MAS is a modern, high-performance Go multi-agent system framework designed for building intelligent applications with native Go performance and type safety.
Documentation | Examples | Chinese Version
go get github.com/voocel/masSet up environment variables:
export LLM_API_KEY="your-api-key"
export LLM_MODEL="gpt-5"
export LLM_BASE_URL="https://api.openai.com/v1" # optionalBasic usage:
package main
import (
"context"
"fmt"
"os"
"github.com/voocel/mas/agent"
"github.com/voocel/mas/llm"
"github.com/voocel/mas/runtime"
"github.com/voocel/mas/schema"
)
func main() {
// Create LLM model
model, _ := llm.NewLiteLLMAdapter(
os.Getenv("LLM_MODEL"),
os.Getenv("LLM_API_KEY"),
os.Getenv("LLM_BASE_URL"),
)
// Create agent
agent := agent.NewAgent("assistant", "AI Assistant", model)
// Execute
ctx := runtime.NewContext(context.Background(), "session-1", "trace-1")
response, _ := agent.Execute(ctx, schema.Message{
Role: schema.RoleUser,
Content: "Hello, how are you?",
})
fmt.Printf("Agent: %s\n", response.Content)
}Agent with tools:
import "github.com/voocel/mas/tools/builtin"
// Create agent with calculator tool
calculator := builtin.NewCalculator()
agent := agent.NewAgent("math-agent", "Math Assistant", model,
agent.WithTools(calculator),
)
response, _ := agent.Execute(ctx, schema.Message{
Role: schema.RoleUser,
Content: "Calculate 15 * 8 + 7",
})Multi-agent collaboration:
import "github.com/voocel/mas/orchestrator"
// Create orchestrator
orch := orchestrator.NewOrchestrator()
// Add agents
orch.AddAgent("researcher", researcher)
orch.AddAgent("writer", writer)
// Execute with routing
result, _ := orch.Execute(ctx, orchestrator.ExecuteRequest{
Input: schema.Message{
Role: schema.RoleUser,
Content: "Research and write about AI trends",
},
Type: orchestrator.ExecuteTypeAuto, // Auto-route to best agent
})Workflow orchestration:
import "github.com/voocel/mas/workflows"
// Create workflow steps
preprocessStep := workflows.NewFunctionStep(
workflows.NewStepConfig("preprocess", "Data preprocessing"),
func(ctx runtime.Context, input schema.Message) (schema.Message, error) {
// Processing logic
return processedMessage, nil
},
)
// Build chain workflow
workflow := workflows.NewChainBuilder("data-pipeline", "Data processing pipeline").
Then(preprocessStep).
Then(analysisStep).
Then(summaryStep).
Build()
// Execute workflow
orch.AddWorkflow("data-pipeline", workflow)
result, _ := orch.Execute(ctx, orchestrator.ExecuteRequest{
Input: inputMessage,
Target: "data-pipeline",
Type: orchestrator.ExecuteTypeWorkflow,
})Agent - Autonomous intelligent entities with:
- Smart Decision Making: LLM-powered reasoning and tool selection
- Dynamic Behavior: Adapts to different scenarios and contexts
- Tool Integration: Access to various tools and capabilities
- Memory Management: Maintains conversation history and context
Workflow - Structured process orchestration with:
- Deterministic Execution: Predefined steps and control flow
- Parallel Processing: Concurrent execution of independent tasks
- Conditional Branching: Dynamic routing based on conditions
- Composability: Steps can be functions, agents, or other workflows
Flexible Composition:
- Agent in Workflow: Workflow steps can be agents
- Workflow in Agent: Agents can call workflows as tools
- Hybrid Execution: Mix agents and workflows in the same task
Apache License 2.0 License