AI Agents

What Are AI Agents?

AI Agents are specialized, customizable AI assistants built on top of Nova that you can train with your own domain-specific knowledge. Unlike Nova’s general-purpose capabilities, each agent becomes an expert in your specific area by learning from the documents, papers, and data you provide.

Think of it this way:

• Nova: A general-purpose AI interface ready to help with any task
• AI Agents: Domain experts you create—a LAMMPS specialist, a genomics pipeline expert, or a quantum chemistry advisor—each trained on your specific documentation

Prerequisites: Active CamberCloud account with platform access. No coding experience required, though familiarity with your research domain is helpful.

Core Capabilities

AI Agents combine Nova’s computational power with your domain expertise through three key capabilities:

1. Custom Knowledge Bases

Build agents that understand your research context:

• Document ingestion: Upload PDFs, papers, manuals, and technical documentation
• Multi-source integration: Connect to URLs, Stash files, Google Drive, SharePoint, and Amazon S3
• Automatic indexing: Retrieval-Augmented Generation technology processes and indexes your documents
• Contextual retrieval: Agents automatically find and reference relevant sections from your knowledge base

2. Specialized Instructions & Behavior

Configure how your agent thinks and responds:

• Custom system prompts: Define your agent’s role, expertise, and communication style
• Workflow guidance: Embed standard operating procedures and best practices
• Domain-specific reasoning: Train agents to approach problems using your field’s methodologies
• Team-shared expertise: Create consistent expert assistants accessible to your entire research group

3. Seamless Nova Integration

Access your agents directly within Nova conversations:

• @mention invocation: Call agents using @username.agent_alias syntax
• Contextual responses: Agents automatically retrieve relevant knowledge while answering
• Transparent citations: See which documents informed each response
• Full platform access: Agents inherit Nova’s capabilities for job management, file operations, and app launching

How to Use AI Agents

Step 1: Create Your Agent

1️⃣ Navigate to the AI Agents section in the platform sidebar 2️⃣ Click “Create Agent” to open the configuration dialog 3️⃣ Define your agent’s identity:

• Agent Name: Human-readable name like “LAMMPS Simulation Expert”
• Agent Alias: Short handle for @mentions like “lammps_agent”
• Description: Explain what the agent does and when to use it
• Instructions: Write the system prompt that defines your agent’s expertise and behavior. Keep in mind that longer prompts do not necessarily improve performance, and it’s best to keep the prompt length to several thousand characters.

Writing Effective Instructions

Your agent’s system prompt defines its expertise and behavior. Use this template:

You are a [ROLE] for [CONTEXT].

Your expertise includes:
- [Domain area 1]
- [Domain area 2]
- [Specific tools or methods]

When answering questions:
- [Behavior guideline 1]
- [Behavior guideline 2]
- [When to defer to humans]

You have access to [KNOWLEDGE BASE DESCRIPTION].

You are a protein crystallography expert for the Smith Lab.

Your expertise includes:
- Crystal growth and optimization protocols
- X-ray diffraction data collection
- Structure refinement with Phenix and CCP4
- Our lab's Rigaku MicroMax-007 equipment

When answering questions:
- Reference specific protocols from the knowledge base
- Cite relevant papers when applicable
- Warn about common pitfalls we've documented
- Suggest consulting Dr. Smith for novel situations

You have access to our complete protocol library and methods papers.

You are a data analysis expert for the Marketing Analytics team.

Your expertise includes:
- Customer segmentation analysis
- Time series forecasting
- A/B test interpretation
- Our Snowflake data warehouse schema

When answering questions:
- Use our standard methodology for statistical tests
- Reference our data dictionary for column definitions
- Provide confidence intervals with all estimates
- Flag any data quality concerns

Instruction tips:

• Name specific tools, methods, and terminology
• Specify when the agent should defer to humans
• Define what’s out of scope and include safety guidelines
• Keep it focused—longer prompts aren’t always better

Step 2: Build the Knowledge Base

1️⃣ Select the “Knowledge Base” tab on your agent’s page 2️⃣ Click “Add Resources” to choose your integration method:

Supported file formats: PDF, Word (.docx), Text (.txt), Markdown (.md), Excel (.xlsx), CSV

Monitor indexing status—documents move from “Processing” to “Indexed” when ready. 3️⃣ Deploy the agent.

How it works: When you query an agent, it searches the knowledge base for relevant content, retrieves the most relevant sections, uses that context to formulate its response, and cites sources in its answers.

Knowledge base tips:

• Use well-structured documents with clear headings
• Include context and rationale, not just procedures
• Begin with 5-10 high-quality documents, test responses, then add more based on gaps
• Group related documents together and use descriptive file names

Step 3: Invoke Your Agent

Open a Nova chat, type @ to view available agents, select your agent or continue typing its alias, then ask your question: @username.agent_alias how do I configure a LAMMPS NVT ensemble?

