Agent Development Tutorial
Build sophisticated AI agents using the Fiberwise platform. Learn to create content generation agents, integrate LLM providers, and deploy intelligent automation workflows.
⏱️ 45 minutes
📚 Intermediate
🤖 Agent Development
📋 What You'll Build
A content generation agent that:
- ✅ Creates blog posts, social media content, and emails
- ✅ Integrates with multiple LLM providers (OpenAI, Anthropic)
- ✅ Supports different tones and content lengths
- ✅ Includes workflow automation and error handling
Final Result: A production-ready AI agent installed to your Fiberwise platform (no UI - pure agent functionality).
📋 Prerequisites: Your Setup Checklist
Before you begin, you need a fully configured Fiberwise environment. This is the foundation for building any app on the platform.
🔧 Required Setup
✅ All Set?
Once all boxes are checked, you are ready to proceed. If not, please complete the linked guides first.
Step 1: Create Your Agent App
Initialize Agent Project
# Create and enter project directory
mkdir content-generator-agent
cd content-generator-agent
# Initialize basic app structure
fiber init
# The CLI creates:
# ├── app_manifest.yaml # App configuration
# ├── agents/ # Agent implementations
# │ └── content_generator.py
# └── index.js # App entry pointCreate Content Generator Agent
Edit agents/content_generator.py:
from fiberwise_sdk import FiberAgent
class ContentGenerator(FiberAgent):
"""Generates marketing content with AI assistance"""
def run_agent(self, input_data: dict, llm_provider: LLMProviderService):
topic = input_data.get('topic', '')
content_type = input_data.get('type', 'blog_post')
tone = input_data.get('tone', 'professional')
length = input_data.get('length', 'medium')
if not topic:
return {'status': 'error', 'message': 'Topic is required'}
# Generate content based on type
if content_type == 'blog_post':
return self._generate_blog_post(topic, tone, length, llm_provider)
elif content_type == 'social_media':
return self._generate_social_media(topic, tone, llm_provider)
elif content_type == 'email':
return self._generate_email(topic, tone, llm_provider)
else:
return {'status': 'error', 'message': f'Unknown content type: {content_type}'}
def _generate_blog_post(self, topic, tone, length, llm_provider):
"""Generate a blog post"""
length_guide = {
'short': 'about 300-500 words',
'medium': 'about 800-1200 words',
'long': 'about 1500-2000 words'
}
prompt = f"""
Write a {tone} blog post about {topic}.
Length: {length_guide.get(length, 'medium length')}
Include:
- Engaging title
- Introduction that hooks the reader
- 3-4 main sections with subheadings
- Conclusion with call-to-action
- SEO-friendly structure
Format as markdown.
"""
try:
content = llm_provider.complete(prompt)
return {
'status': 'success',
'content': content,
'content_type': 'blog_post',
'metadata': {
'topic': topic,
'tone': tone,
'length': length,
'word_count': len(content.split()),
'generated_at': self.get_current_time()
}
}
except Exception as e:
return {'status': 'error', 'message': f'Generation failed: {str(e)}'}
def _generate_social_media(self, topic, tone, llm_provider):
"""Generate social media posts"""
prompt = f"""
Create 3 different social media posts about {topic} in a {tone} tone.
Format:
1. Twitter/X post (280 characters max)
2. LinkedIn post (professional, 1-2 paragraphs)
3. Instagram caption (engaging, with hashtags)
Each should be engaging and platform-appropriate.
"""
try:
content = llm_provider.complete(prompt)
return {
'status': 'success',
'content': content,
'content_type': 'social_media',
'metadata': {
'topic': topic,
'tone': tone,
'platforms': ['twitter', 'linkedin', 'instagram'],
'generated_at': self.get_current_time()
}
}
except Exception as e:
return {'status': 'error', 'message': f'Generation failed: {str(e)}'}
def _generate_email(self, topic, tone, llm_provider):
"""Generate marketing email"""
prompt = f"""
Write a marketing email about {topic} in a {tone} tone.
Include:
- Compelling subject line
- Personal greeting
- Clear value proposition
- Call-to-action
- Professional signature placeholder
Keep it concise and actionable.
