GoAgent

GoAgent is a generic multi-agent framework implemented in Go, supporting multi-agent collaboration, memory management, and tool invocation.

Architecture Diagram

Embedding Gateway Service (FastAPI)

The Embedding Service is a standalone vector embedding service for GoAgent, supporting multiple backends:

Embedding Service Features:

• High Performance: Supports Ollama local deployment and SentenceTransformers cloud deployment
• Smart Caching: Redis cache + text normalization to avoid cache misses
• Batch Processing: Supports batch vector generation for improved efficiency
• Auto Normalization: Vectors automatically normalized to unit vectors for accurate cosine similarity
• Health Check: Built-in health check endpoint

Configuration File: services/embedding/.env

BACKEND_TYPE=ollama              # Backend type: ollama / transformers
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=qwen3-embedding:0.6b
MODEL_NAME=qwen3-embedding:0.6b
EMBEDDING_DIM=1024
REDIS_URL=redis://localhost:6379
CACHE_TTL=86400
HOST=0.0.0.0
PORT=8000

Code Locations:

• services/embedding/app.py - Service main program
• services/embedding/config.py - Configuration management
• internal/storage/postgres/embedding/client.go - Go client

Memory Distillation model

Tech Stack

Core Technologies

• Language: Go 1.26+
• Database: PostgreSQL 15+ with pgvector extension
• Concurrency: errgroup, sync
• Protocol: Custom AHP Protocol
• Embedding Service: FastAPI + Ollama/SentenceTransformers
• Cache: Redis

Main Components

Component	Purpose	Code Location
Agent System	Leader/Sub Agent collaboration	internal/agents/
Protocol Layer	Inter-agent communication and heartbeat	internal/protocol/ahp/
Memory System	Session, task, and distilled memory	internal/memory/
Storage Layer	PostgreSQL + pgvector	internal/storage/postgres/
Tool System	Tool registry and invocation	internal/tools/
Workflow Engine	DAG workflow orchestration	internal/workflow/engine/
Embedding Service	Vector embedding generation	services/embedding/

Dependencies

• github.com/jackc/pgx/v5 - PostgreSQL driver
• github.com/google/uuid - UUID generation
• github.com/stretchr/testify - Testing framework
• golang.org/x/sync - Concurrent extensions
• gopkg.in/yaml.v3 - YAML parsing
• fastapi - Embedding service framework
• redis - Cache support

Configuration

1. LLM Configuration

Config File: examples/travel/config.yaml

llm:
  provider: openrouter        # LLM provider: openai / ollama / openrouter
  api_key: ""                  # API Key (recommended: use env var OPENROUTER_API_KEY)
  base_url: https://openrouter.ai/api/v1
  model: meta-llama/llama-3.1-8b-instruct
  timeout: 60                  # Request timeout (seconds)
  max_tokens: 2048              # Max response tokens

Code Location: internal/llm/client.go:80-100

2. Agent Configuration

agents:
  leader:
    id: leader-travel
    max_steps: 10              # Max execution steps
    max_parallel_tasks: 4      # Max parallel tasks
    enable_cache: true          # Enable caching

  sub:
    - id: agent-destination
      type: destination         # Agent type: destination/food/hotel/itinerary
      triggers: [destination]   # Trigger keywords
      max_retries: 3             # Max retry attempts
      timeout: 30                # Timeout (seconds)

Code Location: internal/agents/leader/agent.go:30-50

3. Database Configuration

storage:
  enabled: true               # Enable PostgreSQL storage
  type: postgres
  host: localhost
  port: 5433                # Docker default port is 5433
  user: postgres
  password: postgres
  database: goagent
  
  pgvector:
    enabled: true             # Enable pgvector
    dimension: 1024           # Vector dimension

Code Location: internal/storage/postgres/pool.go:35-50

4. Embedding Service Configuration

embedding:
  service_url: http://localhost:8000    # Embedding service address
  model: qwen3-embedding:0.6b          # Model name
  dimension: 1024                       # Vector dimension
  timeout: 30                           # Request timeout (seconds)

Code Location: internal/storage/postgres/embedding/client.go:30-50

5. Memory Configuration

memory:
  enabled: true               # Enable memory system
  enable_distillation: true   # Enable memory distillation
  distillation_threshold: 3   # Trigger distillation every N rounds

