This finishes the fix for the race condition between opening a
GrpcWorkerConnection and registering agent types on that worker. Now,
instead of failing to register, we return from the call (with the
expectation that we will finish registration as we set up the
connection)
Part 1: #5494
Part 2: #5514
---------
Co-authored-by: Ryan Sweet <rysweet@microsoft.com>
Fixed a wrong class name style in the document and removed trailing whitespace.
Co-authored-by: Wei Jen Lu <weijenlu@microsoft.com>
Co-authored-by: Eric Zhu <ekzhu@users.noreply.github.com>
<!-- Thank you for your contribution! Please review
https://microsoft.github.io/autogen/docs/Contribute before opening a
pull request. -->
<!-- Please add a reviewer to the assignee section when you create a PR.
If you don't have the access to it, we will shortly find a reviewer and
assign them to your PR. -->
Adds the following blurb to the model clients docs
## API Keys From Environment Variables
In the examples above, we show that you can provide the API key through
the `api_key` argument. Importantly, the OpenAI and Azure OpenAI clients
use the [openai
package](3f8d8205ae/src/openai/__init__.py (L260)),
which will automatically read an api key from the environment variable
if one is not provided.
- For OpenAI, you can set the `OPENAI_API_KEY` environment variable.
- For Azure OpenAI, you can set the `AZURE_OPENAI_API_KEY` environment
variable.
This is a good practice to explore, as it avoids including sensitive api
keys in your code.
## Why are these changes needed?
<!-- Please give a short summary of the change and the problem this
solves. -->
## Related issue number
<!-- For example: "Closes #1234" -->
## Checks
- [ ] I've included any doc changes needed for
https://microsoft.github.io/autogen/. See
https://microsoft.github.io/autogen/docs/Contribute#documentation to
build and test documentation locally.
- [ ] I've added tests (if relevant) corresponding to the changes
introduced in this PR.
- [ ] I've made sure all auto checks have passed.
Right now we rely on opening the channel to associate a ClientId with an
entry on the gateway side. This causes a race when the channel is being
opened in the background while an RPC (e.g. MyAgent.register()) is
invoked.
If the RPC is processed first, the gateway rejects it due to "invalid"
clientId.
This fix makes this condition less likely to trigger, but there is still
a piece of the puzzle that needs to be solved on the Gateway side.
<!-- Thank you for your contribution! Please review
https://microsoft.github.io/autogen/docs/Contribute before opening a
pull request. -->
<!-- Please add a reviewer to the assignee section when you create a PR.
If you don't have the access to it, we will shortly find a reviewer and
assign them to your PR. -->
## Why are these changes needed?
It is useful to rapidly validate any changes to a team structure as
teams are built either via drag and drop or by modifying the underlying
spec
You can now “validate” your team. The key ideas are as follows
- Each team is based on some Component Config specification which is a
pedantic model underneath.
- Validation is 3 pronged based on a ValidatorService class
- Data model validation (validate component schema)
- Instantiation validation (validate component can be instantiated)
- Provider validation, component_type validation (validate that provider
exists and can be imported)
- UX: each time a component is **loaded or saved**, it is automatically
validated and any errors shown (via a server endpoint). This way, the
developer immediately knows if updates to the configuration is wrong or
has errors.
> Note: this is different from actually running the component against a
task. Currently you can run the entire team. In a separate PR we will
implement ability to run/test other components.
<img width="1360" alt="image"
src="https://github.com/user-attachments/assets/d61095b7-0b07-463a-b4b2-5c50ded750f6"
/>
<img width="1368" alt="image"
src="https://github.com/user-attachments/assets/09a1677e-76e8-44a4-9749-15c27457efbb"
/>
<!-- Please give a short summary of the change and the problem this
solves. -->
## Related issue number
Closes#4616
<!-- For example: "Closes #1234" -->
## Checks
- [ ] I've included any doc changes needed for
https://microsoft.github.io/autogen/. See
https://microsoft.github.io/autogen/docs/Contribute#documentation to
build and test documentation locally.
- [ ] I've added tests (if relevant) corresponding to the changes
introduced in this PR.
- [ ] I've made sure all auto checks have passed.