"""
try:
content = llm_provider.complete(prompt)
return {
'status': 'success',
'content': content,
'content_type': 'email',
'metadata': {
'topic': topic,
'tone': tone,
'generated_at': self.get_current_time()
}
}
except Exception as e:
return {'status': 'error', 'message': f'Generation failed: {str(e)}'}
def get_current_time(self):
"""Get current timestamp"""
from datetime import datetime
return datetime.now().isoformat()🧪 Step 2: Test Your Agent Locally
Test Basic Content Generation
# Test blog post generation
fiber activate agents/content_generator.py --input-data '{
"topic": "AI automation for small businesses",
"type": "blog_post",
"tone": "friendly",
"length": "medium"
}'
# Expected output:
# ✅ Agent activated successfully
# ✅ Content generated: ~800 word blog post
# ✅ Metadata includes generation timestampTest Different Content Types
# Generate social media posts
fiber activate agents/content_generator.py --input-data '{
"topic": "productivity tips for remote work",
"type": "social_media",
"tone": "casual"
}'
# Generate marketing email
fiber activate agents/content_generator.py --input-data '{
"topic": "new product launch",
"type": "email",
"tone": "professional"
}'Expected Results
- ✅ Successful execution - No Python errors or exceptions
- ✅ Generated content - Content appears in response JSON
- ✅ Proper formatting - Content follows requested tone and structure
- ✅ Metadata tracking - Response includes generation timestamp and provider info
🚀 Step 3: Install Your Agent
Configure App Manifest
# Content Generator Agent App
app:
name: Content Generator
app_slug: content-generator-agent
version: 1.0.0
description: AI-powered content generation agent
icon: fas fa-pen-fancy
category: ai-tools
minPlatformVersion: 1.6.0
# Agent definitions
agents:
- name: content_generator
description: Generates marketing content with AI assistance
agent_type: content-generation
system_prompt: "You are a professional content creator..."
capabilities: ["blog_posts", "social_media", "email_marketing"]Install to Platform
# Install your agent app
fiber install app .
# The CLI will:
# 1. ✅ Validate app_manifest.yaml
# 2. ✅ Create agent in platform database
# 3. ✅ Upload agent implementation
# 4. ✅ Make agent available for useVerify Installation
# List installed agents
fiber agent list
# Should show: content_generator - active
# Test installed agent
fiber agent activate content_generator --input-data '{
"topic": "AI in marketing",
"type": "blog_post",
"tone": "professional"
}'
# Should generate content using installed agent📊 Step 4: Explore Your Agent's Data
Data Explorer API Integration Flow
graph TD
A[🔧 app-data-explorer.js] --> B[🌐 API Calls to Platform]
B --> C[📍 GET /api/apps/:appId/:modelSlug]
C --> D[📊 Fetch Activations Data]
B --> E[📍 GET /api/apps/:appId/models]
E --> F[📋 Get Available Data Models]
D --> G[📄 Parse JSON Responses]
F --> G
G --> H[🎨 Render Data Tables]
H --> I[📱 Interactive UI Components]
I --> J[🔍 Filter by Content Type]
I --> K[📅 Sort by Date]
I --> L[👁️ View Full Content]
I --> M[📁 Export to CSV]
classDef component fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
classDef api fill:#e8f5e8,stroke:#388e3c,stroke-width:2px
classDef data fill:#fff3e0,stroke:#f57c00,stroke-width:2px
classDef ui fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
classDef features fill:#e0f2f1,stroke:#00796b,stroke-width:2px
class A component
class B,C,E api
class D,F,G data
class H,I ui
class J,K,L,M features
Enhanced Data Explorer & Dynamic Model Flow
graph TD
AA[🤖 Agent Activation] --> AB[🆕 Create Dynamic Model (modelSlug)]
AB --> AC[🗄️ app_models Table]
AB --> AD[🗄️ app_model_fields Table]
AA --> AE[⚡ Create Activation Record]
AE --> AF[🗄️ app_model_items Table]
AF --> AG[📄 data: {field1, field2, ...}]
AH[🔧 app-data-explorer.js] --> AI[🌐 GET /api/apps/:appId/models]
AI --> AJ[📋 List Models (with modelSlug)]
AH --> AK[🌐 GET /api/apps/:appId/:modelSlug/fields]
AK --> AL[📋 List Model Fields]
AH --> AM[🌐 GET /api/apps/:appId/:modelSlug]
AM --> AN[📊 Fetch Model Data]
AN --> AO[📄 Parse JSON Records]
AL --> AO
AO --> AP[🎨 Render Table: Dynamic Fields]
AP --> AQ[📱 Interactive UI: Filter, Sort, View]
classDef agent fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
classDef db fill:#e8f5e8,stroke:#388e3c,stroke-width:2px
classDef api fill:#fff3e0,stroke:#f57c00,stroke-width:2px
classDef data fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
classDef ui fill:#e0f2f1,stroke:#00796b,stroke-width:2px
