v0.2.11

Highlights

Two major upgrades of conversation programming (experimental features under active development):

• Finite state machine (FSM) based group chat: allow more fine-grained control of speaker order transition in group chat.
• Multiple sequential chats: program a sequence of chats and carry previous chat results to new chats.

AutoGen Studio updates:

• Workflow export.
• Skill editing.
• CSV support.
• Windows testing.

Other improvements:

• Allow timeout for code execution in windows.
• Make every agent receive the termination message in group chat.
• Improvement in documentation.
• Fix an import issue.

Thanks to @joshkyh @freedeaths @qingyun-wu @victordibia @pretherford @ekzhu @jackgerrits @samuelogboye @Xtrah @olgavrou @thinkall @abhaymathur21 and all the other contributors!

What's Changed

• Disable default code execution on society_of_mind and web_surfer. by @afourney in #1516
• Add notebooks section on website by @jackgerrits in #1495
• Update pyproject.toml for Poetry mismatch by @agamm in #1454
• Update tests for code_utils and use ThreadPoolExecutor for code execution in windows by @ekzhu in #1472
• remove useless code in groupchat; simplify the handling of termination msg by @sonichi in #1532
• Fix example for LLM configuration docs by @jackgerrits in #1528
• fix(path): Update Installation.md -Update Dockerfile path for Autogen base image build by @samuelogboye in #1425
• Update outdated link in README.md by @Xtrah in #1541
• Fix formatting of notebook contributing.md by @jackgerrits in #1540
• Custom Model Client docs follow-up by @olgavrou in #1545
• Fix docstrings of a_initiate_chat by @thinkall in #1539
• Autogenstudio Updates [CSV support, Workflow Export, Skill Editing, Windows Testing ] by @victordibia in #1475
• Graph group chat by @joshkyh in #857
• Update Docker.md by @abhaymathur21 in #1554
• patch to graph groupchat by @sonichi in #1555
• Import oai.client directly in ConversableAgent by @AkashKulkarni4444 in #1557
• New initiate_chats Interface for Managing Dependent Chats in ConversableAgent by @qingyun-wu in #1402

New Contributors

• @agamm made their first contribution in #1454
• @samuelogboye made their first contribution in #1425
• @Xtrah made their first contribution in #1541
• @abhaymathur21 made their first contribution in #1554
• @AkashKulkarni4444 made their first contribution in #1557

Full Changelog: v0.2.10...v0.2.11

v0.2.10

Breaking change

Change code_execution_config default in ConversableAgent to False to match the default value change of last_n_messages.

Highlights

• Custom model client for extensibility of the inference mechanism
• SocietyOfMindAgent: demonstrating using nested chat to compose a more capable single agent based on multi-agent chat
• Improvement of tool call and function call and GPTAssistantAgent
• Documentation improvement for function call, gallery, FAQ, notebooks etc.

Thanks to @olgavrou @afourney @davorrunje @jtrugman @ekzhu @namanbarkiya @maxim-saplin @jackgerrits @Yanni8 @victordibia @eltociear @pmalarme and all the other contributors!

What's Changed

• Update function call doc with example of not using decorator syntax. by @ekzhu in #1441
• Improve docs/gallery card component by @namanbarkiya in #1445
• FAQ, working with LLM endpoints and explaining why config-list is a list by @maxim-saplin in #1451
• fixed wrong doc link by @Yanni8 in #1449
• Removed "Tool Call Id" from main content string. by @afourney in #1471
• docs: initial Jupyter support for website docs, move config notebook by @jackgerrits in #1448
• Adds a SocietyOfMindAgent that presents as a single agent, but runs GroupChat as an inner-monologue by @afourney in #890
• Function calling upgrade by @davorrunje in #1443
• fix: unit test should not call private function by @olgavrou in #1494
• Bump autogenbench version. by @afourney in #1485
• update readme to add note on required Quarto Version. Update readme t… by @victordibia in #1493
• Update README.md by @eltociear in #1491
• Implement Overwrite Tools Functionality in GPTAssistantAgent by @jtrugman in #1434
• Update Contribute.md, #1502 by @victordibia in #1508
• Fix image print for auto feedback from code notebook by @pmalarme in #1389
• deprecate using None for code_execution_config by @jackgerrits in #1506
• Added new models to token_count_utils by @afourney in #1511
• change code_execution_config default by @sonichi in #1518
• Fix tests for GPT assistant by @davorrunje in #1505
• fix broken links from moving oai utils notebook by @jackgerrits in #1497
• Custom Model Client support by @olgavrou in #1345