Closes#5297
---------
Co-authored-by: Ryan Sweet <rysweet@microsoft.com>
Co-authored-by: Jacob Alber <jaalber@microsoft.com>
Co-authored-by: Jacob Alber <jacob.alber@microsoft.com>
Co-authored-by: Eric Zhu <ekzhu@users.noreply.github.com>
Add the following additional configuration options to
DockerCommandLineCodeExectutor:
- **extra_volumes** (Optional[Dict[str, Dict[str, str]]], optional): A
dictionary of extra volumes (beyond the work_dir) to mount to the
container. Defaults to None.
- **extra_hosts** (Optional[Dict[str, str]], optional): A dictionary of
host mappings to add to the container. (See Docker docs on extra_hosts)
Defaults to None.
- **init_command** (Optional[str], optional): A shell command to run
before each shell operation execution. Defaults to None.
## Why are these changes needed?
See linked issue below.
In summary: Enable the agents to:
- work with a richer set of sys admin tools on top of code execution
- add support for a 'project' directory the agents can interact on
that's accessible by bash tools and custom scripts
## Related issue number
Closes#5363
## Checks
- [x] I've included any doc changes needed for
https://microsoft.github.io/autogen/. See
https://microsoft.github.io/autogen/docs/Contribute#documentation to
build and test documentation locally.
- [x] I've added tests (if relevant) corresponding to the changes
introduced in this PR.
- [x] I've made sure all auto checks have passed.
This PR improves documentation on custom agents
- Shows example on how to create a custom agent that directly uses a
model client. In this case an example of a GeminiAssistantAgent that
directly uses the Gemini SDK model client.
- Shows that that CustomAgent can be easily added to any agentchat team
- Shows how the same CustomAgent can be made declarative by inheriting
the Component interface and implementing the required methods.
Closes#5450
## Why are these changes needed?
These changes are needed because currently there's no generic way to add
`tools` to autogen studio workflows using the existing DSL and schema
other than inline python.
This API will be quite verbose, and lacks a discovery mechanism, but it
unlocks a lot of programmatic use-cases.
## Related issue number
https://github.com/microsoft/autogen/issues/5170
Co-authored-by: Victor Dibia <victordibia@microsoft.com>
Co-authored-by: Eric Zhu <ekzhu@users.noreply.github.com>
Have seen discussion on Discord regarding confusion about multi-modal
support in v0.4. This change adds a small note on how to use multi-modal
messages with agents.
<!-- Thank you for your contribution! Please review
https://microsoft.github.io/autogen/docs/Contribute before opening a
pull request. -->
<!-- Please add a reviewer to the assignee section when you create a PR.
If you don't have the access to it, we will shortly find a reviewer and
assign them to your PR. -->
## Why are these changes needed?
<!-- Please give a short summary of the change and the problem this
solves. -->
The current implementation tries to recreate the metadata but it does it
in an incomplete way. This PR uses SK built-in kernel from function
decorator to infer the callable from the `run_json` and makes better use
of the pydantic schemas for the input and output to infer the schema of
the kernel function.
## Related issue number
<!-- For example: "Closes #1234" -->
Closes#5458
## Checks
- [ ] I've included any doc changes needed for
https://microsoft.github.io/autogen/. See
https://microsoft.github.io/autogen/docs/Contribute#documentation to
build and test documentation locally.
- [ ] I've added tests (if relevant) corresponding to the changes
introduced in this PR.
- [ ] I've made sure all auto checks have passed.
---------
Co-authored-by: Leonardo Pinheiro <lpinheiro@microsoft.com>
Co-authored-by: Eric Zhu <ekzhu@users.noreply.github.com>
<!-- Thank you for your contribution! Please review
https://microsoft.github.io/autogen/docs/Contribute before opening a
pull request. -->
<!-- Please add a reviewer to the assignee section when you create a PR.
If you don't have the access to it, we will shortly find a reviewer and
assign them to your PR. -->
Add clear notes on how to specify api key and modifying specifications
in AGS.
Add diagrams explaining how to switch between visual builder and JSON
mode
## Why are these changes needed?
<!-- Please give a short summary of the change and the problem this
solves. -->
## Related issue number
<!-- For example: "Closes #1234" -->
@nour-bouzid
## Checks
- [ ] I've included any doc changes needed for
https://microsoft.github.io/autogen/. See
https://microsoft.github.io/autogen/docs/Contribute#documentation to
build and test documentation locally.