class AA,AE agent
class AB,AC,AD,AF db
class AH,AI,AK,AM api
class AJ,AL,AN,AO data
class AP,AQ ui
Agent Activation Data Structure Flow
graph TD
A[🤖 Content Generator Agent] --> B[📊 Creates Activation Records]
B --> C[🗄️ app_model_items Table]
C --> D[📝 item_id: UUID]
C --> E[📅 created_at: Timestamp]
C --> F[👤 created_by: User ID]
C --> G[📄 data: JSON Content]
G --> H[🎯 status: success/error]
G --> I[📝 content_type: blog_post/social_media/email]
G --> J[🎨 topic: User Input]
G --> K[🎭 tone: professional/casual/friendly]
G --> L[📏 word_count: Generated Count]
G --> M[📄 content: Generated Text]
H --> N[🔍 Data Explorer Display]
I --> N
J --> N
K --> N
L --> N
M --> N
classDef agent fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
classDef database fill:#e8f5e8,stroke:#388e3c,stroke-width:2px
classDef metadata fill:#fff3e0,stroke:#f57c00,stroke-width:2px
classDef content fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
classDef display fill:#e0f2f1,stroke:#00796b,stroke-width:2px
class A agent
class B,C database
class D,E,F,G metadata
class H,I,J,K,L,M content
class N display
Start Your Platform
Let's see the data your agent has been creating by starting the Fiberwise platform and using the built-in data explorer:
# Start your Fiberwise platform
fiber start
# Platform will be available at:
# http://localhost:8000Access the Data Explorer
Navigate to your app's data explorer to see all the content your agent has generated:
- Open your browser: Go to
http://localhost:8000 - Navigate to Apps: Click on "Apps" in the main navigation
- Find your agent app: Look for "Content Generator" in the apps list
- Open Data Explorer: Click "Data" next to your app name
📊 What You'll See
The data explorer will show you:
- Activations Model - All agent executions and their results
- Generated Content - Blog posts, social media content, emails
- Metadata - Topics, tones, timestamps, content types
- Performance Data - Success rates, generation times
Explore Your Agent's Output
In the data explorer, you can:
🔍 Browse Generated Content
View all blog posts, social media content, and emails your agent has created
📈 Analyze Performance
See success rates, response times, and content quality metrics
🗂️ Filter by Type
Sort content by type (blog_post, social_media, email) or tone
📱 View Details
Click "View" on any item to see full content and metadata
Export Your Data
Download your agent's generated content for analysis or backup:
Data Export & CLI Integration Flow
graph TD
A[💻 fiber data export Command] --> B[🔍 Identify App & Model]
B --> C[🌐 GET /api/apps/:appId/:modelSlug]
C --> D[📊 Fetch All Activation Records]
D --> E{📄 Export Format?}
E -->|csv| F[📊 Convert to CSV Format]
E -->|json| G[📱 Convert to JSON Format]
F --> H[📝 Create CSV Headers]
H --> I[🔧 Map Data Fields]
I --> J[💾 Write CSV File]
G --> K[🔧 Structure JSON Data]
K --> L[💾 Write JSON File]
J --> M[✅ Export Complete]
L --> M
classDef cli fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
classDef api fill:#e8f5e8,stroke:#388e3c,stroke-width:2px
classDef format fill:#fff3e0,stroke:#f57c00,stroke-width:2px
classDef csv fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
classDef json fill:#e0f2f1,stroke:#00796b,stroke-width:2px
classDef complete fill:#ffebee,stroke:#d32f2f,stroke-width:2px
class A,B cli
class C,D api
class E format
class F,H,I,J csv
class G,K,L json
class M complete
CSV Export Data Transformation Flow
graph TD
A[🗄️ Raw Database Records] --> B[📊 JSON Data Extraction]
B --> C[🔧 Field Mapping Process]
C --> D[🆔 activation_id ← item_id]
C --> E[📊 status ← data.status]
C --> F[🎯 content_type ← data.content_type]
C --> G[📝 topic ← data.metadata.topic]
C --> H[🎭 tone ← data.metadata.tone]
C --> I[📏 word_count ← data.metadata.word_count]
C --> J[📅 generated_at ← data.metadata.generated_at]
C --> K[📄 content ← data.content (truncated)]
D --> L[📊 CSV Row Generation]
E --> L
F --> L
G --> L
H --> L
I --> L
J --> L
K --> L
L --> M[💾 Final CSV File]
classDef database fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
classDef extraction fill:#e8f5e8,stroke:#388e3c,stroke-width:2px
classDef mapping fill:#fff3e0,stroke:#f57c00,stroke-width:2px
classDef csv fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
classDef output fill:#e0f2f1,stroke:#00796b,stroke-width:2px
class A database
class B extraction
class C,D,E,F,G,H,I,J,K mapping
class L csv
class M output
