New AUTOGEN 3.0 Update | Amazing UI
Summary
TLDRオートジェン・スタジオの最新バージョンに関するこの動画では、視覚的に大きく進化した新機能や改善点について解説しています。特に、ドラッグ&ドロップによるエージェントのワークフローの設定、プロファイラビューによるデバッグ機能の強化、さらには新しいコスト解析やツールの可視化機能が導入され、使いやすさが大幅に向上しています。また、再利用可能なテンプレートやワークフローの簡単なエクスポート・デプロイが可能となり、複雑なマルチエージェントシステムのプロトタイピングと管理が容易になりました。視聴者は、このバージョンに対する期待を高められる内容となっています。
Takeaways
- 🚀 Autogen Studioの新バージョンは大幅に改善され、前バージョンよりも多くの新機能が追加されています。
- 🖥️ ドラッグ&ドロップUIを使用して、エージェントワークフローを簡単に指定・評価できるようになりました。
- 📊 新機能として、メトリクスの可視化やメッセージのプロファイリングが導入され、エージェントの動作をより詳細に確認できるようになりました。
- 📂 Autogen Studioでは、複数のエージェント間でスキルやツールを簡単に割り当て可能になっています。
- 🛠️ デバッグがより容易になり、エージェントの失敗原因やワークフローコストが明確に表示されます。
- 🗂️ 再利用可能なエージェントコンポーネントのギャラリーがあり、テンプレートの再利用と共有が可能です。
- 👨💻 コード不要で、Pythonアプリケーションにワークフローをエクスポート・デプロイできる新機能が追加されています。
- 🎯 Playgroundビューを使用してタスク実行やデバッグができ、ワークフローを迅速にプロトタイプ化できます。
- 🤖 Autogen Studioは、マルチエージェントシステムのデザインと最適化に向けた新しいツールを提供しています。
- 📈 プロファイラビューで、エージェントの動作と成果物の生成がすべて可視化され、複雑なシステムでも効率的に管理できます。
Q & A
Autogen Studioとは何ですか?
-Autogen Studioは、マルチエージェントワークフローのプロトタイピング、デバッグ、および評価を迅速に行うためのノーコード開発ツールです。直感的なドラッグ&ドロップUIを提供し、エージェントのワークフローを簡単に設計できます。
Autogen Studioの最新バージョンで何が改善されましたか?
-最新バージョンでは、ドラッグ&ドロップによるUIの大幅な改善、プロファイリング機能の追加、エージェント間のメッセージやアクションの可視化、コストの追跡などが強化されました。
新バージョンのAutogen Studioでどのような機能が追加されましたか?
-エージェント間のメッセージやアクションの可視化、タスクの進行状況の監視、ワークフローのコスト計算、ツールの成功・失敗を追跡するプロファイリング機能などが追加されました。
エージェントワークフローを作成する際にどのようなモデルやツールを使用できますか?
-GPT-4 Turboなどのモデルを選択して使用することができ、コンテンツ生成、画像生成、ウェブ検索などのスキルをエージェントに追加することが可能です。
Autogen Studioの新しいプロファイリング機能とは何ですか?
-プロファイリング機能は、エージェントのメッセージ、アクション、コスト、成功したツールや失敗したツールを可視化し、ワークフローのデバッグや最適化をサポートします。
Autogen Studioはどのようにワークフローのエクスポートをサポートしていますか?
-Autogen Studioでは、ワークフローをJSON形式でエクスポートでき、Pythonアプリケーションで実行したり、Dockerコンテナとしてデプロイしたりすることが可能です。
Autogen Studioの最新バージョンでの課題は何ですか?
-マルチエージェントワークフローの設計は複雑であり、特にエージェントの失敗や問題点を特定するのが困難ですが、新しいプロファイリングツールでこれを解決しようとしています。
Autogen Studioで作成されたワークフローのデバッグ方法はどのように改善されましたか?
-デバッグ機能が強化され、エージェントが実行するタスクの進行状況や生成されたファイル(画像、コード、ドキュメントなど)をリアルタイムで観察できるようになりました。
Autogen Studioの『Playground View』とは何ですか?
