GTEdgeAI/gt-ai-os-community
3
0
Fork 0
Code Issues Packages Releases Wiki Activity

Add How to build GT AI OS Agents

gtadmin
2026-01-10 03:24:41 +00:00
parent 0833b99fc7
commit ed2da0029c
1 changed files with 260 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

260
How-to-build-GT-AI-OS-Agents.md Normal file

@@ -0,0 +1,260 @@
# Demo Agent Templates
Pre-built AI agents you can import to get started quickly. Use them as-is or as templates for creating your own.
## Table of Contents
- [Available Demo Agents](#available-demo-agents)
- [Prerequisites](#prerequisites)
- [Downloading the Agent Files](#downloading-the-agent-files)
- [Importing Agents](#importing-agents)
- [Using the Agents](#using-the-agents)
- [Web Research Agent](#web-research-agent)
- [System Prompt Reviewer](#system-prompt-reviewer)
- [Document Chat Agent](#document-chat-agent)
- [Creating Your Own Agents](#creating-your-own-agents)
- [Method 1: Create in the App and Export](#method-1-create-in-the-app-and-export)
- [Method 2: Edit a Demo Agent CSV](#method-2-edit-a-demo-agent-csv)
- [CSV Format Reference](#csv-format-reference)
- [Changing Models](#changing-models)
---
## Available Demo Agents
| Agent | Description | Model | Provider |
|-------|-------------|-------|----------|
| **Web Research Agent** | Searches the web and provides well-sourced answers with citations | `groq/compound` | Groq |
| **System Prompt Reviewer** | Analyzes and improves system prompts for AI agents | `moonshotai/kimi-k2-instruct-0905` | Groq |
| **Document Chat Agent** | Chat with your uploaded documents using RAG | `nvidia/llama-3.3-nemotron-super-49b-v1` | NVIDIA |
---
## Prerequisites
Before importing these agents, you need API keys configured for the model providers.
### Required API Keys
| Agent | Required API Key |
|-------|------------------|
| Web Research Agent | Groq API Key |
| System Prompt Reviewer | Groq API Key |
| Document Chat Agent | NVIDIA API Key |
### Setting Up API Keys
Follow **Step 6** in the [Control Panel Guide](Control-Panel-Guide#step-6-add-your-external-ai-inference-provider-api-key) to add your API keys:
- **Groq API Key:** Required for Web Research Agent and System Prompt Reviewer
- **NVIDIA API Key:** Required for Document Chat Agent
After adding API keys, verify models are available by following **Step 7** in the [Control Panel Guide](Control-Panel-Guide#step-7-verify-models-are-available).
---
## Downloading the Agent Files
Download the CSV files for the agents you want to import:
| Agent | Download | Required API Key |
|-------|----------|------------------|
| Web Research Agent | [Download CSV](https://github.com/GT-Edge-AI-Internal/gt-ai-os-community/releases/download/v2.0.33/internet-search-agent.csv) | Groq |
| System Prompt Reviewer | [Download CSV](https://github.com/GT-Edge-AI-Internal/gt-ai-os-community/releases/download/v2.0.33/system-prompt-reviewer.csv) | Groq |
| Document Chat Agent | [Download CSV](https://github.com/GT-Edge-AI-Internal/gt-ai-os-community/releases/download/v2.0.33/document-chat-agent.csv) | NVIDIA |
Click the download link and the file will download automatically.
---
## Importing Agents
Follow the [Importing Agent Configuration](Tenant-App-Guide#importing-agent-configuration) steps in the Tenant App Guide:
1. Open the Tenant App: http://localhost:3002
2. Log in with your credentials
3. Click **Agents** in the left sidebar
4. Click the **Agent Configuration** tab
5. Click the **Import** button
6. Drag and drop the CSV file or click **Choose Files**
7. Click **Import Agent**
The agent will appear in your agents list, ready to use.
**Tip:** You can import multiple CSV files at once.
---
## Using the Agents
### Web Research Agent
**What it does:** Searches the internet and provides comprehensive answers with properly formatted citations and source links.
**How to use:**
1. Go to **Agents** → **Favorite Agents** or **Agent Configuration**
2. Click on the Web Research Agent
3. Ask any question you'd research online
4. The agent will search the web and respond with:
- Confidence level based on source quality
- Inline citations [1], [2], etc.
- Full source list with clickable links
**Example prompts:**
- "What are the latest developments in AI?"
- "Current US economic indicators"
- "Recent climate change research findings"
> **Important:** This agent requires the `groq/compound` model because it has web search capability. Other models cannot search the internet.
---
### System Prompt Reviewer
**What it does:** Analyzes system prompts for AI agents and provides actionable feedback to improve them.
**How to use:**
1. Select the System Prompt Reviewer from your agents list
2. Paste a system prompt you want to improve
3. The agent will analyze it for clarity, structure, specificity, and guardrails
4. You'll receive an improved version with explanations
**Example prompts:**
- "Review this prompt: [paste your prompt]"
- "How can I make my agent more focused?"
- "What makes a good system prompt?"
---
### Document Chat Agent
**What it does:** Answers questions based on documents you've uploaded to datasets using RAG (Retrieval-Augmented Generation).
**Setup required:** Before using this agent, you need to create a dataset and connect it:
1. Create a dataset and upload documents by following [Creating a Dataset](Tenant-App-Guide#creating-a-dataset) and [Uploading Documents](Tenant-App-Guide#uploading-documents-to-a-dataset)
