Simba is an open-source, self-hosted customer service assistant built for teams that need full control over the AI answering questions from their own knowledge base, with evaluation and monitoring designed in rather than bolted on.

What it is

Simba is a customer service assistant that answers user questions from your own documents through a retrieval-augmented generation pipeline. It ships as several cooperating pieces: a chat widget delivered through the npm package simba-chat-widget, a FastAPI backend service, a vector store, an LLM, and an ingestion path built on Celery workers with Redis as the task queue. A dashboard runs alongside the API for managing documents, monitoring conversations, and reading performance metrics. The registry lists the project as TypeScript under Apache-2.0, while the README's own instructions show a Python backend (pip install simba-core, simba server) plus a Next.js frontend.

The problem it solves is measurement and lock-in. Most customer service bots are black boxes: the operator cannot tell whether retrieval pulled the right passages, whether the generated answer stayed faithful to them, or how long the whole thing took. Simba is built evaluation-first, so retrieval accuracy, generation quality, and latency are tracked out of the box, and every pipeline component is swappable rather than fixed by a vendor. It is the open, self-hosted replacement for the closed, vendor-locked customer service assistant whose behaviour cannot be inspected or tuned.

Key capabilities

• Built-in evaluation framework covering retrieval metrics (precision, recall, relevance), generation metrics (faithfulness, answer relevancy, latency), and conversation analytics such as user satisfaction and resolution rates.
• Swappable RAG components: vector store from Qdrant, FAISS, or Chroma; embeddings from OpenAI, HuggingFace, or Cohere; LLM from OpenAI, Anthropic, or local models; rerankers from Cohere, ColBERT, or cross-encoders; parsers including Docling, Unstructured, and PyMuPDF.
• Drop-in website integration through npm install simba-chat-widget, which exports a SimbaChat component that can be mounted in an existing React application.
• Modern dashboard reached at http://localhost:3000 for document management, conversation monitoring, and performance analysis.
• Production-oriented runtime behaviour: streaming responses and asynchronous document ingestion handled by Celery workers over Redis.
• Docker deployment for either CPU or NVIDIA GPU targets via DEVICE=cpu make build && make up and DEVICE=cuda make build && make up.
• Claude Code setup commands (/setup --all, /setup --backend, /setup --frontend, /setup --services) that install dependencies and start infrastructure.

Who uses it and how

• Support teams that own a documentation set or knowledge base and want the assistant's answers grounded in that material rather than in generic model knowledge.
• Teams that treat answer quality as a measurable quantity, using retrieval and generation metrics to iterate on chunking, embedding models, and rerankers instead of guessing.
• Self-hosters with GPU hardware, who run the DEVICE=cuda build path to keep the LLM workload on their own machines.
• Developers adding a support widget to an existing website, who install simba-chat-widget and render the chat component in their own front end.
• Claude Code users, who bring the whole stack up with /setup --all, which installs Python, frontend, and npm package dependencies and starts the infrastructure services.

Getting started

The recommended route is Docker: clone the repository, create a .env file containing OPENAI_API_KEY, then run DEVICE=cpu make build && make up (or DEVICE=cuda for NVIDIA hardware) and visit http://localhost:3000. A manual installation path also exists through pip install simba-core followed by simba server and simba front.

How it compares

No comparable product is named in the facts supplied for this listing, so Simba stands alone in this registry across the customer service, evaluation, and RAG topics it carries. Any comparison against paid, hosted customer service platforms would have to be drawn from outside this page's sources.

When to use it — and when not to

A self-hoster must operate a vector store (Qdrant, FAISS, or Chroma), Redis, Celery ingestion workers, the FastAPI service, a frontend, and an LLM dependency — either an external provider key such as OPENAI_API_KEY or a local model — which is a real operational surface, not a single container. Multi-tenant support, webhook integrations, and the advanced analytics dashboard are still open roadmap items, so teams needing per-tenant isolation or outbound event hooks today should not pick it yet, and the mismatch between the registry's TypeScript language tag and the README's Python backend is worth confirming before adoption.

git clone https://github.com/GitHamza0206/simba.git
cd simba

OPENAI_API_KEY=your_openai_api_key

# CPU
DEVICE=cpu make build && make up

# NVIDIA GPU
DEVICE=cuda make build && make up

pip install simba-core

simba server
simba front

/setup --all

/setup --backend    # Python dependencies only
/setup --frontend   # Next.js + simba-chat only
/setup --services   # Start Docker infrastructure only

npm install simba-chat-widget

import { SimbaChat } from 'simba-chat-widget';

function App() {
  return (
    <SimbaChat
      apiUrl="https://your-simba-instance.com"
      theme="light"
    />
  );
}

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│   Your Website  │────▶│   Simba API     │────▶│  Vector Store   │
│  (npm package)  │     │   (FastAPI)     │     │  (Qdrant/FAISS) │
└─────────────────┘     └─────────────────┘     └─────────────────┘
                               │                        ▲
                               │                        │
                               ▼                        │
                        ┌─────────────────┐     ┌───────┴─────────┐
                        │      LLM        │     │     Celery      │
                        │ (OpenAI/Local)  │     │   (Ingestion)   │
                        └─────────────────┘     └─────────────────┘
                                                        │
                                                        ▼
                                                ┌─────────────────┐
                                                │      Redis      │
                                                │  (Task Queue)   │
                                                └─────────────────┘

# CPU
DEVICE=cpu make build && make up

# NVIDIA GPU
DEVICE=cuda make build && make up