-Playground Viewは、タスクの実行やワークフローのデバッグを行うためのインターフェースで、エージェントの動作や生成物を視覚的に確認できます。
Autogen Studioのエージェントはどのような用途に使用できますか?
-エージェントは、ユーザープロキシ、コンテンツ生成、画像生成、QAエージェントなど様々な役割を持つことができ、グループチャットやドキュメント作成などのタスクに応用可能です。
Outlines
🛠️ オートジェンスタジオの大幅な改善
オートジェンスタジオの新バージョンについての紹介。このツールはマルチエージェントワークフローのプロトタイプ作成、デバッグ、評価を簡単に行えるノーコード開発ツールです。新バージョンはドラッグ&ドロップのインターフェースを採用し、直感的な操作性が大幅に向上しています。オートジェンスタジオの目標は、より複雑なシステムを管理しやすくすることです。エージェントを使ったプロセスの効率化が期待されており、プロファイリング機能や可視化が追加され、ワークフロー全体のコストやエージェント間のやりとりが一目でわかるようになっています。
🎛️ プレイグラウンドビューとワークフローのデバッグ
プレイグラウンドビューは、タスク実行やワークフローのデバッグに使われる新機能で、エージェントの行動を簡単に追跡できます。このビューでは、セッションを作成し、ワークフローを割り当て、タスクを実行できます。ワークフローは、単発タスクやマルチターンのタスクとして動作します。デバッグをサポートするために、タスクの進行状況やエージェントのメッセージがリアルタイムで表示され、生成された成果物が確認できます。また、新しいインターフェースが導入され、ユーザーはより多くの情報を容易に取得できるようになっています。
📘 新しい機能とエクスポートの改善
オートジェンスタジオでは、ワークフローをJSONファイルとしてエクスポートでき、Pythonアプリケーションに直接組み込むことが可能になりました。また、エージェント間のやり取りの可視化やコスト計算の詳細な表示など、全体的な操作性が大幅に改善されています。デバッグやシステムの挙動を理解するためのツールが強化され、複数エージェントによるワークフローの作成が簡単になりました。このバージョンでは、エージェントを迅速にプロトタイピングできるだけでなく、将来の研究やデザインパターンの向上に向けた方向性も示されています。
Mindmap
Keywords
💡Autogen Studio
💡マルチエージェントワークフロー
💡ドラッグアンドドロップUI
💡プロファイリング機能
💡タスクデバッグ
💡エージェントテンプレート
💡ノーコード開発
💡エージェント間のメッセージ可視化
💡ワークフローのエクスポート
💡GitHubイシューの可視化
Highlights
Huge improvement from the previous version of Autogen Studio.
Introduction of new features with a more intuitive drag-and-drop UI.
Built off the Autogen framework developed over the past year.
No-code developer tool for rapidly prototyping, debugging, and evaluating multi-agent workflows.
Provides reusable agent components and new profiling capabilities.
Significant visual changes to the interface with added tools for debugging.
Allows drag-and-drop functionality to assign tasks and skills to agents.
New feature that tracks costs, tools executed by agents, and workflow outputs.
Debugging multi-agent workflows with improved observability.
Supports long-term memory and vector databases integration.
Playground view for task execution and workflow debugging.
Supports reusable templates for agents like assistant agents and group chat agents.
New gallery view for sharing and reusing templates.
Support for exporting workflows as JSON config files.
Workflows can be deployed in Python applications or wrapped in Docker containers.
