Journalling
Dashboard

🧠 Story.AI & Workflow Agent: Multi-Agent Mental Health Support System

✨ Inspiration

Mental health solutions are often expensive, inaccessible, and generalized. Story.AI was inspired by the idea of decentralizing and democratizing mental wellness through a personalized, always-available network of intelligent AI agents. Powered by the Fetch.ai Agentverse ecosystem, our goal is to make mental health support scalable, intelligent, and deeply personalized—without replacing human therapists, but amplifying their impact.

🧠 Story.AI: Multi-Agent Mental Health Support System

Story.AI is a comprehensive mental health support platform powered by a network of specialized AI agents, designed for emotional understanding, behavioral transformation, and cognitive restructuring through journaling, reflection, and conversation. Built entirely on Fetch.ai's Agentverse, it’s designed as a modular, composable ecosystem of six collaborative agents.

🔁 Multi-Agent Architecture

Agent Name	Role	Address
Personal Assistant Agent	Central orchestrator that analyzes user queries and routes them to specialized agents	`test-agent://agent1qt7fcalfq80vevam23gz8hlrl9txh6wrcpwn57j670p4sdkdd83j5sgq7ge`
Journal Analysis Agent	Detects emotions and cognitive patterns in journal entries	`test-agent://agent1qf97yx9eg6r7ycgt68r7a90dfp4qraf8jldwg8urktr204l0futnkyckt5f`
Exercise Generator Agent	Generates personalized CBT exercises and morning reflections	`test-agent://agent1qtpcdp4gte33ey026qd3pguuadwf89e9dfwl4ezazq9y7ajvv3z6y9h3662`
Gratitude Agent	Builds gratitude-based practices based on user situations	`test-agent://agent1qdxlnp47vjqn2dkn5tg5l62apd4x7gn4daygl8jlcfc4jk7c3ca4jm83ukx`
Therapy Conversation Agent	Conducts memory-aware therapy-like conversations	`test-agent://agent1qftlfn8w5dskw9rd97kg4rt9ul0dynt8y97d99mcfzxkdxcadmnxyjnhrc3`
Story.AI Guide Agent	Assists in navigation and feature discovery within Agentverse	`test-agent://agent1q0sm7nt0w330v9cutvkqg6dvjmgghejwvmrejydlu2dhq4e0s8l6xnwyzds`

Each agent is designed with domain-specific prompts, structured payloads, and secure asynchronous webhooks for inter-agent communication.

🔧 How We Built Story.AI

⚙ Agent Architecture

Assistant Agent: Understands user intent and routes queries.
Journal Agent: Uses emotion models (e.g. DistilRoBERTa) to extract cognitive patterns.
Exercise Generator: Maps journal insights to mental health exercises using LLM prompting.
Gratitude Agent: Crafts personalized gratitude prompts with situational context.
Therapy Agent: Uses LangChain’s ConversationBufferMemory for therapy-like continuity.
Guide Agent: Enables users to explore and use the system via smart recommendations.

🔗 Inter-Agent Communication

Webhook System: Asynchronous and secure webhook-based message passing.
Structured JSON: All data transferred as validated JSON with typed parameters.
Response Validation: Ensures consistent agent interactions and recoverability.

🌍 Agent Registration & Discovery

Domain-Specific Capabilities: Each agent registered with structured documentation and expected inputs/outputs.
Dynamic Discovery: Agents discover others via Search and Discover APIs on Agentverse.

🤖 LLM Integration

Used Google Gemini Pro for:
- Agent-specific prompts
- Tuned temperature settings
- Structured JSON outputs

💾 Stateful Memory

Therapy Agent: Stores conversational context using LangChain.
Assistant Agent: Maintains routing context and conversation type to ensure smooth flow.

🔧 AI Workflow planning Agent: AI for Agent Builders

Workflow Agent is a specialized agent that helps developers create, manage, and deploy agent-based mental health systems.
Agent Address: test-agent://agent1q0n0uv6zqgzh6pclzyae4qwzzmcx60cxnc5y4cxmj8ggfntne4glygzffs4

It is a meta-agent that supports:

Designing new agent architectures
Guiding developers through agent registration and deployment
Generating agent prompts and payload specs
Validating webhook connections and endpoint behaviors

🧩 Features

Step-by-step flow for building new agents
Recommends ideal payload formats
Generates templated code for webhook handlers
Acts as a bridge between technical devs and the Agentverse platform
Provides live debugging support through LLM-assisted issue resolution

🚧 Challenges We Faced

Challenge	Solution
Agent Coordination	Built robust retry logic and fallback systems for inter-agent routing
Identity Management	Managed secure agent seed phrases using local vaults
Cold Start Routing	Designed a fallback routing strategy when new agents were not pre-indexed
Response Consistency	Unified UI and agent interaction schema with strict JSON validation
Webhook Reliability	Used exponential backoff and ping validation endpoints
Multi-Agent Testing	Created simulation harness to test interactions across agents

🏆 Accomplishments

✅ True agent-based architecture with full Fetch.ai Agentverse integration
✅ Seamless routing via Assistant Agent with >90% intent resolution accuracy
✅ Built and deployed six specialized mental health agents
✅ Enabled discovery of external agents via Guide Agent
✅ Created a developer agent (Workflow Agent) to assist in new agent creation
✅ Built robust webhook messaging and async error recovery system

📚 What We Learned

Building scalable multi-agent architectures requires clear interface design
Webhook-based systems need proactive error handling and response validation
Natural language routing can be effectively achieved with LLMs and JSON output formatting
Agent discovery is essential for building extensible ecosystems
Secure identity and key management is crucial in decentralized agent platforms
Agents should remain small and focused for maintainability and composability

🔮 Future Prospects

For Story.AI

🧘 Routine Builder Agent:
- Use voice-based prompts to co-create morning and evening routines
- Integrate journaling and gratitude flows with personalized activity suggestions
- Serve as a behavioral nudge assistant to increase adherence to mental health plans
💬 Multi-modal Agent Expansion:
- Voice journaling, image-based mood boards, TTS therapy agents
🧠 Condition-Specific Agents:
- Specialized agents for anxiety, depression, ADHD, etc.
🌐 User-Customized Agent Networks:
- Let users assemble their own support team using preferred agents

For Workflow Agent

🤖 AI-Creating-AI Tools:
- A system where agents help create new agents autonomously
- Based on user requirements, Workflow Agent can generate entire agent blueprints with prompt strategies, webhook code, and payload formats
🧪 Simulation-Based Testing Toolkit:
- Build test cases where the Workflow Agent simulates interactions between agents before real deployment
📦 Agent Marketplace SDK:
- Helps developers list and monetize agents within the broader Agentverse

🚀 Final Thoughts

Story.AI and Workflow Agent showcase the power of composable AI agents in solving deeply personal problems like mental health support and system creation. By building on Agentverse, we’ve made a flexible, future-ready platform where both users and developers can thrive.

We believe the future is multi-agent — and we’re just getting started.