Release Notes

Version 1.2.1 (2025-10-15)

🧭 Documentation Reorganization

• Adopted clear separation between implemented docs and design/WIP:
  • website/ → Implemented, user-facing documentation (VitePress)
  • docs/ → Design drafts, migration plans, tasks, and developer/internal docs
• Added docs/README.md to document the policy and entry points
• Added CONTRIBUTING.md for contributor onboarding

🔗 Link Fixes

• Updated README, CLAUDE, CHANGELOG, and release notes to point to canonical pages under website/advanced/* and website/guide/*
• Removed outdated references to duplicated files under docs/

🌐 Website Updates

• API: Added Knowledge API details to AgentOS page (/api/agentos)
• Ensured Workflow History and Performance pages are the canonical references

✅ No Behavior Changes

• This release updates documentation only; runtime behavior is unchanged

✨ New in 1.2.1 (Implemented)

• SSE event filtering for streaming endpoints (A2A)
  • POST /api/v1/agents/:id/run/stream?types=token,complete
  • Emits only requested event types; standard SSE format; context cancel supported
• Content extraction middleware for AgentOS
  • JSON/Form → context injection of content/metadata/user_id/session_id
  • Request size guard via MaxRequestSize; skip paths supported
• Google Sheets toolkit (service account)
  • read_range, write_range, append_rows; JSON or file credentials
• Minimal knowledge ingestion endpoint
  • POST /api/v1/knowledge/content supports text/plain and application/json

Enterprise validation steps: see docs/ENTERPRISE_MIGRATION_PLAN.md.

Version 1.1.0 (2025-10-08)

🎉 Highlights

This release brings powerful new features for production-ready multi-agent systems:

• A2A Interface - Standardized agent-to-agent communication protocol
• Session State Management - Persistent state across workflow steps
• Multi-Tenant Support - Serve multiple users with a single agent instance
• Model Timeout Configuration - Fine-grained timeout control for LLM calls

---

✨ New Features

A2A (Agent-to-Agent) Interface

A standardized communication protocol for agent-to-agent interactions based on JSON-RPC 2.0.

Key Features:

• REST API endpoints (/a2a/message/send, /a2a/message/stream)
• Multi-media support (text, images, files, JSON data)
• Server-Sent Events (SSE) for streaming
• Compatible with Python Agno A2A implementation

Quick Example:

import "github.com/rexleimo/agno-go/pkg/agentos/a2a"

// Create A2A interface
a2a := a2a.New(a2a.Config{
    Agents: []a2a.Entity{myAgent},
    Prefix: "/a2a",
})

// Register routes (Gin)
router := gin.Default()
a2a.RegisterRoutes(router)

📚 Learn More: A2A Interface Documentation

---

Workflow Session State Management

Thread-safe session management for maintaining state across workflow steps.

Key Features:

• Cross-step persistent state storage
• Thread-safe with sync.RWMutex
• Deep copy for parallel branch isolation
• Smart merge strategy to prevent data loss
• Fixes Python Agno v2.1.2 race condition

Quick Example:

// Create context with session info
execCtx := workflow.NewExecutionContextWithSession(
    "input",
    "session-123",  // Session ID
    "user-a",       // User ID
)

// Access session state
execCtx.SetSessionState("key", "value")
value, _ := execCtx.GetSessionState("key")

📚 Learn More: Session State Documentation

---

Multi-Tenant Support

Serve multiple users with a single Agent instance while ensuring complete data isolation.

Key Features:

• User-isolated conversation history
• Optional userID parameter in Memory interface
• Backward compatible with existing code
• Thread-safe concurrent operations
• ClearAll() method for cleanup

Quick Example:

// Create multi-tenant agent
agent, _ := agent.New(&agent.Config{
    Name:   "customer-service",
    Model:  model,
    Memory: memory.NewInMemory(100),
})

// User A's conversation
agent.UserID = "user-a"
output, _ := agent.Run(ctx, "My name is Alice")

// User B's conversation
agent.UserID = "user-b"
output, _ := agent.Run(ctx, "My name is Bob")

📚 Learn More: Multi-Tenant Documentation

---

Model Timeout Configuration

Configure request timeout for LLM calls with fine-grained control.