New Contributors

• @namanbarkiya made their first contribution in #1445
• @jtrugman made their first contribution in #1434
• @pmalarme made their first contribution in #1389

Full Changelog: v0.2.9...v0.2.10

v0.2.9

Highlights

• GroupChat support in AutoGen Studio.
• Introducing AutoGenBench, A Tool for Measuring and Evaluating AutoGen Agents.
• (Experimental) Allow agents to send history cleaning commands in group chat.
• Notebook example for SQL agent and Spider environment.
• New examples added to Gallery, such as stateful development environment in Jupyter Notebook and X-Force IDE.

Thanks to @victordibia @afourney @GregorD1A1 @byronxu99 @cemtu @olimoz @randombet and all the other contributors!

What's Changed

• Autogenstudio - Add GroupChat Support to UI by @victordibia in #1352
• Cleanup and unify Dockerfiles by @gagb in #1333
• Fixed skipping code execution in docker test if already running in a container by @davorrunje in #1383
• docs: new gallery item added. by @cemtu in #1365
• add blog and warning in readme for new breaking changes by @olgavrou in #1380
• test permissions by @olaoluwasalami in #1264
• Fix permission block for deploy-website.yml by @ekzhu in #1397
• Introduces AutoGenBench by @afourney in #1048
• Fixed broken links. by @afourney in #1408
• bump version to 0.2.9 by @sonichi in #1430
• Add SQL agent and Spider environment by @byronxu99 in #1218
• Add A Stateful Dev Environment by Jupiter example to gallery by @randombet in #1376
• Correct small typos in system messages by @pbadeer in #1410
• docs: split apart installation docs into parts by @jackgerrits in #1420
• functionality of manual history cleaning by user proxy added by @GregorD1A1 in #1230

New Contributors

• @cemtu made their first contribution in #1365
• @byronxu99 made their first contribution in #1218
• @pbadeer made their first contribution in #1410
• @jackgerrits made their first contribution in #1420
• @GregorD1A1 made their first contribution in #1230

Full Changelog: v0.2.8...v0.2.9

v0.2.8

Breaking changes

1. use_docker defaults to True.
2. last_n_messages defaults to "auto".

Highlights

1. Support both redis cache and diskcache for agent chat and client.
2. Web surfer agent
3. Allow initiate_chat with no message to take human input.
4. Add dev container for AutoGen Studio.

Thanks to @olgavrou @afourney @vijaykramesh @ekzhu @balakreshnan @bitnom @gagb and all the other contributors!

What's Changed

• set use_docker to default to True by @olgavrou in #1147
• Add dev container for AutoGen Studio by @gagb in #1241
• Allow initiate_chat without passing message by @bitnom in #1244
• [Core] implement redis cache mode by @vijaykramesh in #1222
• More documentation for Cache. Updated FAQ to include database Locked error by @balakreshnan in #1291
• Add a web surfer agent that can search and browse the web. by @afourney in #1093
• bump version to 0.2.8 by @sonichi in #1349
• fix website production css by @cheng-tan in #1366
• Add an AutoGen UI example powered by Panel by @randombet in #1348
• Change last_n_messages to default to 'auto' by @afourney in #1356

New Contributors

• @olgavrou made their first contribution in #1147
• @vijaykramesh made their first contribution in #1222
• @balakreshnan made their first contribution in #1291
• @randombet made their first contribution in #1348

Full Changelog: v0.2.7...v0.2.8

v0.2.7

Headsup: in the next release we will change the default setting of use_docker in code_execution_config to True. It needs to be set to False or None explicitly to disable docker.

Highlights

• Support Python 3.12.
• Support function call style API in function decorator.
• Support AzureOpenAI when using azure endpoints.
• Add usage summary for agents.
• Improvement in docker usage and documentation.
• Bug fix for tool call, async, group chat.
• Improvement in AutoGen studio.
• Filter llm configurations by tag.

Thanks to @ekzhu, @maxim-saplin, @kevin666aa, @r3d91ll, @Yanni8, @yenif, @davorrunje, @shhdgit, @ashutoshpw, @afourney and all the other contributors!

