Dict, Any


def get_vector_collection(persist_dir: str = "./chroma_data", collection_name: str = "knowledge_base"):
    """Initializes a persistent ChromaDB client and resolves the target collection."""
    client = chromadb.PersistentClient(
        path=persist_dir,
        settings=Settings(anonymized_telemetry=False, allow_reset=True)
    )
    return client.get_or_create_collection(
        name=collection_name,
        metadata={"hnsw:space": "cosine"}
    )


def upsert_documents(collection, docs: List[str], metadatas: List[Dict[str, Any]], ids: List[str]):
    """Upserts records to prevent duplicate ID runtime exceptions."""
    collection.upsert(documents=docs, metadatas=metadatas, ids=ids)


def search_documents(collection, query: str, topic_filter: str, limit: int = 3):
    """Executes semantic query with deterministic ChromaDB metadata where filters."""
    return collection.query(
        query_texts=[query],
        n_results=limit,
        where={"topic": {"$eq": topic_filter}}
    )