AI

Build AI agents with tools, prompts, and models — all configurable without code deployment.

What Is the Bosca AI System?

The Bosca AI system is a database-driven agent platform built into the Bosca platform. It lets you create, configure, and manage AI agents entirely through the admin UI — no code deployment required. You define what an agent does by combining a model, a prompt, and a set of tools. When you need to change how an agent behaves, you update its configuration in the UI and the changes take effect immediately.

This approach means that product managers, content strategists, and operations teams can create and iterate on AI-powered features without waiting for engineering releases. Engineers set up the infrastructure and tools once; everyone else can build on top of it.

Key Concepts

The AI system is built around four interconnected building blocks. Each one can be configured independently, and they come together to form complete AI experiences.

Concept	What It Does
Agents	The central unit. An agent combines a model, a prompt, and a set of tools to perform a specific task. Agents can also delegate work to sub-agents for complex multi-step workflows.
Prompts	The instructions that define how an agent behaves. A prompt includes a system message (the agent's persona and rules), a user message template, and optional output schemas for structured responses.
Models	Configurations that point to specific LLM providers and model versions. Swap between Google Gemini, OpenAI GPT, Anthropic Claude, Mistral, and others without changing anything else about the agent.
Tools	Capabilities that agents can use during a conversation. Tools include SQL queries against your data warehouse, content search, image generation, script execution, and connections to external services via MCP servers.

How It All Fits Together

Building an AI agent in Bosca follows a natural progression:

Choose a model. Select which LLM provider and version the agent should use. Different models have different strengths — fast and inexpensive for simple tasks, powerful and capable for complex reasoning.
Write a prompt. Define the agent's personality, instructions, and behavioral rules. Specify what kind of input the agent expects and what format its output should take.
Assign tools. Give the agent capabilities beyond conversation — the ability to query databases, search content, generate images, run scripts, or call external APIs through MCP servers.
Configure the agent. Bring the model, prompt, and tools together under a single agent definition with a name and description.
Chat and iterate. Users interact with the agent through real-time streaming chat. Review conversations, adjust the prompt, swap the model, or add tools — all without any code changes.

Everything in the AI system is stored in the database and managed through the admin UI. There is no need to write code, build containers, or deploy applications to create or modify agents. Changes take effect immediately.

Real-Time Streaming Chat

Users interact with agents through a real-time streaming chat interface. Responses appear token by token as the model generates them, providing a responsive and natural conversation experience.

The chat interface goes beyond simple text. Agents can present interactive elements during a conversation:

Confirmation requests — The agent can ask the user to approve an action before proceeding, such as confirming a database query or approving a content change.
Choice selection — The agent can present a set of options for the user to choose from, guiding the conversation in a structured way.
Visualizations — Agents can render charts, tables, and other visual elements directly in the chat to present data and analysis results.

Hierarchical Agent Composition

Agents can be composed hierarchically. A top-level agent can delegate subtasks to specialized sub-agents, each with their own model, prompt, and tools. This lets you build complex workflows from simple, focused components.

For example, a "Research Assistant" agent might delegate to a "Search Agent" that finds relevant content, a "Data Analysis Agent" that runs queries and produces charts, and a "Summarization Agent" that distills findings into a concise report. Each sub-agent is independently configurable, testable, and reusable across different parent agents.

Built-In Content Agents

The platform ships with several pre-configured agents that integrate directly with the content management system:

Description generation — Automatically generate descriptions for content items based on their body text.
Topic extraction — Identify and tag the key topics covered in a piece of content.
Reading time estimation — Calculate how long it will take a reader to consume a piece of content.
Content generation — Generate full content drafts from specifications, outlines, or briefs.

These built-in agents serve as practical examples of how to configure agents for your own use cases. You can inspect their prompts, models, and tool assignments to learn how they work, then create your own agents following the same patterns.

Explore AI

Agents — Define AI agents that combine models, prompts, and tools for specific tasks
Prompts — Define agent behavior with system prompts, user templates, and output schemas
Models — Configure which AI models your agents use and switch providers without changing code
MCP Servers — Connect agents to external tools and data sources via the Model Context Protocol