What's Changed

• Update studio README.md with badges by @gagb in #1225
• Add documentation and raise exception when registering async reply function in sync chat by @davorrunje in #1208
• Upgrade docusaurus by @olaoluwasalami in #1101
• fix: forward kwargs to conversable agent + fix typing by @joseferben in #1193
• [Core] Sanitize filename before using it as docker image tag. Fix #1069 by @ekzhu in #1127
• [Core] Compatibility with function call style API (Azure OpenAI and Gemini) by @ekzhu in #1227
• Refined the user_proxy description. by @afourney in #951
• update default aoai version by @sonichi in #1229
• Fix bug in async group chat with func/tool call by @sonichi in #1243
• fix: Fixed typo from 'mont' to 'mount' by @eaandersen in #1250
• Improve test for function call in groupchat by @sonichi in #1252
• run sync and async in async by @yenif in #1242
• Added the ability to add tags to the OAI_CONFIG_LIST, and filter by @afourney in #1226
• Docker Documentation by @r3d91ll in #1251
• Replaced asyncio.coroutines.iscoroutinefunction with inspect.iscoroutinefunction by @davorrunje in #1267
• remove name from tool response by @yenif in #1263
• [Core] Update documentation for function call compatibility by @ekzhu in #1237
• add workdir in test by @sonichi in #1274
• remove filter_dict by @sonichi in #1284
• Add usage summary for agents by @kevin666aa in #1269
• [AutoGen Studio] Fix the new workflow process by @shhdgit in #1181
• Switched to AzureOpenAI for api_type=="azure" by @maxim-saplin in #1232
• removed alpine image by @Yanni8 in #1279
• Update README.md by @olivMertens in #1306
• Chore: Updated Blog and removed cd frontend by @ashutoshpw in #1301
• def _prepare_chat for groupchat manager to reset the groupchat by @yenif in #1254
• Fix website accessibility by @cheng-tan in #1153
• Support for Python 3.12 by @ekzhu in #1317

New Contributors

• @joseferben made their first contribution in #1193
• @eaandersen made their first contribution in #1250
• @r3d91ll made their first contribution in #1251
• @shhdgit made their first contribution in #1181
• @Yanni8 made their first contribution in #1279
• @olivMertens made their first contribution in #1306
• @ashutoshpw made their first contribution in #1301
• @cheng-tan made their first contribution in #1153

Full Changelog: v0.2.6...v0.2.7

v0.2.6

This release fixes bugs in tool calls. It also adds support for streaming tool calls.

Thanks to @yenif, @davorrunje and other contributors!

What's Changed

• cleaner definition of tool_responses fixes #1174 by @yenif in #1182
• Documentation: polish docs for tools calls by @davorrunje in #1185
• check tool_returns by @sonichi in #1187
• Change contrib-tests.yml tests to use --skip-openai by @maxim-saplin in #1132
• remove inappropriate api_type by @sonichi in #1190
• Bump follow-redirects from 1.15.2 to 1.15.4 in /website by @dependabot in #1189
• Added support for streaming tool calls by @davorrunje in #1184
• More async tool fixes by @yenif in #1204
• Fix registration of async functions by @davorrunje in #1201

Full Changelog: v0.2.5...v0.2.6

v0.2.5

This PR makes contrib/capability importable and adds support for streamed function calls. The documentation about installation is improved.

Thanks to @btcioner @bitnom @qingyun-wu @yuandong-tian and all the other contributors!

What's Changed

• guide on the usage of docker by @qingyun-wu in #1111
• nbqa adedd to pre-commit, added black and ruff for notebooks by @davorrunje in #1171
• Handle streamed function calls by @bitnom in #1118
• add init.py for capabilities by @btcioner in #1177
• bump version to 0.2.5 by @sonichi in #1180

New Contributors

• @bitnom made their first contribution in #1118
• @btcioner made their first contribution in #1177

Full Changelog: v0.2.4...v0.2.5

v0.2.4

Highlights

• Breaking change: Remove GPT-4 as the default model to avoid unexpected use of GPT-4. Require explicit setting the model when using LLM.
• Generalized teachability and extensibility: Allow any conversable agent to be teachable via a new and generic extensibility mechanism. Example: Teach OpenAI Assistants Through GPTAssistantAgent
• New feature in AutoBuild: support building agents from an agent library and generating agent descriptions for group chat. Example.
• Initial support of OpenAI tool calls in conversable agents.

Thanks to @rickyloynd-microsoft @LinxinS97 @yenif @davorrunje @afourney @ekzhu and all the other contributors!

