Agent Builder Guide
Build agents that actually work
Natural language creation. Precision control. Production-ready automation.
The Agent Builder Philosophy
The Agent Builder is your workspace for crafting intelligent AI assistants that can plan and execute multi-step tasks, connect to your tools like Gmail, Notion, and GitHub, remember and learn from conversations, take real actions on your behalf, and work on a schedule or respond to events. Unlike simple chatbots, agents in Opulent OS Heavy are designed for production-grade autonomy with proper permissions, audit trails, and safety guardrails.
Learn From Lived Experience
Agents improve when they learn from their own interaction histories with grounded rewards, not one‑off prompts. This is the core thesis of “The Era of Experience.” [2]
Built For Production (Not Demos)
The shift is happening now: enterprises are operationalizing agentic workflows and 2025 has been called the “year of AI agents.” Reliability, governance, and observability are table stakes. [3]
Environments Where Work Happens
Reliable outcomes require instrumented workspaces where agents execute, test, and verify changes— moving beyond file I/O to close the loop from plan → action → validation. [1]
Prompt to Build
Describe what you want in natural language and let the system generate a complete agent configuration in seconds.
- ✓ Fast prototyping
- ✓ Non-technical friendly
- ✓ Iterative refinement
- ✓ Best practice defaults
Manual Configuration
Craft every aspect by hand—custom prompts, tool selection, model parameters, and complex workflows.
- ✓ Precision control
- ✓ Token optimization
- ✓ Production tuning
- ✓ Advanced workflows
Six Pillars of Agent Intelligence
Building effective agents is an iterative craft that combines strategic design with tactical implementation. Our Agent Builder provides six core capabilities that work together to create production-ready automation.
1. System Prompt: Defining Personality & Behavior
The system prompt is your agent's foundational instruction set—a combination of job description, personality guide, operating manual, and rule book. It's prepended to every conversation turn, ensuring consistency across long interactions.
Anatomy of an Effective System Prompt
- Identity & Role: Who is the agent and what's their primary purpose?
- Communication Style: Tone, verbosity, format preferences, emoji usage
- Available Tools: When to use each tool and with what parameters
- Standard Procedures: Step-by-step workflows for common tasks
- Constraints: Hard boundaries, approval requirements, safety rules
Token Budget: System prompts cost tokens on every turn. Aim for 200-500 tokens for simple agents, 500-1,500 for standard agents, and 1,500-4,000 for complex agents. Beyond 4,000 tokens, consider moving reference material to the Knowledge Base.
2. Model Selection: Choosing the Right Brain
The AI model is your agent's "brain"—it determines reasoning capability, response style, token costs, and performance on specific tasks. Different models excel at different things.
Cost-Optimized Setup
Primary: Gemini Flash or Claude Haiku for most queries. Fallback to Claude Sonnet only when complexity is detected. Result: 10× cost savings.
Quality-Optimized Setup
Primary: Claude Opus or GPT-4o for all queries. Accept higher costs for consistently excellent results. Ideal for customer-facing agents.
| Task Type | Recommended Models |
|---|---|
| Classification, routing | Gemini Flash, Claude Haiku |
| Summarization, Q&A | Claude Sonnet, GPT-4o |
| Code generation, research | Claude Opus, GPT-4o, Gemini Ultra |
| Hard math, logical proofs | o1, o3, Claude Opus |
3. Default Tools: Built-in Capabilities
Default Tools are core capabilities every agent can access without external integrations. These battle-tested primitives enable agents to move beyond conversation into real action.
Planning & Coordination
- • Task management
- • Workflow orchestration
- • Progress tracking
Development Tools
- • File operations
- • Shell execution
- • Code deployment
Research & Automation
- • Web search
- • Browser automation
- • Screenshot capture
Granular Permissions: Enable only the specific actions your agent needs. For example, allow file reading and editing but disable file deletion for safety. Configure in Configuration → Tool Permissions.
4. Integrations: Connecting External Services
Model Context Protocol (MCP) is an open standard that lets agents plug into Gmail, Notion, GitHub, Slack, and hundreds of other services without custom code. It's like USB for AI—a universal interface that makes integrations plug-and-play.
