Content inputs to be processed according to the query.

[thiago-aixplain]

from aixplain.enums import Function, Supplier

from aixplain.factories import AgentFactory
from aixplain.modules.agent import ModelTool, PipelineTool

agent = AgentFactory.create(
    name="Test Agent 23423",
    tools=[
        ModelTool(
            function=Function.SPEECH_RECOGNITION,
            supplier=Supplier.MICROSOFT,
        ),
        ModelTool(
            function=Function.TRANSLATION,
            supplier=Supplier.MICROSOFT
        )
    ],
    description="Test Agent Description",
    supplier="aiXplain",
    version="1.0.0",
    # api_key="776baec37164e833039d030a5e48edc735e6cd038f04fa9642d39936e570905d",
    llm_id="6646261c6eb563165658bbb1",
)

response = agent.run(
    query="Translate the English texts within the links to Portuguese:",
    content_inputs=[
        "https://aixplain-platform-assets.s3.amazonaws.com/samples/en/CPAC1x2.txt",
        "https://aixplain-platform-assets.s3.amazonaws.com/samples/en/bestofyou.txt"
    ]
)

[tim-nelson]

1. Looks good except I think files would be clearer than content_inputs. Thoughts?
2. Does the user need to specify "within the links", or could it be possible to just say "Translate the English texts to Portuguese"?

[thiago-aixplain]

1. Looks good except I think files would be clearer than content_inputs. Thoughts?

The idea is that the user could set a text, url or file. Any other idea of name?

2. Does the user need to specify "within the links", or could it be possible to just say "Translate the English texts to Portuguese"?

I did not get this question. Can you please explain?

[tim-nelson]

1. My question rephrased
In your example, you write
"query": "Translate the English texts within the links to Portuguese."
Would the query still work if you only wrote
"query": "Translate the English texts to Portuguese."?
Would the Agent know where to get the text?

2. Additional content name
Ah, then how about we keep to just content. I don't think we need _inputs on the end, given we don't use that convention elsewhere. E.g. we write data and not data_inputs.

3. Using the data parameter
Have you considered using data? It may be overly complicated, but to be consistent with Models and Pipelines we'd include query, content and (maybe) session_id in a data parameter. (See the examples below.) Maybe there's a middle ground? I can see a use for data if users can include custom parameters that would be additional key-value pairs in the data dictionary:

• Session IDs for the other agents acting as tools to their 'main' agent?
• Changing configurations for some of the tools the agent uses? (Suppose we have a speech synthesis model that takes voice as an argument and we want to pass John-MALE in one case and a Jane-Female in another.)

What do you think?

MODELS
1. Pass single input.

model.run("Tell me a joke about dogs.")

model.run(data={
    "text": "Tell me a joke about dogs."
})

2. Pass multiple inputs.

model.run(
    {
        "text": "Tell me a joke about dogs.",
        "max_tokens": 10,
        "temperature": 0.5,
    }
)

AGENTS
1. Pass single input.

agent.run("What is the name of the driver who won the Formula One Championship in 2023? Answer in an English audio.")

agent.run(data={
    "query": "What is the name of the driver who won the Formula One Championship in 2023? Answer in an English audio."
})

2. Pass multiple inputs.

agent.run(data={
    "query": "Translate the English texts to Portuguese.",
    "content": [
        "This is a text string to translate.",
        "https://aixplain-platform-assets.s3.amazonaws.com/samples/en/CPAC1x2.txt",
        "/path/to/local/file.txt"
    ],
    "session_id": "2e275c1c-eb11-424a-824a-e0ae12c4b42d"
})

[thiago-aixplain]

1. My question rephrased In your example, you write "query": "Translate the English texts within the links to Portuguese." Would the query still work if you only wrote "query": "Translate the English texts to Portuguese."? Would the Agent know where to get the text?

Yes, it would know where to get the text in both instructions

2. Additional content name Ah, then how about we keep to just content. I don't think we need _inputs on the end, given we don't use that convention elsewhere. E.g. we write data and not data_inputs.

Nice. I will do the change accordingly.

3. Using the data parameter Have you considered using data? It may be overly complicated, but to be consistent with Models and Pipelines we'd include query, content and (maybe) session_id in a data parameter. (See the examples below.) Maybe there's a middle ground? I can see a use for data if users can include custom parameters that would be additional key-value pairs in the data dictionary:

• Session IDs for the other agents acting as tools to their 'main' agent?
• Changing configurations for some of the tools the agent uses? (Suppose we have a speech synthesis model that takes voice as an argument and we want to pass John-MALE in one case and a Jane-Female in another.)

What do you think?

MODELS 1. Pass single input.

model.run("Tell me a joke about dogs.")

model.run(data={
    "text": "Tell me a joke about dogs."
})

2. Pass multiple inputs.

model.run(
    {
        "text": "Tell me a joke about dogs.",
        "max_tokens": 10,
        "temperature": 0.5,
    }
)

AGENTS 1. Pass single input.

agent.run("What is the name of the driver who won the Formula One Championship in 2023? Answer in an English audio.")

agent.run(data={
    "query": "What is the name of the driver who won the Formula One Championship in 2023? Answer in an English audio."
})

2. Pass multiple inputs.

agent.run(data={
    "query": "Translate the English texts to Portuguese.",
    "content": [
        "This is a text string to translate.",
        "https://aixplain-platform-assets.s3.amazonaws.com/samples/en/CPAC1x2.txt",
        "/path/to/local/file.txt"
    ],
    "session_id": "2e275c1c-eb11-424a-824a-e0ae12c4b42d"
})

All right. I can also add a data field to be consistent with models and pipelines. The changes will be available by the end of the day.