Code Location: examples/knowledge-base/main.go:750-760

6. Retrieval Configuration

knowledge:
  chunk_size: 1000             # Document chunk size
  chunk_overlap: 100            # Chunk overlap
  top_k: 10                    # Return top K results
  min_score: 0.6               # Minimum similarity threshold

Code Location: internal/storage/postgres/repositories/knowledge_repository.go:100-120

Quick Start

1. Set Environment

# Set API Key (recommended: use environment variable)
export OPENROUTER_API_KEY="your-api-key"

# Or set in config file (not recommended)

2. Start Database (Optional, for persistence)

# Quick start PostgreSQL + pgvector with Docker
docker run -d \
  --name goagent-db \
  -e POSTGRES_PASSWORD=postgres \
  -e POSTGRES_DB=goagent \
  -p 5433:5432 \
  pgvector/pgvector:pg15

# Verify connection
docker exec -it goagent-db psql -U postgres -d goagent -c "SELECT version();"

3. Start Embedding Service (for vector retrieval)

# Navigate to embedding service directory
cd services/embedding

# Run setup script (install dependencies and model)
./setup.sh

# Start service
./start.sh

# Verify service
curl http://localhost:8000/health

4. Run Examples

# Travel planning example
cd examples/travel
go run main.go

# Knowledge base Q&A example (requires database + embedding service)
cd examples/knowledge-base
go run main.go --save README.md  # Import document
go run main.go --chat             # Start Q&A

Project Structure

goagent/
├── examples/               # Example applications
│   ├── travel/              # Travel planning
│   ├── knowledge-base/       # Knowledge base Q&A
│   └── simple/              # Simple example
├── internal/                # Core implementation
│   ├── agents/              # Agent system
│   │   ├── base/            # Agent base interfaces
│   │   ├── leader/          # Leader Agent
│   │   └── sub/             # Sub Agent
│   ├── protocol/             # AHP protocol
│   ├── storage/              # PostgreSQL + pgvector
│   ├── memory/               # Memory system
│   └── workflow/             # Workflow engine
├── services/                # Standalone services
│   └── embedding/           # Embedding service
│       ├── app.py           # FastAPI service
│       ├── config.py        # Configuration management
│       └── requirements.txt # Python dependencies
├── api/                     # API layer
│   ├── service/             # Service interfaces
│   └── client/              # Client
└── docs/                    # Documentation

Documentation

• Quick Start Guide - Detailed installation and configuration guide
• FAQ - Common issues and solutions
• Architecture Documentation - Complete architecture design
• Embedding Service Documentation - Embedding service details
• Integration Guide - How to integrate into existing projects

Migration Guide (v0.2)

Breaking Changes

1. TaskPlanner.Plan() signature changed

   • Before: Plan(ctx, profile) (2 args)
   • After: Plan(ctx, profile, inputText) (3 args)
   • Fix: Add inputText parameter to all Plan() calls.

2. NewResultAggregator() requires sortBy parameter

   • Before: NewResultAggregator(enableDedupe, maxItems)
   • After: NewResultAggregator(enableDedupe, maxItems, sortBy)
   • Fix: Add leader.SortByNone (or SortByPriority / SortByCreatedAt) as third argument.

3. Fashion-specific constants removed

   • Removed: AgentTypeShoes, AgentTypeHead, AgentTypeAccessory, StyleCasual, StyleFormal, StyleStreet, OccasionDaily, OccasionParty, OccasionDate
   • Fix: Replace with models.StyleTag("...") and models.Occasion("...") string types.

4. Domain types renamed (internal/tools/resources/types/)

   • FashionFilters → ResourceFilters
   • FashionItem → ResourceItem
   • AgentProfile → AgentUserProfile
   • AgentRecommendation → TaskRecommendation
   • OutfitSuggestion → Suggestion
   • AgentTrend → Trend

5. Validation default schema type changed

   • Before: "fashion"
   • After: "default"
   • Fix: Update config if explicitly setting validation.schema_type.

Examples

• Travel Planning - Multi-agent collaboration
• Knowledge Base Q&A - Vector search
• Simple Example - Basic usage
• Capability Demo - Full feature showcase

Development Guide

Running Tests

# Run all tests
go test ./...

# Run tests for specific package
go test ./internal/agents/...

# Run tests with coverage
go test -cover ./...

Building Project

# Build examples
go build -o bin/simple ./examples/simple
go build -o bin/travel ./examples/travel

# Build CLI tools
go build -o bin/migrate ./cmd/migrate_goagent

Code Standards

# Format code
go fmt ./...

# Run linter
golangci-lint run

---