2. Connect the dataset to the agent by following [Connecting a Dataset to an Agent](Tenant-App-Guide#connecting-a-dataset-to-an-agent)
**How to use:**
1. Select the Document Chat Agent
2. Ask questions about your uploaded documents
3. The agent will search your documents and provide answers based on the content
**Example prompts:**
- "Summarize the main points"
- "What does it say about [topic]?"
- "Find all mentions of [keyword]"
---
## Creating Your Own Agents
You can create custom agents in two ways: through the app interface or by editing CSV files.
### Method 1: Create in the App and Export
The easiest way to create agents is through the Tenant App:
1. Follow [Creating Your First Agent](Tenant-App-Guide#creating-your-first-agent) to build your agent
2. Test and refine your agent by chatting with it
3. When satisfied, export it by following [Exporting Agent Configuration](Tenant-App-Guide#exporting-agent-configuration)
4. Share the exported file with colleagues or use it as a backup
This method lets you use the visual interface to configure all settings, then export for sharing or backup.
### Method 2: Edit a Demo Agent CSV
Use a demo agent as a starting template:
1. Download one of the demo agent CSV files above
2. Open it in a spreadsheet app (Excel, Google Sheets) or text editor
3. Modify the fields:
- `name` - Give it a unique name
- `description` - Update the description
- `model` - Change the model if needed (see [Changing Models](#changing-models))
- `prompt_template` - Customize the system prompt
- `easy_prompts` - Add your own suggested prompts (pipe-separated)
4. Save as CSV
5. Import the modified file
### CSV Format Reference
Agent CSV files use this format:
| Column | Required | Description |
|--------|----------|-------------|
| `name` | Yes | Agent name (max 255 characters) |
| `model` | Yes | Model ID (e.g., `groq/compound`, `nvidia/llama-3.3-nemotron-super-49b-v1`) |
| `description` | No | Brief description of the agent |
| `category` | No | Category slug (e.g., `research`, `development`, `productivity`) |
| `temperature` | No | Creativity level 0.0-2.0 (default: 0.7) |
| `max_tokens` | No | Maximum response length (default: 4096) |
| `prompt_template` | No | System prompt instructions |
| `dataset_connection` | No | `all`, `none`, or `selected` (default: `all`) |
| `selected_dataset_ids` | No | Pipe-separated UUIDs if using `selected` |
| `disclaimer` | No | Warning text shown to users (max 500 characters) |
| `easy_prompts` | No | Suggested prompts, pipe-separated (max 10) |
| `visibility` | No | `individual`, `team`, or `organization` |
| `tags` | No | Comma-separated tags |
**Format notes:**
- Use commas to separate columns
- Wrap values containing commas or newlines in double quotes
- Use pipe `|` to separate multiple values in array fields (e.g., `easy_prompts`)
- Escape double quotes inside values by doubling them: `""`
### Changing Models
You can change which model an agent uses by editing the `model` column in the CSV.
**Groq models** (require Groq API key):
- `groq/compound` - **Web search capability** ⚠️
- `groq/compound-mini` - Smaller web search model ⚠️
- `llama-3.3-70b-versatile` - General purpose
- `llama-3.1-8b-instant` - Fast, lightweight
- `moonshotai/kimi-k2-instruct-0905` - Advanced reasoning
**NVIDIA models** (require NVIDIA API key):
- `nvidia/llama-3.3-nemotron-super-49b-v1` - Balanced performance
- `nvidia/llama-3.1-nemotron-ultra-253b-v1` - Maximum capability
- `meta/llama-3.3-70b-instruct` - General purpose
- `deepseek-ai/deepseek-r1` - Reasoning specialist
⚠️ **Web Search:** Only `groq/compound` and `groq/compound-mini` models have web search capability. For agents that need to search the internet, you must use one of these models.
**To see all available models:**
1. Open Control Panel: http://localhost:3001
2. Go to **Models**
3. The **Model ID** column shows the exact values to use in your CSV
For adding new models, see [Adding More Models](Control-Panel-Guide#adding-more-models) in the Control Panel Guide.
---
## Troubleshooting
### "Model not found" error when importing
The model specified in the CSV isn't available. Either:
- Add the required API key (see [Prerequisites](#prerequisites))
- Change the `model` column to a model you have access to
### Agent doesn't respond
- Check that the model's API key is configured in [Control Panel → API Keys](Control-Panel-Guide#step-6-add-your-external-ai-inference-provider-api-key)
- Verify the API key is valid and not expired
- Check the model is assigned to your tenant in Control Panel → Tenant Access
### Web search not working
Web search only works with Groq Compound models. Verify:
- You have a Groq API key configured
- The agent is using `groq/compound` or `groq/compound-mini`
### Document Chat Agent doesn't find my documents
- Create and upload documents to a dataset first (see [Tenant App Guide](Tenant-App-Guide#creating-a-dataset))
- Connect the dataset to the agent (see [Tenant App Guide](Tenant-App-Guide#connecting-a-dataset-to-an-agent))
- Verify document processing status is "Completed"
For more help, see the [Troubleshooting](Troubleshooting) guide.