AI Agents

The Web Report Server embeds an AI chat panel: users talk to role-based AI agents in their own language.
The agents call real tools against the repository (list data sources, create and execute reports, read logs, edit metadata, etc.) and show their progress steps live in the chat.

The AI agents use your own AI provider: no provider or key is bundled with the product.
With Ollama, a fully local model can be used: no data leaves your server.

AI Providers

Run the Server Manager and select the menu Configuration->Configure AI... to configure the AI Providers, Tools and Agents.

An AI Provider is defined with its Type, its Endpoint, an API Key, a Model name and optional generation parameters (temperature, max tokens).
Four provider types are supported:

  • OpenAI: OpenAI API or any OpenAI-compatible endpoint.
  • Azure: Azure OpenAI deployments.
  • Anthropic: Claude models.
  • Ollama: local models running on your own hardware (offline).

The AI configuration is stored in the /Settings/AI Repository sub-folder.

Several providers can be configured (e.g. a hosted model for the report designers, a local model for the analysts).

The Agents

Each agent is dedicated to a role, with its own system prompt and tool set:

Agent For What it does Sample prompts
Report Designer Report authors Designs and builds reports from natural-language requests: from the metadata model, from a SQL statement or on NoSQL/LINQ sources. Styles the report views. "Create a report showing sales by country with a bar chart", "What columns are in the Orders table?"
Data Analyst End users Answers data questions by running existing reports and exploring data sources read-only. "What were total sales last month?", "Show me the top 10 customers by revenue"
Schedule & Output Power users Configures report delivery and automation: outputs, devices and schedules. "Create an Excel output for this report saved in my Personal folder", "Email this report every Monday at 8am"
System Admin Administrators Handles administration and troubleshooting: event logs, security and activity audit, file management. "Show errors in today's event log", "Which schedules failed this week?"
Data Source Manager Data modelers Manages the data source metadata model: adds tables, columns, enumerated lists and joins, edits or removes metadata, checks which reports are impacted before any change. "Add a Yes/No enumerated list on Orders.Shipped", "Which reports use Customers.Country?"
Check AI Data Source Manager for the details of the metadata edition.

Security

  • The agents act as the logged-in user: they can only see and edit what the user's Security Groups allow (folder rights, personal folder, source folder Edit right for metadata changes).
  • Action tools are gated behind skills that the agent must explicitly load: reads are cheap, writes are deliberate.
  • Destructive operations require an explicit user confirmation, and every metadata write is backed up automatically.
Check Security Groups for the rights configuration.

The Semantic Layer

The quality of the AI answers depends directly on the quality of your metadata model.
Fill the Description properties of your connections, tables, columns, joins and enumerated lists: the agents read them to understand your data model.

Check Descriptions: the semantic layer for more information.

AI Tasks in Reports

AI can also be used inside a report execution with the dedicated Task Templates: AI Agent, AI Summary, AI Modify HTML Table, AI Cancel Output (e.g. adding an executive summary generated from the data at the top of a scheduled report).

Check Tasks & ETL for more information. The samples are available in the \Reports\Samples\07-AI Tasks Repository sub-folder.

Customization

The AI configuration is fully customizable in the /Settings/AI Repository sub-folder:

  • Prompts/*.md: the system prompt of each agent. Edit them to add company-specific rules and vocabulary.
  • Samples/*.md: the sample prompts proposed to the users in the chat panel (one prompt per line).
  • AIConfiguration.xml: the agents, tools and skills definitions (advanced).

The agent names, descriptions and tool progress labels displayed in the chat are translated in the user locale via the repository translations.

Contribute to Seal Report

Give a Star!

You like Seal Report? The simplest contribution takes one click: star the project on GitHub to increase its visibility and help the community grow.

Enhance the documentation and samples

Want to share your experience or to add a new topic ? Something is not clear in the documentation ? Do you have an interesting recipe ?
As the site for the documentation is part of the solution, you are welcome to Pull a Request at GitHub.

Sponsor the project

If you are using Seal Report in a business application, please consider to sponsor the product to ensure its maintenance, quality and independence.
Please check Support and Sponsor.

And more...

You are also welcome to help us in several different areas:
  • Beta testing.
  • Libraries of Data Sources and Reports on standard databases.
  • Additional View templates.
  • New features development.
Please feel free to contact us at Ariacom or use GitHub if you are interested in contributing to Seal Report.