- [ ] I've added tests (if relevant) corresponding to the changes
introduced in this PR.
- [ ] I've made sure all auto checks have passed.
<!-- Thank you for your contribution! Please review
https://microsoft.github.io/autogen/docs/Contribute before opening a
pull request. -->
<!-- Please add a reviewer to the assignee section when you create a PR.
If you don't have the access to it, we will shortly find a reviewer and
assign them to your PR. -->
It is often helpful to inspect the raw request and response to/from an
LLM as agents act.
This PR, does the following:
- Updates TeamManager to yield LLMCallEvents from core library
- Run in an async background task to listen for LLMCallEvent JIT style
when a team is run
- Add events to an async queue and
- yield those events in addition to whatever actual agentchat
team.run_stream yields.
- Update the AGS UI to show those LLMCallEvents in the messages section
as a team runs
- Add settings panel to show/hide llm call events in messages.
- Minor updates to default team
<img width="1539" alt="image"
src="https://github.com/user-attachments/assets/bfbb19fe-3560-4faa-b600-7dd244e0e974"
/>
<img width="1554" alt="image"
src="https://github.com/user-attachments/assets/775624f5-ba83-46e8-81ff-512bfeed6bab"
/>
<img width="1538" alt="image"
src="https://github.com/user-attachments/assets/3becbf85-b75a-4506-adb7-e5575ebbaec4"
/>
## Why are these changes needed?
<!-- Please give a short summary of the change and the problem this
solves. -->
## Related issue number
<!-- For example: "Closes #1234" -->
Closes#5440
## Checks
- [ ] I've included any doc changes needed for
https://microsoft.github.io/autogen/. See
https://microsoft.github.io/autogen/docs/Contribute#documentation to
build and test documentation locally.
- [ ] I've added tests (if relevant) corresponding to the changes
introduced in this PR.
- [ ] I've made sure all auto checks have passed.
<!-- Thank you for your contribution! Please review
https://microsoft.github.io/autogen/docs/Contribute before opening a
pull request. -->
<!-- Please add a reviewer to the assignee section when you create a PR.
If you don't have the access to it, we will shortly find a reviewer and
assign them to your PR. -->
## Why are these changes needed?
<!-- Please give a short summary of the change and the problem this
solves. -->
The current stream processing of SK model adapter returns on the first
function call chunk but this behavior is incorrect end ends up returning
with an incomplete function call. The observed behavior is that the
function name and arguments are split into different chunks and this
update correctly processes the chunks in this way.
## Related issue number
<!-- For example: "Closes #1234" -->
Fixes the reply in #5420
## Checks
- [ ] I've included any doc changes needed for
https://microsoft.github.io/autogen/. See
https://microsoft.github.io/autogen/docs/Contribute#documentation to
build and test documentation locally.
- [ ] I've added tests (if relevant) corresponding to the changes
introduced in this PR.
- [ ] I've made sure all auto checks have passed.
---------
Co-authored-by: Leonardo Pinheiro <lpinheiro@microsoft.com>
Don't throw an exception when model makes a mistake. Use retries, and if
not succeeding after a fixed attempts, fall back to the previous sepaker
if available, or the first participant.
Resolves#5453
<!-- Thank you for your contribution! Please review
https://microsoft.github.io/autogen/docs/Contribute before opening a
pull request. -->
<!-- Please add a reviewer to the assignee section when you create a PR.
If you don't have the access to it, we will shortly find a reviewer and
assign them to your PR. -->
## Why are these changes needed?
<!-- Please give a short summary of the change and the problem this
solves. -->
Semantic kernel prepends the plugin name to the tool name when passing
the tools to model clients and this is causing a mismatch between tool
names in SK and the AssistantAgent. Since plugin names are optional, we
have opted to remove it.
## Related issue number
<!-- For example: "Closes #1234" -->
Closes#5420
## Checks
- [ ] I've included any doc changes needed for
https://microsoft.github.io/autogen/. See
https://microsoft.github.io/autogen/docs/Contribute#documentation to
build and test documentation locally.