# Export activations data to CSV
fiber data export content-generator-agent activations --format csv --output agent-content.csv
# Export with specific date range
fiber data export content-generator-agent activations \
--format csv \
--start-date "2025-01-01" \
--end-date "2025-12-31" \
--output agent-content-2025.csv
# Export as JSON for programmatic use
fiber data export content-generator-agent activations --format json --output agent-data.jsonCSV Structure
Your exported CSV will contain columns like:
| activation_id | status | content_type | topic | tone | word_count | generated_at | content |
|---|---|---|---|---|---|---|---|
| act_123456 | success | blog_post | AI automation | professional | 842 | 2025-08-15T10:30:00Z | # AI Automation for Modern... |
| act_123457 | success | social_media | productivity tips | casual | 156 | 2025-08-15T11:15:00Z | 🚀 Boost your productivity... |
💡 Use Your Exported Data For:
- Content Analysis - Analyze word counts, topics, and performance
- Quality Review - Review and edit generated content
- Publishing Workflows - Import into CMS or social media schedulers
- Training Data - Use successful outputs to improve prompts
- Reporting - Show content generation metrics to stakeholders
🚀 Next Steps
Complete Agent Development & Deployment Lifecycle
graph TD
A[💻 Local Development] --> B[🧪 Local Testing]
B --> C[🚀 Platform Installation]
C --> D[🌐 Production Usage]
D --> E[📊 Data Analysis]
E --> F[🔄 Iteration & Improvement]
F --> A
A --> A1[🏗️ fiber init]
A --> A2[🤖 Agent Implementation]
A --> A3[⚙️ Manifest Configuration]
B --> B1[⚡ fiber activate]
B --> B2[🔍 Validation & Debugging]
B --> B3[✅ Local Success]
C --> C1[📦 fiber install app]
C --> C2[🗄️ Database Integration]
C --> C3[🌐 API Endpoints]
D --> D1[📱 Web Interface]
D --> D2[🔧 FIBER.agents.activate]
D --> D3[📊 Real-time Results]
E --> E1[🔍 app-data-explorer]
E --> E2[📁 CSV Export]
E --> E3[📈 Performance Analysis]
F --> F1[🔧 Code Updates]
F --> F2[📦 fiber app update]
F --> F3[🚀 Re-deployment]
classDef phase fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
classDef dev fill:#e8f5e8,stroke:#388e3c,stroke-width:2px
classDef test fill:#fff3e0,stroke:#f57c00,stroke-width:2px
classDef deploy fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
classDef usage fill:#e0f2f1,stroke:#00796b,stroke-width:2px
classDef analysis fill:#ffebee,stroke:#d32f2f,stroke-width:2px
classDef iterate fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
class A,B,C,D,E,F phase
class A1,A2,A3 dev
class B1,B2,B3 test
class C1,C2,C3 deploy
class D1,D2,D3 usage
class E1,E2,E3 analysis
class F1,F2,F3 iterate
End-to-End Agent Execution Flow with FIBER SDK
graph TD
A[👤 User Request] --> B[🌐 Web Interface/CLI]
B --> C[🔧 FIBER.agents.activate]
C --> D[🌐 POST /api/v1/activations]
D --> E[🗄️ Create Activation Record]
E --> F[🤖 Load Agent Code]
F --> G[💉 Dependency Injection]
G --> H[🧠 LLMProviderService]
G --> I[📊 FiberApp SDK]
G --> J[🗄️ DatabaseService]
H --> K[🎯 Agent.run_agent Method]
I --> K
J --> K
K --> L[🧠 llm_provider.complete]
L --> M[📄 Generated Content]
M --> N[📊 Create Metadata]
N --> O[💾 Update Activation]
O --> P[⚡ WebSocket Event]
P --> Q[🌐 Real-time UI Update]
O --> R[🔍 Data Explorer View]
R --> S[📁 CSV Export]
classDef user fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
classDef interface fill:#e8f5e8,stroke:#388e3c,stroke-width:2px
classDef api fill:#fff3e0,stroke:#f57c00,stroke-width:2px
classDef platform fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
classDef agent fill:#e0f2f1,stroke:#00796b,stroke-width:2px
classDef llm fill:#ffebee,stroke:#d32f2f,stroke-width:2px
classDef result fill:#f1f8e9,stroke:#558b2f,stroke-width:2px
class A user
class B,Q interface
class C,D api
class E,F,G,H,I,J platform
class K,M,N agent
class L llm
class O,P,R,S result
Immediate Next Steps
- Experiment with prompts - Try different content types and tones
- Add error logging - Implement comprehensive error tracking
- Create agent variations - Build specialized content agents
- Update your agent - Make changes and redeploy:
# Update deployed agent fiber app update .
Continue Learning
🏆 Congratulations!
You've successfully:
- ✅ Built an AI-powered content generation agent
- ✅ Integrated with LLM providers securely
- ✅ Implemented error handling and metadata tracking
- ✅ Installed a production-ready agent (no UI needed)
- ✅ Learned agent development best practices
You're now ready to build even more powerful AI agents!