pip install hystersis cohere

import cohere
from hystersis import Hystersis

hystersis = Hystersis(api_key="your-hystersis-key")
co = cohere.Client(api_key="your-cohere-key")

def chat_with_memory(user_id: str, message: str) -> str:
    # 1. Retrieve relevant memories
    memories = hystersis.search(
        query=message,
        user_id=user_id,
        limit=10
    )

    # 2. Rerank memories with Cohere for better relevance
    if memories:
        documents = [m["content"] for m in memories]
        rerank_response = co.rerank(
            model="rerank-v3.5",
            query=message,
            documents=documents,
            top_n=5
        )
        reranked_memories = [memories[r.index] for r in rerank_response.results]
    else:
        reranked_memories = []

    # 3. Build context from reranked memories
    memory_context = "\n".join([
        f"- {m['content']}" for m in reranked_memories
    ])

    # 4. Generate response with Cohere
    response = co.chat(
        model="command-r-plus",
        message=message,
        preamble=f"You are a helpful assistant with memory. Relevant context:\n{memory_context}"
    )

    # 5. Store the conversation as memory
    hystersis.create_memory(
        content=f"User: {message} | Assistant: {response.text}",
        user_id=user_id,
        compression_mode="extract"
    )

    return response.text

# Usage
result = chat_with_memory("user_123", "I prefer dark mode in my IDE")
print(result)

def search_with_reranking(user_id: str, query: str, limit: int = 20):
    # 1. Retrieve more results from Hystersis
    memories = hystersis.search(
        query=query,
        user_id=user_id,
        limit=limit * 3  # Get 3x more candidates
    )

    if not memories:
        return []

    # 2. Rerank with Cohere
    documents = [m["content"] for m in memories]
    rerank_response = co.rerank(
        model="rerank-v3.5",
        query=query,
        documents=documents,
        top_n=limit
    )

    # 3. Return reranked results
    return [
        {
            **memories[r.index],
            "rerank_score": r.relevance_score
        }
        for r in rerank_response.results
    ]

results = search_with_reranking("user_123", "project deadlines")
for r in results:
    print(f"{r['content']} (score: {r['rerank_score']:.3f})")

def create_memory_with_embeddings(user_id: str, content: str):
    # Generate embeddings with Cohere
    embed_response = co.embed(
        model="embed-v4",
        texts=[content],
        input_type="search_document"
    )

    # Store in Hystersis
    memory = hystersis.create_memory(
        content=content,
        user_id=user_id,
        metadata={
            "embedding_model": "cohere-embed-v4",
            "embedding_dim": len(embed_response.embeddings[0])
        }
    )

    return memory

def stream_chat(user_id: str, message: str):
    memories = hystersis.search(query=message, user_id=user_id, limit=5)
    memory_context = "\n".join([f"- {m['content']}" for m in memories])

    # Stream response from Cohere
    response = co.chat_stream(
        model="command-r-plus",
        message=message,
        preamble=f"You have persistent memory. Context:\n{memory_context}"
    )

    full_response = ""
    for event in response:
        if event.event_type == "text-generation":
            print(event.text, end="", flush=True)
            full_response += event.text

    # Store in memory
    hystersis.create_memory(
        content=f"User: {message} | Assistant: {full_response}",
        user_id=user_id,
        compression_mode="extract"
    )

    return full_response