AI-Co-Pilot.md

layout	title	nav_order
default	AI Search Assistant Guide	10

Table of Contents

{: .text-delta } - TOC {:toc}

AI Search Assistant Guide
Enterprise Edition

{: .warning } Please contact SWIRL for access to SWIRL Enterprise.

---

Configuring SWIRL AI Search Assistant, Enterprise Edition

Roles for Generative AI / Large Language Models

SWIRL AI Search defines four core roles for GAI/LLMs. SWIRL AI Search Assistant adds a fifth role, "chat," which can be assigned to any sufficiently capable LLM.

Role	Description	Default Provider
reader	Generates embeddings for SWIRL’s Reader LLM to re-rank search results	spaCy
query	Provides query completions for transformations	OpenAI GPT-3.5 Turbo
connector	Answers direct questions (not RAG)	OpenAI GPT-3.5 Turbo
rag	Generates responses using Retrieval-Augmented Generation (RAG) with retrieved data	OpenAI GPT-4
chat	Powers SWIRL AI Search Assistant messaging	OpenAI GPT-4

Adding Chat to an AI Provider

1. Open the AI Providers management page: http://localhost:8000/swirl/aiproviders (default local installation).

   

2. Edit an AI provider by appending its id to the URL, e.g.: http://localhost:8000/swirl/aiproviders/4/

   

3. Ensure the following in the provider’s configuration:

   • active is set to true
   • api_key contains a valid API key
   • model and config values are correctly filled
   • "chat" is included in the tags list
   • "chat" is included in the defaults list

Example: OpenAI GPT-4 Configured for Chat & RAG

{
    "id": 16,
    "name": "OpenAI GPT-4",
    "owner": "admin",
    "shared": true,
    "date_created": "2024-03-04T15:15:16.940393-05:00",
    "date_updated": "2024-03-04T15:15:16.940410-05:00",
    "active": true,
    "api_key": "<your-openai-api-key>",
    "model": "gpt-4",
    "config": {},
    "tags": ["query", "connector", "rag", "chat"],
    "defaults": ["rag", "chat"]
}

Launching Assistant

Once the AI provider is configured correctly, Assistant should be accessible via a browser.

For a default installation, go to: http://localhost:8000/galaxy/chat

For more details, see the AI Search, Enterprise Edition section: Connecting to Enterprise GAI and LLMs.

GAI/LLM Requirements

SWIRL AI Search Assistant expects AI providers to support:

• Chat history in reverse chronological order, following the format used by the OpenAI Chat Completions API.
• Prompt size of at least 3K tokens per message, with 6K+ preferred.
• Recommended models: OpenAI GPT-4 (gpt-4, gpt-4o), Anthropic Claude 3, or Google Gemini Pro/Ultra.

Other LLMs may also work if they support chat history as described above. If you test alternative models, please let us know what works (or doesn’t)!