Skip to content
⚠️ This is documentation for AG2 Classic (autogen package), which is in maintenance mode. For the latest version, use AG2 v1.0+ (ag2 package) at docs.ag2.ai.

Release History — AG2 Classic and the Path to v1.0#

AG2 Classic is the original AG2 framework — autogen.agentchat with ConversableAgent, GroupChat, LLMConfig, and the surrounding agents, tools, and capabilities. It is published on PyPI as the autogen package and lives in the ag2classic repository.

This page records how AG2 Classic evolved and the path that led to AG2 v1.0.

Where things stand today

  • AG2 Classic (the autogen package) is in maintenance mode — it receives critical and security fixes only.
  • AG2 v1.0+ (the ag2 package) is the actively developed line. Its documentation lives at docs.ag2.ai.

Background#

For most of its life, AG2 shipped a single framework built around ConversableAgent and GroupChat. As the project matured, two goals emerged:

  • Tidy up the original framework through a round of deprecations, then move it to long-term maintenance.
  • Build a new, redesigned framework — developed under autogen.beta and centred on a new Agent — that would become the official AG2 at v1.0.

The releases below trace that transition from v0.12 through to v1.0.

v0.12 — Deprecation notices#

This release opened the deprecation and feedback period that prepared the original framework for maintenance mode.

  • Deprecation announcements. The features listed below were marked as deprecated and scheduled for removal in v0.14.
  • Community feedback window opened. Users were invited to flag any deprecation-bound feature that was critical to their workflow before v0.14.
  • Beta development continued in parallel.

Deprecated agents#

Agent Alternative
GPTAssistantAgent ConversableAgent
LLaVAAgent AG2 v1.0 provides native multimodal support on Agent
WebSurferAgent (contrib) autogen.agents.experimental.WebSurferAgent
TextAnalyzerAgent ConversableAgent with an appropriate system message
MathUserProxyAgent ConversableAgent with tool calling
SocietyOfMindAgent GroupChat patterns
AgentOptimizer ConversableAgent with tool calling
RetrieveAssistantAgent AssistantAgent
QdrantRetrieveUserProxyAgent RetrieveUserProxyAgent with vector_db='qdrant'
SwarmAgent ConversableAgent (already incorporated)
RealtimeAgent Relied on deprecated realtime API endpoints

Deprecated swarm functions and the swarm module#

The entire autogen.agentchat.contrib.swarm_agent module was removed in v0.14. All swarm orchestration is now provided by group chat.

Feature Alternative
initiate_swarm_chat(), run_swarm() (and async variants) run_group_chat
AFTER_WORK, AfterWork, AfterWorkOption Group chat handoffs (Handoffs, transition targets)
ON_CONDITION, OnCondition autogen.agentchat.group.OnCondition / OnContextCondition

Deprecated capabilities and modules#

Feature Reason
ImageGeneration capability Depended on the deprecated TextAnalyzerAgent
agent_eval module (generate_criteria, quantify_criteria, CriticAgent, QuantifierAgent, SubCriticAgent) Experimental evaluation framework

Deprecated interoperability#

Feature Alternative
CrewAIInteroperability LangChain or PydanticAI interop

Deprecated experimental tools#

Tool Alternative
PythonCodeExecutionTool AG2 v1.0 CodeExecutionTool
FirecrawlTool Crawl4AITool or BrowserUseTool
SearxngSearchTool DuckDuckGoSearchTool or TavilySearchTool
WebSearchPreviewTool AG2 v1.0 WebSearchTool

What was not deprecated#

The core framework and actively maintained features carried through unchanged, and they remain supported in AG2 Classic today:

  • Core agents: ConversableAgent, AssistantAgent, UserProxyAgent
  • Group chat: GroupChat, GroupChatManager, run_group_chat, and all orchestration patterns including handoffs (Handoffs, OnCondition, OnContextCondition, and transition targets)
  • RAG: RetrieveUserProxyAgent, vector database integrations (ChromaDB, Qdrant, MongoDB, Couchbase, PGVector), graph RAG (Neo4j, FalkorDB)
  • Capabilities: Teachability, VisionCapability, TransformMessages, ToolsCapability
  • Experimental agents: WebSurferAgent (experimental), DeepResearchAgent, ReasoningAgent, DocAgent, CaptainAgent, A2UIAgent, and platform agents (Discord, Slack, Telegram)

Notebooks and documentation removed in v0.14#

A number of notebooks that referenced unavailable or outdated models (e.g. gpt-3.5-turbo, gpt-4-vision-preview, gpt-4-turbo-preview) were removed in v0.14, along with the associated blog posts and user guide pages for the deprecated features above.

v0.13 — Transition period#

  • Community feedback incorporated. Deprecation decisions were revisited based on feedback gathered during the v0.12 window.
  • Beta API refinements. Development and stabilisation of the beta API continued, and its orchestration capabilities were introduced.
  • This was the last release in which the deprecated features were still available.

v0.14 — Final Classic release#

  • Deprecated features removed. Everything marked deprecated in v0.12 (and not rescued by community feedback) was removed. See the lists above.
  • Beta moved to Release Candidate (RC). The beta API became feature-complete and entered final testing.
  • This was the last feature release of the original AG2 codebase.

v1.0 — The new AG2#

  • The redesigned framework became stable and took over as AG2 v1.0, published as the ag2 package.
  • The original framework became AG2 Classic — it continues as the autogen package in maintenance mode, developed in the ag2classic repository.
  • Documentation — Primary documentation now covers AG2 v1.0 at docs.ag2.ai; this AG2 Classic documentation remains available at classic.docs.ag2.ai.

Migrating to AG2 v1.0#

AG2 Classic will be supported with critical and security fixes. When you're ready to move to the new framework — with its redesigned Agent, tools, and multi-agent capabilities — head to docs.ag2.ai for the v1.0 guides and migration notes.

Community feedback#

As an open-source project, we welcome feedback from the AG2 Classic community. If you rely on Classic and have questions, run into issues, or want to share what you need, please open an Issue or reach out on our Discord server.