Key Features:

• Default: 60 seconds
• Range: 1 second to 10 minutes
• Supported models: OpenAI, Anthropic Claude
• Context-aware timeout handling

Quick Example:

// OpenAI with custom timeout
model, _ := openai.New("gpt-4", openai.Config{
    APIKey:  apiKey,
    Timeout: 30 * time.Second,
})

// Claude with custom timeout
claude, _ := anthropic.New("claude-3-opus", anthropic.Config{
    APIKey:  apiKey,
    Timeout: 45 * time.Second,
})

📚 Learn More: Model Configuration

---

🐛 Bug Fixes

• Workflow Race Condition - Fixed parallel step execution data race
  • Python Agno v2.1.2 had shared session_state dict causing overwrites
  • Go implementation uses independent SessionState clones per branch
  • Smart merge strategy prevents data loss in concurrent execution

---

📚 Documentation

All new features include comprehensive bilingual documentation (English/中文):

• A2A Interface Guide - Complete protocol specification
• Session State Guide - Workflow state management
• Multi-Tenant Guide - Data isolation patterns
• Model Configuration - Timeout settings

---

🧪 Testing

New Test Suites:

• session_state_test.go - 543 lines of session state tests
• memory_test.go - Multi-tenant memory tests (4 new test cases)
• agent_test.go - Multi-tenant agent test
• openai_test.go - Timeout configuration test
• anthropic_test.go - Timeout configuration test

Test Results:

• ✅ All tests passing with -race detector
• ✅ Workflow coverage: 79.4%
• ✅ Memory coverage: 93.1%
• ✅ Agent coverage: 74.7%

---

📊 Performance

No Performance Regression - All benchmarks remain consistent:

• Agent instantiation: ~180ns/op (16x faster than Python)
• Memory footprint: ~1.2KB per agent
• Thread-safe concurrent operations

---

⚠️ Breaking Changes

None. This release is fully backward compatible with v1.0.x.

---

🔄 Migration Guide

No migration needed - All new features are additive and backward compatible.

Optional Enhancements:

1. Enable Multi-Tenant Support:

   // Add UserID to agent configuration
   agent := agent.New(agent.Config{
       UserID: "user-123",  // NEW
       Memory: memory.NewInMemory(100),
   })

2. Use Session State in Workflows:

   // Create context with session
   ctx := workflow.NewExecutionContextWithSession(
       "input",
       "session-id",
       "user-id",
   )

3. Configure Model Timeout:

   // Add Timeout to model config
   model, _ := openai.New("gpt-4", openai.Config{
       APIKey:  apiKey,
       Timeout: 30 * time.Second,  // NEW
   })

---

📦 Installation

go get github.com/rexleimo/agno-go@v1.1.0

---

🔗 Links

• GitHub Release: v1.1.0
• Full Changelog: CHANGELOG.md
• Documentation: https://agno-go.dev

---

Version 1.0.3 (2025-10-06)

🧪 Improved

Testing & Quality

• Enhanced JSON Serialization Tests - Achieved 100% test coverage for utils/serialize package
• Performance Benchmarks - Aligned with Python Agno performance testing patterns
• Comprehensive Documentation - Added bilingual package documentation

Performance

• ToJSON: ~600ns/op, 760B/op, 15 allocs/op
• ConvertValue: ~180ns/op, 392B/op, 5 allocs/op
• Agent Creation: ~180ns/op (16x faster than Python)

---

Version 1.0.2 (2025-10-05)

✨ Added

GLM (智谱AI) Provider

• Full integration with Zhipu AI's GLM models
• Support for GLM-4, GLM-4V (vision), GLM-3-Turbo
• Custom JWT authentication (HMAC-SHA256)
• Synchronous and streaming API calls
• Tool/function calling support
• Test coverage: 57.2%

Quick Example:

model, _ := glm.New("glm-4", glm.Config{
    APIKey:      os.Getenv("ZHIPUAI_API_KEY"),
    Temperature: 0.7,
})

---

Version 1.0.0 (2025-10-02)

🎉 Initial Release

Agno-Go v1.0 is a high-performance Go implementation of the Agno multi-agent framework.

Core Features

• Agent - Single autonomous agent with tool support (74.7% coverage)
• Team - Multi-agent collaboration with 4 modes (92.3% coverage)
• Workflow - Step-based orchestration with 5 primitives (80.4% coverage)

LLM Providers

• OpenAI (GPT-4, GPT-3.5, GPT-4 Turbo)
• Anthropic (Claude 3.5 Sonnet, Claude 3 Opus/Sonnet/Haiku)
• Ollama (Local models)

Tools & Storage

• Calculator, HTTP, File tools
• In-memory conversation storage (93.1% coverage)
• Session management
• ChromaDB vector database

Performance

• Agent creation: ~180ns/op (16x faster than Python)
• Memory footprint: ~1.2KB per agent
• Test coverage: 80.8% average

---

Previous Versions

See CHANGELOG.md for complete version history.

---

Last Updated: 2025-10-08