Release Notes

Version 1.2.9 (2025-11-14)

✨ Highlights

EvoLink provider for text, images, and video via pkg/agno/models/evolink/*, with new EvoLink Media Agents example pages showing end-to-end media workflows.
Knowledge upload chunking: POST /api/v1/knowledge/content now accepts chunk_size and chunk_overlap (JSON, text/plain query params, and multipart form fields) and records these values together with chunker_type in chunk metadata.
AgentOS HTTP documentation now covers custom health endpoints, /openapi.yaml, /docs, and server.Resync() guidance on the AgentOS API page.

📚 Documentation

Updated website/api/agentos.md (and localized variants) with Knowledge chunking parameters, examples, and HTTP surface tips.
Added EvoLink Media Agents example pages in English and Chinese for media-focused workflows.

✅ Compatibility

Additive release; no breaking API changes. Knowledge chunking parameters are optional and default to previous behavior when omitted.

Version 1.2.8 (2025-11-10)

✨ Highlights

Run Context propagation across hooks, tools, and telemetry with run_context_id included in streaming events for trace correlation.
Session state persists AGUI UI substate and returns it via GET /sessions/{id}.
Vector indexing:
- Pluggable VectorDB providers (Chroma default; Redis optional).
- VectorDB migration CLI for idempotent setup (migrate up/down).
Embeddings: VLLM provider (local/remote) via common interface.
MCPTools: optional tool_name_prefix to prefix registered tools.

🔧 Improvements

Redis removed as a default dependency for VectorDB; enabling Redis registers provider without affecting defaults.
Team model inheritance propagates only the primary model; auxiliary flags remain agent-scoped unless opted-in.

🐛 Fixes

Correctly bind model responses to active steps; fix unbound/zero-value histories.
Team tool determination aligns with OS schema; preserves member tools.
Async DB knowledge filters respect composite predicates and timeouts.
Toolkit import errors are structured; no panics on missing modules.
AgentOS error responses standardized for consistent contract tests.

🧪 Tests

Added coverage for Run Context, AGUI persistence, team inheritance, MCP prefixing, and VLLM embeddings.
Optional Redis tests gated; skipped when dependency absent.

✅ Compatibility

Additive update; public APIs unchanged. Optional features are off by default.

Version 1.2.7 (2025-11-03)

✨ Highlights

Go-native session service mirroring the Python AgentOS /sessions API with Postgres-backed CRUD, Chi routing, and health endpoints (guide).
Deployment assets across environments: dedicated Dockerfile, Docker Compose stack with Postgres, and a Helm chart for Kubernetes rollouts.
Documentation updates and test-session-api.sh helper script to validate endpoints locally and in CI.

🔧 Improvements

Postgres store implementation with typed DTOs and transaction-safe operations aligned with existing AgentOS schemas.
Configuration guidance covering DSN wiring, environment variables, and workflow scripts for the new session runtime.

🧪 Tests

Contract suite comparing Go responses to Python fixtures plus dedicated Postgres store tests.

✅ Compatibility

Additive update; the Go session runtime is optional and coexists with the Python service.

Version 1.2.6 (2025-10-31)

✨ Highlights

Session parity: session reuse endpoint, GET/POST /sessions/{id}/summary with sync/async modes, history filters (num_messages, stream_events), and run metadata for cache hits and cancellation reasons.
Response caching for agents and teams with in-memory LRU store plus configurable summary manager defaults.
Media attachments pipeline for agents, teams, and workflows (WithMediaPayload) with validation helpers.
Storage adapters for MongoDB and SQLite alongside Postgres, sharing the same JSON contracts.
Toolkit expansions: Tavily Reader/Search, Claude Agent Skills, Gmail mark-as-read, Jira worklogs, ElevenLabs speech synthesis, and enhanced file tooling.
Culture knowledge manager for curating organisational knowledge asynchronously.

🔧 Improvements

Workflow engine persists cancellation reasons, supports resume-from checkpoints, and handles media-only payloads.
AgentOS session APIs expose summary endpoints, reuse semantics, and history pagination with SSE toggles.
MCP client forwards media attachments and caches capability manifests for faster tool execution.

🧪 Tests

Added coverage for cache layer, summary manager, storage drivers, workflow resumptions, and new toolkits.

✅ Compatibility

Additive changes; existing APIs remain backward compatible.

