Changelog¶
v0.1.18¶
Features¶
- Now
agentdojo.functions_engine.Function
stores also the return type of the function and the full docstring.
Breaking Changes¶
- None
Bug Fixes¶
- None
v0.1.17¶
Features¶
- None
Breaking Changes¶
- None
Bug Fixes¶
- Fix breaking changes due to updated APIs from some LLM providers.
v0.1.16¶
Features¶
- Add the possibility to pass a custom pipeline element object as an LLM to a
PipelineConfig
.
Breaking Changes¶
- None
Bug Fixes¶
- Updated some checks in the banking suite.
v0.1.15¶
Features¶
- None
Breaking Changes¶
- None
Bug Fixes¶
- Fix the security checks for Workspace's injections tasks 4 and 5. The checks were not precise and injection tasks 4 would fail in some edge cases even if the model was executing it correctly.
v0.1.14¶
Features¶
None
Breaking Changes¶
Ellipsis
is not supported anymore as a type for function calls. This was done to simplify the code as any value can be passed as an argument to a function call, since is going to be discarded by the utility/security checks anyways.
Bug Fixes¶
- Fix issue with
FunctionCall
not being serializable when saving to JSON. This issue was introuced in v0.1.13 when theFunctionCall
object was changed from aTypedDict
v0.1.13¶
Features¶
- Add support for nested function calls in the
FunctionCall
object. This allows for more complex function calls: if any of theargs
in aFunctionCall
object is anotherFunctionCall
, the function call will be executed recursively.
Breaking Changes¶
None
Bug Fixes¶
None
v0.1.12¶
Features¶
None
Breaking Changes¶
- Change
FunctionCall
from aTypedDict
to a PydanticBaseModel
. This was done to allow for distinguishing function calls at runtime (in preparation) for supporting nested function calls. This affects users who are implementing new LLMs, new runtimes, and pipeline components that use function calls. To update your code, you will need to access thefunction
andargs
attributes of theFunctionCall
object instead of using dictionary access.
Bug Fixes¶
None