Step 4: Review Knowledge-Enhanced Responses

The agent responds using both its base knowledge and your custom documentation.

Agent Use Cases

Domain-Specific Research Assistants

1. Molecular Dynamics Expert Scenario: Your lab runs frequent LAMMPS simulations but team members have varying expertise levels.

Solution: Create a LAMMPS agent trained on:

• Official LAMMPS documentation
• Your lab’s simulation protocols
• Published papers using your methods
• Troubleshooting guides from past projects

Benefits: New students get expert guidance instantly; consistent methodology across the team.

2. Genomics Pipeline Specialist Scenario: You’ve developed custom bioinformatics workflows with specific tool configurations.

Solution: Build an agent with knowledge of:

• Your pipeline documentation
• Tool-specific parameter explanations
• Quality control thresholds and interpretation
• Common error resolutions

Benefits: Team members can query best practices without interrupting senior researchers.

3. Quantum Chemistry Advisor Scenario: Complex quantum chemistry calculations require choosing between multiple methods and basis sets.

Solution: Create an agent informed by:

• Computational chemistry textbooks
• Benchmark studies for your molecule types
• Lab-specific computational resources and constraints
• Published methodologies from your field

Benefits: Make informed method decisions with guidance grounded in authoritative sources.

Example: Building a Research Assistant

Here’s a walkthrough of creating a research assistant agent from scratch.

Problem: Your lab has accumulated years of protocols, papers, and tribal knowledge. New team members spend weeks getting up to speed, and even experienced researchers forget details of rarely-used procedures.

Solution: Build a Research Assistant agent trained on your lab’s documentation. Team members can ask questions in natural language and get accurate, sourced answers instantly.

1. Gather your knowledge base — Collect lab protocols (PDF, Word, Markdown), published papers from your group, equipment manuals and SOPs, meeting notes with methodological decisions, and troubleshooting guides. Start with 5-10 high-quality documents.

2. Create the agent — Navigate to AI Agents, click Create Agent, and configure:

3. Test your agent:

@myusername.crystal_expert What's our standard protocol
for growing lysozyme crystals? Include the specific
PEG concentrations we use.

User: @lab.crystal_expert How do I prepare the cryoprotectant
for flash-freezing?

Agent: Based on Protocol #CR-007 in our knowledge base,
the standard cryoprotectant preparation is:

1. Start with mother liquor from the crystal drop
2. Add glycerol to 20% (v/v) final concentration
3. Mix gently to avoid precipitation
4. Transfer crystal using a nylon loop

Note: For PEG-based conditions, you may need to increase
glycerol to 25% per Dr. Smith's 2023 optimization study.

User: @lab.crystal_expert My crystals are growing too fast
and have poor diffraction. What should I try?

Agent: According to our troubleshooting guide (Section 4.2),
rapid crystal growth often indicates supersaturation issues.
Recommended adjustments:

1. Reduce protein concentration by 20-30%
2. Lower precipitant concentration by 5%
3. Try seeding with microcrystals from a previous batch

Our 2022 paper on optimization (Smith et al., Acta Cryst)
found that slower growth at 4C improved resolution
from 3.2A to 2.1A for similar cases.

4. Share with your team — Go to your agent’s settings, under Sharing select your team, and team members can invoke: @teamname.crystal_expert

Best Practices

Building Effective Knowledge Bases

1. Start with authoritative sources: Official docs, peer-reviewed papers, textbooks
2. Add lab-specific context: Standard operating procedures and internal protocols
3. Keep it current: Update regularly with new findings

Writing Effective Instructions

Do: Clearly state domain expertise, specify terminology conventions, define when to defer to humans Don’t: Use overly broad instructions or assume unstated lab conventions

Agent Management

• Deploy incrementally: Test with a few documents before adding your entire library
• Share across teams: Grant access to colleagues in the same domain
• Iterate: Refine instructions based on how the agent performs
• Review agent responses periodically and add documents that fill knowledge gaps

Related Documentation

• Nova AI Interface - Your primary way to interact with the Camber platform
• Apps Catalog - Applications that agents can launch and configure
• Stash File System - Store agent knowledge base documents
• Job Attributes Reference - Configure compute resources for agent-launched jobs
• Data Connectors - Integrate external data sources into agent workflows

Getting Started

Ready to build your first specialized research assistant?

1. Identify your domain: What specific expertise do you need to encode?
2. Gather resources: Collect 3-5 key documents to start your knowledge base
3. Create your agent: Define its role and upload your initial documents
4. Test and refine: Try example questions and adjust instructions as needed
5. Share with your team: Grant access and gather feedback

Transform your hard-won expertise into an AI assistant that’s available 24/7. Create your first agent today.

Need help? Join our Slack community or contact support.