Version 1.2.5 (2025-10-20)

✨ Highlights

新增 8 个模型提供商：Cohere、Together、OpenRouter、LM Studio、Vercel、Portkey、InternLM、SambaNova（支持同步与流式、函数调用）
新增评估系统（场景评测、指标汇总、多模型对比），媒体处理（图片元数据；音/视频占位），调试工具（请求/响应精简转储），云部署占位（NoopDeployer）
集成注册表（注册/列表/健康检查）与通用 utils（JSONPretty、Retry）

🔧 修复与改进

Airflow 工具返回结构对齐 Airflow REST API v2：total_entries、dag_run_id、logical_date
站点首页图片缺失：将 hero 图片从 /logo.svg 更改为 /logo.png

🧪 测试

为新模型与模块补充了聚焦单测；维持现有基准测试

✅ 兼容性

全部为增量功能，无破坏性变更

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:

Enable Multi-Tenant Support:

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

Use Session State in Workflows:

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

Configure Model Timeout:

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

📦 Installation

bash

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

Release Notes ​

Version 1.2.9 (2025-11-14) ​

✨ Highlights ​

📚 Documentation ​

✅ Compatibility ​

Version 1.2.8 (2025-11-10) ​

✨ Highlights ​

🔧 Improvements ​

🐛 Fixes ​

🧪 Tests ​

✅ Compatibility ​

Version 1.2.7 (2025-11-03) ​

✨ Highlights ​

🔧 Improvements ​

🧪 Tests ​

✅ Compatibility ​

Version 1.2.6 (2025-10-31) ​

✨ Highlights ​

🔧 Improvements ​

🧪 Tests ​

✅ Compatibility ​

Version 1.2.5 (2025-10-20) ​

✨ Highlights ​

🔧 修复与改进 ​

🧪 测试 ​

✅ 兼容性 ​

Version 1.2.1 (2025-10-15) ​

🧭 Documentation Reorganization ​

🔗 Link Fixes ​

🌐 Website Updates ​

✅ No Behavior Changes ​

✨ New in 1.2.1 (Implemented) ​

Version 1.1.0 (2025-10-08) ​

🎉 Highlights ​

✨ New Features ​

A2A (Agent-to-Agent) Interface ​

Workflow Session State Management ​

Multi-Tenant Support ​

Model Timeout Configuration ​

🐛 Bug Fixes ​

📚 Documentation ​

🧪 Testing ​

📊 Performance ​

⚠️ Breaking Changes ​

🔄 Migration Guide ​

📦 Installation ​

🔗 Links ​

Version 1.0.3 (2025-10-06) ​

🧪 Improved ​

Testing & Quality ​

Performance ​

Version 1.0.2 (2025-10-05) ​

✨ Added ​

GLM (智谱AI) Provider ​

Version 1.0.0 (2025-10-02) ​

🎉 Initial Release ​

Core Features ​

LLM Providers ​

Tools & Storage ​

Performance ​

Previous Versions ​

Release Notes

Version 1.2.9 (2025-11-14)

✨ Highlights

📚 Documentation

✅ Compatibility

Version 1.2.8 (2025-11-10)

✨ Highlights

🔧 Improvements

🐛 Fixes

🧪 Tests

✅ Compatibility

Version 1.2.7 (2025-11-03)

✨ Highlights

🔧 Improvements

🧪 Tests

✅ Compatibility

Version 1.2.6 (2025-10-31)

✨ Highlights

🔧 Improvements

🧪 Tests

✅ Compatibility

Version 1.2.5 (2025-10-20)

✨ Highlights

🔧 修复与改进

🧪 测试

✅ 兼容性

Version 1.2.1 (2025-10-15)

🧭 Documentation Reorganization

🔗 Link Fixes

🌐 Website Updates

✅ No Behavior Changes

✨ New in 1.2.1 (Implemented)

Version 1.1.0 (2025-10-08)

🎉 Highlights

✨ New Features

A2A (Agent-to-Agent) Interface

Workflow Session State Management

Multi-Tenant Support

Model Timeout Configuration

🐛 Bug Fixes

📚 Documentation

🧪 Testing

📊 Performance

⚠️ Breaking Changes

🔄 Migration Guide

📦 Installation

🔗 Links

Version 1.0.3 (2025-10-06)

🧪 Improved

Testing & Quality

Performance

Version 1.0.2 (2025-10-05)

✨ Added

GLM (智谱AI) Provider

Version 1.0.0 (2025-10-02)

🎉 Initial Release

Core Features

LLM Providers

Tools & Storage

Performance

Previous Versions