Agents
Agents are the core building blocks of SciLaxy. Each agent is an LLM-powered workflow that can use tools, access knowledge bases, and collaborate with other agents.
Agent Types
SciLaxy supports several agent types, each optimized for different tasks:
ReAct Agent
The default agent type. Uses a tool-calling loop where the LLM reasons about the task, selects a tool, observes the result, and repeats until the task is complete.
Best for: general-purpose tasks, question answering, tool-assisted workflows.
Custom Agents
Build your own agent workflows by composing reusable components into a graph. Define nodes, edges, and conditional routing to create specialized multi-step workflows. See Custom Agents for details.
Creating an Agent
- Open the SciLaxy web interface
- Click the + button in the sidebar
- Fill in the agent details:
- Name — A descriptive name for the agent
- Description — What the agent does (used for delegation)
- Model — Which LLM to use (e.g.,
gpt-4o,claude-sonnet-4-20250514) - System Prompt — Instructions that guide the agent's behavior
- Select the tools the agent should have access to
- Optionally attach a knowledge base or skills
Agent Configuration
Key configuration options:
Builtin Agents
SciLaxy ships with several pre-configured agents:
- Default ReAct — General-purpose agent with web search, code execution, and file reading
- Root Agent — Autonomous exploration Agent that can discover content, create findings, and schedule follow-up tasks
- Slides — Presentation generation agent
Agent Scopes
Agents have two visibility scopes:
- System agents — Created by administrators, available to all users. Cannot be edited by regular users.
- User agents — Created by individual users, only visible to the creator. Fully editable.
When a user forks an agent from the Marketplace, a user-scoped copy is created.