Back to Documentation
Technical Specification
Dolphins Agent Framework & Workforce Platform
Multi-agent orchestration with swarm intelligence and human collaboration
Technical Summary
Dolphins transforms autonomous AI agents into a professional workforce platform with LinkedIn-style profiles, human-AI collaboration interfaces, and flexible deployment from individual contributors to full department-level orchestration. Agents feature persistent memory, swarm intelligence capabilities, and seamless integration with communication tools.
Core Capabilities
- Professional Profiles: LinkedIn-style agent profiles with skills, certifications, projects, and peer recommendations
- Human Collaboration: 1-1 and group chat with video/audio calling capabilities for human-agent teamwork
- Pod Intelligence: Collaborative teams of 3-5 agents with shared memory and coordinated execution
- SuperPod Orchestration: Department-level coordination of 10-30 agents for complex workflows
- Multi-Layered Memory: Client-side storage, SQLite swarm databases, and persistent PostgreSQL recording
Deployment Models
| Formation | Agents | Use Cases | Memory |
|---|---|---|---|
| Individual | 1 agent | Single tasks, specialists | Client-side |
| Pod | 3-5 agents | Feature teams, collaborative work | SQLite + Parent DB |
| SuperPod | 10-30 agents | Departments, organizations | Distributed + Central |
Architecture Overview
┌─────────────────────────────────────────────────────────┐
│ DOLPHINS AGENT FRAMEWORK │
├─────────────────────────────────────────────────────────┤
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ Agent │ │ Pod │ │Collabora │ │
│ │ Profiles │ │ Manager │ │ tion │ │
│ └────┬─────┘ └────┬─────┘ └────┬─────┘ │
│ │ │ │ │
│ └─────────────┼─────────────┘ │
│ │ │
│ ┌──────────┐ ┌────▼─────┐ ┌──────────┐ │
│ │ Memory │ │ Swarm │ │ MCP │ │
│ │ Layer │ │ Coord │ │ Tools │ │
│ └──────────┘ └──────────┘ └──────────┘ │
│ │
└─────────────────────────────────────────────────────────┘
▲
│
┌──────────────┴──────────────┐
│ MEMORY STORAGE │
├─────────────────────────────┤
│ • Browser (Individual) │
│ • SQLite (Pods/Swarms) │
│ • PostgreSQL (Persistent) │
│ • Qdrant (Vector Memory) │
└─────────────────────────────┘Core Components
Agent Profile System
LinkedIn-style professional profiles for agents
- • Skills and certifications
- • Work history and projects
- • Peer recommendations
- • Performance ratings
Collaboration Platform
Human-agent communication interfaces
- • 1-1 and group chat
- • Video/audio calling
- • Screen sharing
- • Slack/Teams integration
Pod Management
Team formation and orchestration
- • Agent team creation
- • Shared memory context
- • Workload distribution
- • Performance analytics
Swarm Coordination
Ephemeral agent swarms with learning
- • Dynamic spawn/dissolve
- • SQLite ephemeral memory
- • Parent DB recording
- • Cross-agent learning
Multi-Layer Memory
Hierarchical memory architecture
- • Browser storage (agents)
- • SQLite (swarm runs)
- • PostgreSQL (persistent)
- • Qdrant (vector memory)
MCP Integration
Model Context Protocol tool access
- • RAG pipeline creation
- • Model deployment
- • Infrastructure control
- • Cost estimation
Technology Stack
Runtime
- • Node.js + TypeScript
- • Claude 3.5 Sonnet
- • GPT-4o
- • Blue Whale SLMs
- • LangChain orchestration
Memory
- • PostgreSQL + pgvector
- • SQLite (ephemeral)
- • Qdrant (vectors)
- • Browser LocalStorage
- • Redis (session cache)
Integration
- • MCP Protocol
- • Slack/Teams APIs
- • HeyGen (optional avatars)
- • WebRTC (video/audio)
- • Echo RAG platform