Transcripts
I have to say this looks absolutely
amazing this is such a huge improvement
from the previous version of Auden
Studio it seems that it's been a little
while since we really heard something
about autogen Studio or at least
anything significant until now I'm
really excited to go over this paper
with you and see what all there is with
the new version so let's get into it
okay so this is the new paper and as you
can see just from this screenshot that
is going to look so much different than
what it currently does mainly because
there are new features but let's just go
over if you're new to aen Studio and
what it is let's just go over what this
paper is saying and we'll get to these
features the idea of autogen studio is
that it's built off the autogen
framework which has developed over the
past year into something wonderful
however with that comes complexity right
the more you're adding onto the
framework to do more the more complex it
can get and this is where autogen Studio
comes into play so to address this
situation they present aen Studio A no
code developer tool for rapidly
prototyping debugging and evalu evalua
these multi-agent workflows it provides
an intuitive drag and drop UI for agent
workflow specification interacting the
evalu interactive valuation and
debugging of workflows and a gallery of
reable agent components and if you see
with a screenshot right this is nothing
at all what autogen studio used to look
like this is now a drag and drop see
there's an agent a there's an agent B
look this is the um the agent a is a
user proxy and it is executing code B is
a book generation group chat manager and
inside of here there are three different
agents and it looks like they're just
dragging and dropping the model so right
here is GPT 4 Turbo they're just
dragging dropping the models and adding
skills right so this one's getting a
Content agent is getting web search and
image agent is getting an image
generator skill or tool and I have to
say that is an this this screenshot
right here is an amazing Improvement of
what it used to look like now that I see
this I'm loving this but what's
interesting is this introduction of
profiling capabilities with
visualizations of messages and actions
and well this metrics right here metrics
is the main thing here is it looks like
they're they're going to have a UI to
show you um you know what the costs are
what tools were executed by which agents
what are the output for each of the
agent and so forth so if something if
something failed you can now see where
it was and then also how much your
workflow cost you and it is hard to
debug some of these mod multi- agent
workflow especially the more complex
they get right if something fails yeah I
mean you can have you can have logging
and print statements and what or however
you do it but still to see exactly how
the agent that specific agent failed and
why it failed it can be difficult so
looks like here they have a related work
section where they talk about uh agentic
implementation such as react which is
the reason and acting for llms they have
L chain for Harrison Chase uh and so
forth this is just kind of um saying
about what some people have done and
some limitations to have better
architectures for multi-agents all right
and I'm just going to go over these
Concepts really quick just in case you
know some of you a lot of you may
already know these but these are like
going to be the main components of
autogen Studio which is why they're
highlighting here right it's this whole
drag and drop experience now so with the
with the model you know this is you're
going to be able to pick the model to
generate whatever it is image text uh
skills SL tools you know you're going to
I think you can it looks like you can
still if you want to code your own tool
and then you can probably drag and drop
that onto whichever agent and it looks
like they may be implementing memory so
shortterm you know this is the normal or
long-term right so they can we can have
Vector databases which would be nice to
have um with autogen Studio I will'll
look further in the research paper see
if they're going to implement that but
that would be really nice to have and of
course the agent and then you can looks
like you can combine um like workflows
together to create these groups of
Agents or like a group chat or you can
have a single agent or whatever it is
you know and scrolling down a little bit
here what they're kind of just getting
at is that you know there are
limitations um whenever you have all of
these together that like you know it can
be hard to basically have reusable
templates or be able to bootstrap some
workflow you know it can be hard so they
try to get rid of that just by providing
this visual interface and they're going
to and they're going to kind of hone in
on the design goals which is to have
rapid prototyping developer tooling so
these tools um you know we can help
understand what's happening with agent
behaviors and this is going to help us
you know facilitate the Improvement of
these systems by understanding quicker
and easier what's happening and they
have reusable templates now this they
had a gallery section right that present
a gallery of reusable sharable templates
it could be just still in the same
Gallery section but you know that was a
little different so hopefully it's
improved with um with this next version
of autogen studio so they have a
playground view which is going to be
used for tax task execution workflow
debugging and options to export and
deploy the gallery view fac facilitates
the ReUse and sharing of the templates
which is what I just talked about now
this kind of confirms it and then here
they're you know they're just going over
some templates they have for um for
agents like assistant agent a group chat
uh agent and so forth right this is
that's all pretty standard stuff with
autogen saying workflows can be tested
in the build view which you know we
could we did already um in the previous
version I believe or was made the
playground view but more systematically
explored within the playground view okay
so the playground view allows users to