Popular Integrations
- Productivity: Gmail, Notion, Google Calendar, Slack
- Development: GitHub, GitLab, Linear, Jira
- Content: Google Drive, Dropbox, YouTube
- Data: Google Sheets, Airtable, PostgreSQL
Example: Customer Support
Connect Gmail (receive emails) + Notion (create tasks) + Slack (escalate urgent issues). When support emails arrive, the agent categorizes urgency, creates Notion tasks, alerts the team via Slack, and drafts initial responses.
5. Knowledge Base: Teaching Your Agent
A Knowledge Base is a collection of documents, URLs, and data that your agent can search and reference. Using RAG (Retrieval-Augmented Generation), agents become domain experts grounded in your specific context.
How RAG Works
- Documents are split into overlapping chunks (500-1000 tokens)
- Each chunk is converted to a semantic vector (embedding)
- When a user asks a question, the system finds the most relevant chunks
- Retrieved chunks are added to the agent's context
- Agent synthesizes an answer grounded in your documents with citations
Supported Formats
PDF, DOCX, MD, TXT, code files, CSV, JSON, YAML
Upload Methods
Direct upload, URL scraping, integration sync
Best Practices
Keep focused, remove noise, use structure, update regularly
6. Playbooks: Reusable Workflows
Playbooks are parameterized instruction templates that codify repeatable workflows. Define the structure once, then invoke it many times with different inputs—like functions in programming.
Example: Deploy Service to Environment
Template Variables: service_name, environment, previous_version
Steps:
1. Verify {{service_name}} tests pass in CI
2. Build Docker image for {{service_name}}
3. Deploy to {{environment}} cluster
4. Verify health check returns 200 OK
5. Monitor logs for 5 minutes
Rollback: If fails, revert to {{previous_version}}Common Playbook Uses
- • Weekly sprint reports
- • Code review automation
- • Customer onboarding
- • Incident response
- • Content publishing
Invocation Methods
- • Macro trigger in chat
- • UI form with variables
- • Scheduled automation
- • Event-based trigger
Automation with Triggers
Triggers enable agents to act proactively. Instead of waiting for user input, agents can run on schedules, react to events, respond to webhooks, or monitor conditions.
Schedule Triggers
- • Daily at 9am: Morning briefing
- • Every Friday 5pm: Sprint report
- • First of month: Invoice generation
- • Every 15 minutes: Monitor dashboard
Event Triggers
- • New email: Create task
- • PR opened: Run code review
- • Meeting starts: Post agenda
- • Slack mention: Categorize request
Production Best Practices
Start Simple, Then Expand
- 1. Create agent with basic system prompt
- 2. Enable 1-2 integrations (Gmail + Notion is a great combo)
- 3. Test in interactive chat and refine
- 4. Upload key documents to Knowledge Base
- 5. Create playbooks for repeated tasks
- 6. Set up triggers for automation
Security & Governance
- • Grant minimum necessary permissions (OAuth scopes)
- • Enable approval workflows for high-risk operations
- • Monitor audit logs for all tool invocations
- • Use granular tool permissions (enable read, disable delete)
- • Revoke integration access anytime from Settings
Monitoring & Optimization
- • Track token usage and response quality metrics
- • Test retrieval quality after uploading documents
- • Review trigger execution history and failure rates
- • Adjust model selection based on cost vs quality tradeoffs
- • Iterate on system prompt based on real usage patterns
Real-World Use Cases
Personal Productivity Agent
Integrations: Gmail, Google Calendar, Notion
Knowledge Base: Personal goals, project notes
Playbooks: Morning briefing, email triage
Triggers: Daily at 8am, new email arrival
Result: Automated morning briefings, emails converted to tasks, calendar-aware prioritization
Customer Support Automation
Integrations: Gmail, Notion, Slack
Knowledge Base: Product FAQs, policies
Playbooks: Ticket triage, SLA monitoring
Triggers: New support email, every 30 minutes
Result: 80% of tickets handled automatically, SLA compliance, escalation workflows
Code Review Agent
Integrations: GitHub
Knowledge Base: Code style guide, security best practices
Playbooks: PR review checklist
Triggers: New PR opened
Result: Instant style feedback, security checks, consistent review quality
Project Management Assistant
Integrations: GitHub, Linear, Slack, Google Calendar
Knowledge Base: Sprint planning templates
Playbooks: Sprint report, standup summary
Triggers: Daily 9am, weekly Friday 5pm
Result: Automated standup summaries, sprint reports, blocker tracking
Ready to build your first agent?
Start with the Agent Builder and create intelligent automation that works for you—not the other way around.
Docs
Opulent OS