Self-hosted Omni-nli MCP Server: Boost NLI, Enhance AI App Reliability

Omni Nli

Omni-NLI is a self-hosted multi-interface (REST and MCP) server that focuses on natural language inference tasks, used to verify the support, contradiction, or neutral relationship between texts, and can help reduce AI hallucination and improve application reliability.

Artificial intelligence chatbots Research and data #Natural language inference #AI verification #Text consistency #Multi-interface service .Python

rating : 2.5 points

downloads : 5.9K

update time : 2026-03-13

Open Site

What is Omni-NLI?

Omni-NLI is a self-hosted server that provides natural language inference (NLI) capabilities. It can determine the logical relationship between two pieces of text: support, contradiction, or neutrality. Its main purpose is to verify whether AI-generated content is consistent with known facts or context, thereby reducing AI hallucination issues.

How to use Omni-NLI?

You can use Omni-NLI in two ways: 1) As an independent REST API microservice, perform text verification through HTTP requests; 2) As an MCP server, allow AI assistants to directly call the verification function. It is easy to install and supports multiple AI model backends.

Applicable scenarios

Suitable for scenarios that require verifying the accuracy of AI outputs: consistency check of chatbot conversations, verification of document summary accuracy, verification of search result relevance, fact-checking systems, verification of AI assistant credibility, etc.

Main features

Dual interface support

It provides both REST API (for traditional applications) and Model Context Protocol (for AI assistants) interfaces, flexibly adapting to different usage scenarios.

Multi-model backend

Supports multiple AI model backends such as Ollama, HuggingFace (public and private models), and OpenRouter. You can choose the most suitable model according to your needs.

Interpretable output

It not only returns the inference result (support/contradiction/neutrality) but also provides a confidence score and an optional inference process, making the verification result more transparent and credible.

High-performance cache

It has a built-in cache mechanism that caches the same or similar text inference requests, significantly improving the response speed and system throughput.

Easy to deploy

It provides Docker images (CPU and CUDA versions), supports pip installation, has simple configuration, and can be quickly deployed in a production environment.

Advantages

Reduce AI hallucination: Provide a fact verification layer for AI applications and improve output reliability.

Flexible integration: Support traditional REST API and the emerging MCP protocol, adapting to different technology stacks.

Model-agnostic: Not bound to a specific AI model and can switch different backends according to needs.

Highly scalable: Stateless design, supports horizontal scaling, and is suitable for high-concurrency scenarios.

Open source and transparent: Under the MIT license, the code is completely open and can be customized and optimized.

Limitations

Dependent on model quality: The inference accuracy is limited by the quality of the selected AI model.

Requires domain adaptation: For professional domains, targeted model fine-tuning may be required.

Early development stage: Currently in an early version, there may be instability.

Resource consumption: Running AI models requires certain computing resources (CPU/GPU).

Semantic understanding limitation: Based on statistical patterns and cannot completely replace human logical reasoning.

How to use

Installation

Install Omni-NLI via pip and choose different backend supports according to your needs.

Start the server

Run the startup command, and the server will start locally and listen on the port.

Configure the model (optional)

Configure the AI model to be used as needed, supporting local models or cloud APIs.

Send a verification request

Send a text verification request through the REST API or MCP interface.

Usage examples

Chatbot consistency check

Verify whether the new response of the AI assistant contradicts the previous conversation content to ensure the consistency of the conversation logic.

Document summary verification

Check whether the document summary generated by AI accurately reflects the original content and avoid distorting the original meaning of the summary.

Fact checking

Verify whether the factual statements generated by AI are consistent with the known fact database.

Search result relevance verification

Check whether the documents returned by the search engine truly answer the user's query.

Frequently Asked Questions

What is the difference between NLI and logical reasoning?

How to improve the inference accuracy?

Which languages are supported?

What is the response time?

How to integrate with existing AI applications?

Related resources

Official documentation

Complete API reference, configuration guide, and advanced usage.

GitHub repository

Source code, issue tracking, and contribution guide.

Example projects

Practical usage examples and demonstration code.

Docker image

Pre-built Docker container image.

PyPI package

Python package installation page.

🚀 Omni-NLI

A multi-interface (REST and MCP) server for natural language inference

🚀 Quick Start

Omni-NLI is a self-hostable server that provides natural language inference (NLI) capabilities via RESTful and the Model Context Protocol (MCP) interfaces. It can be used both as a very scalable standalone stateless microservice (via the REST API) and also as an MCP server for AI agents to implement a verification layer for AI-based applications.

1. Installation

pip install omni-nli[huggingface]

2. Start the Server

omni-nli

3. Evaluate NLI (with REST API)

curl -X POST \
  -H "Content-Type: application/json" \
  -d '{
    "premise": "A football player kicks a ball into the goal.",
    "hypothesis": "The football player is asleep on the field."
  }' \
  http://127.0.0.1:8000/api/v1/nli/evaluate

Example response:

{
    "label": "contradiction",
    "confidence": 0.99,
    "model": "microsoft/Phi-3.5-mini-instruct",
    "backend": "huggingface"
}

4. Evaluate NLI (with MCP Interface)

✨ Features

Helps mitigate LLM hallucinations by verifying if the generated content is supported by facts
Supports models provided by different backends, including Ollama, HuggingFace (public and private/gated models), and OpenRouter
Supports REST API (for traditional applications) and MCP (for AI agents) interfaces
Fully configurable and very scalable, with built-in caching
Provides confidence scores and (optional) reasoning traces for explainability

See ROADMAP.md for the list of implemented and planned features.

⚠️ Important Note Omni-NLI is in early development, so bugs and breaking changes are expected. Please use the issues page to report bugs or request features.