create sessions which again we were able
to do in the previous version attach
workflows to the session and then run
tasks so either single shot or
multi-turn basically group chats or just
a onetoone um or just an Interactive in
with one agent aen Studio provides two
features to support debugging first it
provides an observe view whereas task
progress messages and actions performed
by agents are streamed to the interface
and all generated artifacts are
displayed so files uh such as images
code and documents and now let's go
ahead and look at the new interface and
here all they're really saying is the
backend API the frontend web UI and kind
of what it looks like but look at just
look at how this looks this looks abs
absolutely amazing this is a huge
upgrade from the way it used to look and
the amount of information that you can
gather from this so it looks like
they're in the playground session right
now and so they asked you to create a
children's uh PDF book with four pages
each describing the weather in Seattle
so then the it says the agents have
completed the task right so this again
this was uh part of the group chat
manager the all these agents so it's
saying the agents have completed the
task and then the children's PDF book
titled weather in Seattle uh has been
created with descriptions and everything
the book should now be available as
Seattle weather children book.pdf on
your system you can now View and ensure
that it meets your expectations if
everything looks good that completes our
task if you need any further assistance
or modifications please let me know so
it gave a total of seven files so I
guess just uh like it created the
children's book compiled with seven
images and and on the right side here is
talking about the profiler which which
now it says the group chat manager right
these the amounts of tokens that were
generated it cost 15 cents the uh then
it breaks it down you know so the
content this is probably that the agent
that was creating the images the user
proxy or no I'm sorry here's the image
generator uh you know this is only 1
cent and the content is 12 cents so I
can't imagine the image generator only
being one penny um but then they have a
QA um agent so this is it this is
amazing right the way this looks and
then um how many the total messages
between all of the agents so it looks
like maybe they were uh between the user
proxy and the group chat there were
probably you know here's about 17 and
then the user proxy probably had like
six so you know a fair amount of you
know it's been a fair amount of messages
sent from everybody and then tool and
then here's all the tool calls right it
says which ones were successful and
which ones failed so you know it looks
like uh it some of these failed and the
thing about this is it's telling you and
giving you the metrics for what has
happened with this right that's the
amazing part about this and now you can
deploy workflows auten Studio does
enable users to export workflows as a
Json config file which they kind of did
have before but the what they what
you're able to do now is an exported
workflow uh can be put into any python
application right so if you just have
some python code or you know how to do
that then you can execute this uh as an
API endpoint using the autogen studio
CLI or wrapped in a Docker container
right so they you know you just you
don't have to actually um pip install
autogen you can just do from autogen
Studio import workflow manager you can
give it the Json file that you exported
and then it can run it right and then um
you know this is basically the workflow
that you had created so you can say what
is the height of the Eiffel Tower for
instance this is probably just a simple
onetoone agent interaction but I mean
this is such an improvement of what you
were able to do and then now finally
we're going to come down to the usage
and evaluation and what it looks like
here is the this is the autogen studio
uh GitHub issue visualization right so
they kind of grouped uh everything
together and you know said the plot of
GitHub issues from the autogen studio
repo user feedback range from support
with workflow uh authoring tools to
General installation so they kind of
grouped everything together and probably
tried to figure out how can we solve the
problem for everybody that was kind of
that we were kind of having with autogen
Studio because at the end of the day it
was kind of I don't want to say bare it
did get you it allowed you to prototype
um agents really quickly but even like
after you did that right there wasn't
much it felt like there wasn't much more
you can do with it like something was
kind of missing and it looks like
they're trying to fix all of that with
this new version of autogen and then
we're they're talking about design
patterns and research directions um so
you know we kind of already talked about
the Define and compose workflow this is
allows users to author workflows by
basically dragging and dropping
everything together to create the
multiagent workflow and then the
debugging and sensemaking tools um this
is what we've also talked about like the
profiler views all the metrics these
help you debug interpret and uh
understand the behavior and output of
your systems using the no code method
right again because this is no code you
don't have the typical logging methods
which aren't easy with multi-agents so
they came up with something so that we
could have we could see more of what's
Happening under the hood you know and
then they have export and deployment
there's uh collaboration and sharing you
know and how do we understand uh the
multi-agent system designs you know and
then optimizing the multi-agent systems
you know so this is kind of the the
questions that they're asking that they
want to improve upon in the future well
this is autogen 3.0 and I would actually
say that because this is such an
improvement from the previous version I
am really excited for this I hope you
are excited too and in the meantime I
have a couple courses you can take
related to Ai and one of them is autogen
thank you for watching I'll see you next
video
5.0 / 5 (0 votes)