subrina AI

Enterprise Ready GenAI

subrina applies your enterprise knowledge seamlessly and accurately in real time.

Developer APIs
Go to the App

Learn more about subrina AI below


subrina AI is the latest breakthrough in GenAI and RAG...

Unlike other chatbot style assistants, subrina is powered by our proprietary state of the art RAG System - using your company documents as a single source of truth.

Designed for the enterprise - subrina AI offers privacy, security and readiness for scale.

subrina can do everything that's expected of an AI Assistant and more. Not only does our dedicated RAG technology improve information accuracy, but subrina uses our own Large Language Model (LLM), which is superior for both comprehension and cognitive reasoning - this ensures subrina can apply and explain your knowledge base precisely and intelligently.

subrina is designed to be akin to a human expert, fully versed in your company's unique processes, knowledge base, and training materials, applying this wealth of information with precision in real-time.

• Access and leverage your entire corporate knowledge repository with unmatched accuracy.

• Generate Tables through natural lanuage or a click of a button.

• Assist in writing copy, or emails, using your company procedures as a reference.

• Dedicated language switcher between English and Japanese.

...and that's just the beginning. subrina AI is the key to unlocking potentials you've yet to imagine.

Knowledge Retrieval database

subrina's knowledge comes from the information you put into it. Making AI trustworthy.

Not only will subrina use your company's knowledge as a single source of truth... but subrina's combined technologies are far more accurate at retrieving and applying it than anything else out there.

Furthermore, a built in citation system ensures that you know where the information is coming from within your company's knowledge base. Never be uncertain about the information you're being provided again!


Simple, easy to understand pricing structure.

subrina AI App
£12 per seat / per month (excl. VAT)
• Unlimited Interactions and Queries
There is no limit on chatting with subrina and getting help using your organisations knowledge.
• Unlimited Knowledge Uploads
Import the whole of your company's knowledge to your account (it is never shared)
• Easy to use Knowledge Base UI and document upload system
Linked with API Knowledge Docs
• Intuitive User management and invite system
• Enterprise Encryption Security and Privacy
Data is never sold, shared or used for training.
Subscribe Now
Developer APIs
Pay As You Go
• subrina Instruct MoE LLM 4K 48B
Trained for logical inference and improved comprehension within a chat interaction format.
Input: £0.008 / 1000 tokens
Output: £0.035 / 1000 tokens
• subrina RAG v1.2
Reduce misinformation with our superior knowledge retrieval service.
Use as a drop-in replacement for DIY RAG - Easy to setup, fully managed, and much more accurate than traditional RAG solutions.
Document Upload & Processing (£0.03 / 1,000 tokens),
Storage (£0.18 per GB / month),
Knowledge Retrieval (£0.0004 per request).
• Enterprise Encryption Security and Privacy
Data is never sold, shared or used for training.
Min £5 Top Up (incl. VAT)


Want to see the numbers of how subrina stacks up to the competition?

LLaMa 2 70B (Meta AI)GPT-3.5 (openAI)Grok-1 (X-AI)Palm 2 (Google)Claude 2 (Anthropic)GPT-4 (openAI)subrina AI (sub-10)
Knowledge Dissemination Accuracy Index*
Factual Accuracy using Knowledge Retrieval
68.5% †
0-shot pretrained
0-shot retrieval
Multiple-choice questions in 57 subjects (professional & academic)
5-shot + CoT
0-shot CoT@1
Resistance to coerced misinformation.
Adversarial MCQ
(RLHF Fine Tuned)
- - 69%
Adversarial MCQ
(RLHF Fine Tuned)
Adversarial MCQ
A-Level Biology Paper - - - - - 74.7%
(Full Capabilities incl. Vision and Analysis)
(Full Capabilities incl. Vision and Compute)

* - The Knowledge Dissemination Accuracy Index (KDAI) is an internally designed bechmark which assesses AI systems for factual accuracy, focusing on four competencies: interpreting potentially awkwardly worded questions, effective and sensitive information retrieval from a knowledge base, formulating factually correct responses based on this information in relation to the question, and identifying when information is insufficient to avoid speculation. The KDAI encompasses 1,500 Wikipedia extracts which are unlabeled, unstructured and uploaded to the knowledge system. There are also 1,500 corresponding questions with little to no context, requiring clear factual answers derived from these extracts during the AI's inference process.

† - The completion of the KDAI test using the RAG system on the gpt-4-1106-preview assistants API was hindered due to financial and quota limitations, with the cost being $5.23 for 31 questions and a subsequent reach of the message limit quota. Additionally, the average response latency was approximately 24 seconds per question (subrina had a response latency of < 4 seconds). Consequently, for comparative analysis, the test was conducted using the gpt-4-1106-preview via the standard chat completions API, utilizing its pretrained knowledge.

- CoT = Chain of Thought Prompt. @1 = greedy sampling - we took subrina's first answer only.

- x-shot = The count of accurate exemplars provided within the context to guide the model in recognizing the necessary pattern for a given test. A 5-shot approach includes five accurate examples, whereas a 0-shot approach does not provide any examples at all, which typically yields significantly less precise outcomes.

- subrina scores 83.61% with the 'moral scenarios' MMLU category omitted.

- TruthfulQA Note: Adversarial MCQ Version is much more difficult for an LLM to achieve a high score in.

Contact us now to learn even more!

Contact us at

We are a British Foundation AI company, offering access to our LLM and Knowledge Tech via the subrina AI App and metered APIs.

Go to the App