mirror of
https://github.com/microsoft/semantic-kernel.git
synced 2026-10-12 02:19:06 +00:00
Evan Mattson
afb1e15724
### Motivation and Context
This PR relaxes the API signatures of all agent invocation methods
(`get_response`, `invoke`, `invoke_stream`) by changing the `messages`
parameter from keyword-only to positional-or-keyword. This enhances
ergonomics and Pythonic usability by allowing callers to pass messages
either positionally or as a keyword argument:
```python
agent.get_response("Why is the sky blue?")
agent.get_response(messages="Why is the sky blue?")
```
With our current use of **kwargs, it is possible to explicitly pass None
for messages (like `agent.get_response(messages=None)`). However, this
does not feel idiomatic in Python. If a caller wishes to invoke the
agent on an existing thread (which already contains all relevant
context), it is more natural and Pythonic to simply write:
```python
response = await agent.get_response()
```
This change makes the API more user-friendly, and removes unnecessary
verbosity when no new messages need to be provided.
Impact
- Backward compatible for all previous usage.
- Enables concise and intuitive API calls for primary message arguments.
- Subclasses overriding the prior keyword-only signature will continue
to work at runtime, but static type-checkers may require updating their
signatures to match the new base definition.
- No runtime breaking changes are expected.
| Scenario | Breaks at runtime? | Breaks at type-check? | Notes |
|----------------------------------|----------------------|-------------------------|--------------------------------------------------|
| Subclass w/keyword-only | No | Yes | Type checker error, but runtime
OK |
| Subclass matching base | No | No | Fine |
| Calls expecting keyword-only | No | No | Fine |
| Calls expecting positional | No (now allowed) | No | Improved
ergonomics |
| Decorators relying on kw-only | No | No (unless strict sig) | Only if
decorator is brittle, rare |
| Code generation/docs | No | No | Needs updating docs, not an actual
code break |
| Reflection-based invocation | No | No | Slight difference in behavior,
but not breaking |
Documentation and usage examples will be updated to reflect the improved
invocation style.
<!-- Thank you for your contribution to the semantic-kernel repo!
Please help reviewers and future users, providing the following
information:
1. Why is this change required?
2. What problem does it solve?
3. What scenario does it contribute to?
4. If it fixes an open issue, please link to the issue here.
-->
### Description
Relaxes the API signatures of all agent invocation methods.
<!-- Describe your changes, the overall approach, the underlying design.
These notes will help understanding how your code works. Thanks! -->
### Contribution Checklist
<!-- Before submitting this PR, please make sure: -->
- [X] The code builds clean without any errors or warnings
- [X] The PR follows the [SK Contribution
Guidelines](https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md)
and the [pre-submission formatting
script](https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md#development-scripts)
raises no violations
- [X] All unit tests pass, and I have added new tests where possible
- [X] I didn't break anyone 😄
AutoGen Conversable Agent (v0.2.X)
Semantic Kernel Python supports running AutoGen Conversable Agents provided in the 0.2.X package.
Limitations
Currently, there are some limitations to note:
- AutoGen Conversable Agents in Semantic Kernel run asynchronously and do not support streaming of agent inputs or responses.
- The
AutoGenConversableAgentin Semantic Kernel Python cannot be configured as part of a Semantic KernelAgentGroupChat. As we progress towards GA for our agent group chat patterns, we will explore ways to integrate AutoGen agents into a Semantic Kernel group chat scenario.
Installation
Install the semantic-kernel package with the autogen extra:
pip install semantic-kernel[autogen]
For an example of how to integrate an AutoGen Conversable Agent using the Semantic Kernel Agent abstraction, please refer to autogen_conversable_agent_simple_convo.py.