- [ ] I've added tests (if relevant) corresponding to the changes
introduced in this PR.
- [ ] I've made sure all auto checks have passed.
---------
Co-authored-by: Leonardo Pinheiro <lpinheiro@microsoft.com>
Restoring the grpc + Orleans server into the project
## Why are these changes needed?
This is the distributed agent runtime for .NET that can manage routing
messages amongst a fleet of grpc agent runtimes.
## Related issue number
## Checks
- [ ] I've included any doc changes needed for
https://microsoft.github.io/autogen/. See
https://microsoft.github.io/autogen/docs/Contribute#documentation to
build and test documentation locally.
- [x] I've added tests (if relevant) corresponding to the changes
introduced in this PR.
- [x] I've made sure all auto checks have passed.
---------
Co-authored-by: Jack Gerrits <jack@jackgerrits.com>
Co-authored-by: Jacob Alber <jaalber@microsoft.com>
<!-- Thank you for your contribution! Please review
https://microsoft.github.io/autogen/docs/Contribute before opening a
pull request. -->
<!-- Please add a reviewer to the assignee section when you create a PR.
If you don't have the access to it, we will shortly find a reviewer and
assign them to your PR. -->
## Why are these changes needed?
- Add ability to test teams in Team Builder view
- Update Gallery (add deep research default team, fix bug with gallery
serialization)
- UI fixes
- improve drag drop component experience
- improve new session experience (single click rather than 3 clicks to
create a session)
- fix bug with stop reason not being shown in some cases
<img width="1738" alt="Image"
src="https://github.com/user-attachments/assets/4b895df2-3bad-474e-bec6-4fbcbf1c4346"
/>
<img width="1761" alt="Image"
src="https://github.com/user-attachments/assets/65f52eb9-e926-4168-88fb-d2496c159474"
/>
<!-- Please give a short summary of the change and the problem this
solves. -->
## Related issue number
Closes#5392
<!-- For example: "Closes #1234" -->
## Checks
- [ ] I've included any doc changes needed for
https://microsoft.github.io/autogen/. See
https://microsoft.github.io/autogen/docs/Contribute#documentation to
build and test documentation locally.
- [ ] I've added tests (if relevant) corresponding to the changes
introduced in this PR.
- [ ] I've made sure all auto checks have passed.
Introduce a sample chat application using AgentChat and FastAPI,
demonstrating single-agent and team chat functionalities, along with
state persistence and conversation history management.
Resolves#5423
---------
Co-authored-by: Victor Dibia <victor.dibia@gmail.com>
Co-authored-by: Victor Dibia <victordibia@microsoft.com>
Presently MagenticOne and the m1 CLI use the LocalCommandLineExecutor
(presumably copied from the agbench code, which already runs in Docker).
This pr defaults m1 to Docker, and adds a code_executor parameter to
MagenticOne, which defaults to local for now to maintain backward
compatibility -- but this behavior is immediately deprecated.
- Updated HumanEval template to use AgentChat
- Update templates to use config.yaml for model and other configuration
- Read environment from ENV.yaml (ENV.json still supported but
deprecated)
- Temporarily removed WebArena and AssistantBench. Neither had viable
Templates after `autogen_magentic_one` was removed. Templates need to be
update to AgentChat (in a future PR, but this PR is getting big enough
already)
A series of changes to the
`python/packages/autogen-ext/src/autogen_ext/agents/web_surfer/_multimodal_web_surfer.py`
file have been made to better support smaller models.
This includes changes to the prompts, state descriptions, and ordering
of messages.
Regression tasks with OpenAI models shows no change in GAIA scores,
while scores for Llama are significantly improved.
Get's SelectorGroupChat working for llama by:
1. Using a UserMessage rather than a SystemMessage
2. Normalizing how roles are presented (one agent per line)
3. Normalizing how the transcript is constructed (a blank line between
every message)
This PR fixes:
A prompting bug when no control had focus.
Awkward prompt phrasing.
Renamed page_down to scroll_down to better match other prompting and
agent descriptions.
Some agent descriptions were split over multiple lines in the M1
orchestrator. This PR ensures that each description appears on one, and
only one, line. This makes it easier for smaller models to understand.