What's Changed

• [Core] check if docker is installed if user_docker is specified. by @ekzhu in #1145
• Let run tests directly via python test_xyz.py, fix for conftest import via sys.path.append by @maxim-saplin in #1144
• Remove GPT-4 as the default model. by @afourney in #1072
• [AutoBuild] Supporting build agents from library; supporting generating agent descriptions by @LinxinS97 in #1039
• [Core] [Tool Call] adjust conversable agent to support tool_calls by @yenif in #974
• bump version to 0.2.4 by @sonichi in #1160
• Add codespell to pre-commit hooks and fix spelling of existing files by @davorrunje in #1161
• contributor wall & news update by @sonichi in #1165
• Teachability for any agent by @rickyloynd-microsoft in #1091

New Contributors

• @yenif made their first contribution in #974

Full Changelog: v0.2.3...v0.2.4

v0.2.3

Highlights

• Decorator for function calling makes function calls easier to define. Doc
• Address breaking changes introduced by new versions of openai.
• Introducing AgentOptimizer: An Agentic Way to Train Your LLM Agent
• Improvement in group chat, multimodal, GPTAssistant, retrieval chat, AutoBuild, testbed, config list loading utils. Blogpost about agent descriptions.
• Guidance + AutoGen example: constrained generation + multi-step reasoning
• The sample app AutoGen Assistant is renamed into AutoGen Studio with upgrades.
• Sample notebook for using AutoGen in Microsoft Fabric.
• Sample app of simple chat.
• Improvement in docs and tests.

Thanks to @davorrunje @skzhang1 @JieyuZ2 @afourney @BeibinLi @IANTHEREAL @LinxinS97 @cool-RR @LeoLjl @kevin666aa @gagb @victordibia @dragosMC91 @thinkall @danyrouh @olaoluwasalami @yuandong-tian @ShobhitVishnoi30 @ekzhu @maxim-saplin @rlam3 @bitsnaps and all the other contributors!

What's Changed

• Simple chat app by @danyrouh in #921
• Add privacy and cookies link to documentation website by @victordibia in #966
• Add collate file and more tests from autogpt into testbed by @LeoLjl in #915
• Allow users to specify the Docker image to use with Testbed by @afourney in #986
• fixed a typo fron -> from by @kar2905 in #992
• Website, update privacy link by @victordibia in #994
• [AutoBuild] address issue 941 954; add new feature; add debug information by @LinxinS97 in #944
• fix link that returns 404 by @olaoluwasalami in #945
• Updated get_max_token_limit with latest models and numbers by @afourney in #972
• Add missing assistant UI documentation steps by @dragosMC91 in #935
• correcting typo by @halr9000 in #931
• Update client.py by @eltociear in #934
• Enable allow_repeat_speaker to be a list of agents by @afourney in #905
• Add MATH tests to testbed by @kevin666aa in #914
• Fix exception cause in agent_builder.py by @cool-RR in #1007
• Update privacy link docusaurus.config.js by @victordibia in #1016
• Add async test by @gagb in #520
• Add examples constrained generation + multi-step reasoning via Guidance + AutoGen by @gagb in #937
• add links to sdk #910 by @olaoluwasalami in #958
• Change default model in LMM by @BeibinLi in #1032
• Fix/typo by @KazooTTT in #1034
• Update README to incorporate using existing docker image by @yuandong-tian in #1040
• Autobuild notebook minor fix by @AdiG1123 in #1049
• Update AutoGen Sample App | Rename AutoGen Assistant -> AutoGen Studio by @victordibia in #998
• Partial fix for 960 by @afourney in #963
• Add decorator for function calling by @davorrunje in #1018
• Add default value to the logprobs parameter by @davorrunje in #1044
• Fix exception causes all over the codebase #1007 by @cool-RR in #1014
• Add additional docs in retrieval agent if required by @ShobhitVishnoi30 in #1028
• Add poetry setup by @rlam3 in #824
• Fix mode dump with different pydantic version by @kevin666aa in #1063
• AgentOptimizer by @skzhang1 in #1053
• Filter out candidates with the same name but different instructions, … by @IANTHEREAL in #925
• [Core] Fix openai dependency error introduced when merging #1044 by @ekzhu in #1071
• fix: Fix typo by @hugo-syn in #1068
• [Docs/Contributing] Clarification on running non-OpenAI tests by @maxim-saplin in #1052
• Reuse docker image #929 by @olaoluwasalami in #956
• Agent description blog post by @afourney in #1092
• [Core] Extending inline docs for openai_utils.py, boosting coverage in test_utils.py, fixing #762 by @maxim-saplin in #1046
• [Dotnet] Add dotnet build workflow by @LittleLittleCloud in #946
• add example to website by @olaoluwasalami in #860
• dev alias for frontend develop for AutoGen Studio by @brotsky in #1083
• Update Ecosystem.md by @bitsnaps in #1104
• Fix function call use case in website #1107 by @ekzhu in #1108
• openai base url default to None; read files when env vars exist; remove deprecated models by @sonichi in #1110
• Skip tests that depend on OpenAI via --skip-openai by @maxim-saplin in #1097
• move dotnet pipeline to the root of workflow by @LittleLittleCloud in #1116
• [Core] Improve config_list_from_json by @BeibinLi in #1026
• Minor fix groupchat by @kevin666aa in #904
• [AutoGen Studio] Allow multiline strings in chat input by @dragosMC91 in #1105
• Update tutorial.ipynb by @eltociear in #1119
• Fix minor typo in blog post by @PyroGenesis in #1122
• [Core] Throw an error when the OAI_CONFIG_LIST is missing. by @afourney in #1082
• Chore: upgrade base image to the latest Debian (Bullseye -> Bookworm) by @davorrunje in #1133
• fix typo agent_chat.md by @lordlinus in #1131
• update global.json in dotnet-run-openai-test-and-notebook.yml by @LittleLittleCloud in #1135
• [Core] Fix update_usage_summary when response doesn't have usage attribute by @kevin666aa in #1008
• Add a sample notebook for using AutoGen in Microsoft Fabric by @thinkall in #940
• bump version to 0.2.3 by @sonichi in #1140

