Docs
Powered by

AI Concierge – Technical Overview

Technical documentation for AI Concierge, a privacy-first, in-browser conversational AI platform.

1. Overview

AI Concierge is a domain-specific conversational AI agent designed for secure deployment within trusted enterprise environments. It supports both browser-based and on-premises use cases and is tailored for organizational functions such as HR helpdesks, IT support, and compliance Q&A.

The platform operates on enterprise-specific data and runs open-source large language models, including:

  • Llama 4 Maverick: Customer service, general Q&A, multilingual agents
  • Deepseek R1 70B: Developer support, code generation, documentation bots
  • Llama 3.3 70B: Legal, HR, writing-intensive tasks, or any scenario needing high linguistic precision

AI Concierge is ideal for teams building AI copilots, automating tasks, or integrating LLM-powered tools into business operations.

2. Data Privacy & Security

AI Concierge is engineered with a privacy-first architecture:

  • User interactions are not stored, logged, or used for model retraining.
  • Session data remains local—either in the user's browser or customer-controlled infrastructure.
  • No telemetry, analytics, or backend monitoring is included by default.
  • Internal administrators cannot view user chat history or prompt content.

This closed-loop model supports use in regulated industries like healthcare, finance, and government.

3. Key Features

🔒 Privacy-First Design

  • No data is stored on central servers or used for retraining.
  • Conversations are processed in-session and stored locally.
  • Enables safe use in regulated industries.

🏢 Enterprise-Scoped AI

  • Tailored for internal functions like HR, IT, and compliance.
  • Deployed via UI or API within your trusted runtime.
  • Grounded in your proprietary enterprise data.

💬 Persistent and Seamless UX

  • Chat history is saved in-browser, not in the cloud.
  • Sessions persist until explicitly cleared.
  • Revisit previous prompts via a built-in history log.

🚀 Lightweight, Flexible Deployment

  • Runs in-browser with no infrastructure required.
  • Also supports hybrid or on-premise setups.
  • Ideal for pilot projects and internal copilots.

🔧 Open-Source Model Backbone

  • Powered by Llama 3.3 and DeepSeek-R1.
  • Models are customizable to your use case and data.
  • Future support planned for model switching in-session.

🧪 Free Trial Access

  • Includes:
    • Serverless Endpoints for model serving
    • Model Execution & Indexing Units (MEIUs) including Retrieval-Augmented Generation (RAG)
  • Best for teams with predictable usage patterns.

4. Accessing the Platform

  1. Create or receive a user account.

Tip: If you need help, visit our documentation to create an account or receive a user account.

  1. Select an access tier:
    • Trial version
    • Production deployment
  2. Log in to the platform:
    • Trial – DeepSeek R1
    • Production – DeepSeek R1

After login, the chat interface will load automatically.

5. User Interface Overview

UI Element Description
Chat History View and revisit past prompts. See: Chat History
Clear Window Clears the current window while retaining session memory.
Start New Conversation Resets chat context and removes any attached PDFs.
Document Upload Upload up to 3 PDFs (max 1MB each) for RAG-based context. See: Attach File
Generate Button Submits the current prompt to the selected model.
Save Conversation Stores the current chat session locally in the browser.
Download PDF Exports the conversation transcript as a PDF file.
Copy to Clipboard Copies the most recent prompt and response to your clipboard.
Account Menu Manage team access, user profile, and terms/policies.

6. Supported Browsers

  • Google Chrome / Microsoft Edge
  • Mozilla Firefox
  • Apple Safari

7. Tips for Best Results

  • Upload .PDFs to enable Mini-RAG-style context injection.
  • Maximum: 3 files per session, each under 1MB.
  • Hover over a file to remove it (click the “X”).
  • For better outputs:
    • Be specific (e.g., “Explain how AI is used in radiology”).
    • Break complex questions into steps.
    • Add background/context when necessary.

Refer to the Meta Llama 3 Prompt Format Guide for advanced prompt structuring.

8. Security & Data Protection

  • All chat history is stored in your browser only.
  • Uploaded documents are deleted when history is cleared.
  • For extra data protection:
    • Enable disk encryption (e.g., BitLocker for Windows, FileVault for macOS).
    • Always log out on shared or public systems.

9. Troubleshooting

Issue Recommended Action
UI is slow or unresponsive Refresh the page or check the Sunrise Status Page.
File won’t upload Ensure it’s a .pdf file under 1MB; maximum of 3 files per session.
Poor response quality Rephrase your prompt; try using specific or multi-step questions.
Session reset unexpectedly Confirm browser storage is enabled; avoid private/incognito mode.

For questions or support, contact your system administrator or the technical support team.

© Copyright 2026 Sunrise. All rights reserved.