Build a chatbot with memory

This recipe walks through building a customer-facing chatbot that answers questions from your own indexed documents, with source citations. Full working code in Python.

What you’ll build

A FastAPI endpoint that takes a user message, forwards it to Memic’s chat endpoint, and returns a grounded answer plus citations. Under 40 lines of code.

Prerequisites

Python 3.10+
A Memic environment with some documents already uploaded
An API key for that environment
pip install memic fastapi uvicorn

The code

import os
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from memic import Memic

app = FastAPI()
memic = Memic(api_key=os.environ["MEMIC_API_KEY"])


class ChatRequest(BaseModel):
    messages: list[dict]  # [{"role": "user", "content": "..."}]


class Citation(BaseModel):
    file_name: str
    page: int | None
    passage: str


class ChatResponse(BaseModel):
    answer: str
    citations: list[Citation]


@app.post("/chat", response_model=ChatResponse)
def chat(req: ChatRequest):
    try:
        result = memic.chat(messages=req.messages)
    except Exception as e:
        raise HTTPException(500, f"Memic call failed: {e}")

    return ChatResponse(
        answer=result.answer,
        citations=[
            Citation(
                file_name=c.file_name,
                page=c.page,
                passage=c.passage,
            )
            for c in result.citations
        ],
    )

Run it:

MEMIC_API_KEY=mk_live_... uvicorn main:app --reload

Call it:

curl http://localhost:8000/chat \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "user", "content": "What is our refund policy?"}
    ]
  }'

Adding conversation history

Memic is stateless. To have a real conversation, your frontend or session store needs to keep the message array and pass it on every turn:

# In your session:
session.messages.append({"role": "user", "content": user_input})
result = memic.chat(messages=session.messages)
session.messages.append({"role": "assistant", "content": result.answer})

Rendering citations in a UI

Each citation points back to a specific passage in a specific file. A reasonable UI pattern:

<div>
  <p>{answer}</p>
  <div className="sources">
    {citations.map(c => (
      <a href={`/files/${c.file_id}#page=${c.page}`}>
        📄 {c.file_name} (page {c.page})
      </a>
    ))}
  </div>
</div>

Going further

Multi-tenant: swap the API key per customer — see Per-customer isolation
Custom system prompts: use managed prompts to maintain your assistant’s persona in the dashboard without redeploys
Retrieval-only: if you want full control over the LLM call, use memic.search(...) instead of memic.chat(...) and plug the results into your own OpenAI/Anthropic call

Documentation Index

​What you’ll build

​Prerequisites

​The code

​Adding conversation history

​Rendering citations in a UI

​Going further