New Contributors

• @danyrouh made their first contribution in #921
• @kar2905 made their first contribution in #992
• @dragosMC91 made their first contribution in #935
• @halr9000 made their first contribution in #931
• @cool-RR made their first contribution in #1007
• @KazooTTT made their first contribution in #1034
• @yuandong-tian made their first contribution in #1040
• @AdiG1123 made their first contribution in #1049
• @davorrunje made their first contribution in #1018
• @rlam3 made their first contribution in #824
• @ekzhu made their first contribution in #1071
• @hugo-syn made their first contribution in #1068
• @brotsky made their first contribution in #1083
• @bitsnaps made their first contribution in #1104
• @lordlinus made their first contribution in #1131

Full Changelog: v0.2.2...v0.2.3

v0.2.2

This release fixes a crucial breaking change introduced by for openai>=1.1.0 in dealing with function call. Please upgrade to this release if you are using 0.2.0 or 0.2.1.

Highlights

• New example notebook demoing video transcript translate with whisper.
• A preliminary gallery page on the website.
• Improvement to group chat: allow receiving message from agents that are not participants; allow a separate description field distinct from system message - expected to greatly improve the speaker selection quality; allow async group chat; disallow function/tool use for group chat manager.
• Improvement to agent builder: change default model and config loading process.
• Improvement to cost calculation when working with old cache.
• Issue template.

Special thanks to @kevin666aa for the fast fix of the breaking change. Thanks to @chengxuan233 @skzhang1 @afourney @tyler-suard-parker @kittyandrew @LinxinS97 @thinkall and all the other contributors!

What's Changed

• Update cache if cost not calculated by @kevin666aa in #903
• Add test for async group chat by @tyler-suard-parker in #902
• Adding a notebook demostrating video transcript translate with whisper in AutoGen by @chengxuan233 in #881
• Website Gallery page building by @skzhang1 in #850
• Allow GroupChat to receive messages from outside. by @afourney in #912
• Add issue forms by @thinkall in #918
• Add a comment to explain the role of UNSTRUCTURED_FORMATS by @thinkall in #909
• Change default model and config loading process in AgentBuilder by @LinxinS97 in #917
• Added an agent description field distinct from the system_message. by @afourney in #736
• Fix monitoring and observability gallery docs link by @DMCS20 in #926
• Make groupchat & generation async, actually by @kittyandrew in #543
• Enhanced Independent Coverage Reporting by @IANTHEREAL in #886
• Raise error when function as llm_config passed to GroupChatManager by @kevin666aa in #911
• bump version to 0.2.2 by @sonichi in #906
• Convert ChatCompletionMessage to Dict after completion by @kevin666aa in #791

New Contributors

• @tyler-suard-parker made their first contribution in #902
• @chengxuan233 made their first contribution in #881
• @DMCS20 made their first contribution in #926
• @kittyandrew made their first contribution in #543

Full Changelog: v0.2.1...v0.2.2