TXTSearchTool with ollama

[fre391]

Search is NOT limited to given txt file.

`
from crewai_tools import TXTSearchTool

txt_search_tool = TXTSearchTool(
txt="kunst.txt",
config=dict(
llm=dict(
provider="ollama",
config=dict(
model="llama3.1",
),
),
embedder=dict(
provider="ollama",
config=dict(
model="mxbai-embed-large",
),
),
)
)
`

No error message and at the end the best result is shown, but in between (Verbose) it will also show snippets from other sources, f.e. of a PDF which was searched by XMLSearchTool earlier by using a seperate script....

[iNLyze]

@fre391, nice solution, though I am not yet getting it to work. I am trying to apply the text tool to a local file like so:

tool_txt_search = TXTSearchTool(
    txt=path/'nonsense.txt',
    config=config, 
    verbose=True
)

My config:

config=dict(
    llm=dict(
        provider="ollama",  # Change this to your LLM provider
        config=dict(
            model="llama3.1",  # Specify the model you want to use
        ),
    ),
    embedder=dict(
        provider="ollama",  # Change this to your LLM provider
        config=dict(
            model="mxbai-embed-large",  # Specify the embedding model 
        ),
    ),
)

Running this from jupyter currently (on a server via ssh). Unfortunately, the instantiation never seems to finish. I am not seeing any text output and nothing in journalctl -u ollama either.

How long does it take you to instantiate the tool? Any idea, why it might get stuck?

[kspviswa]

+1 . Also facing the same issue.

[siddas27]

I got everything working with this:

os.environ['OPENAI_API_BASE'] = 'http://localhost:11434'
os.environ['OPENAI_MODEL_NAME'] = 'ollama/llama3.2'
os.environ['OPENAI_API_KEY'] = 'NA'

Note: Using this will make this your default llm, no need to configure everything separately

[kspviswa]

I got everything working with this:

os.environ['OPENAI_API_BASE'] = 'http://localhost:11434'
os.environ['OPENAI_MODEL_NAME'] = 'ollama/llama3.2'
os.environ['OPENAI_API_KEY'] = 'NA'

Note: Using this will make this your default llm, no need to configure everything separately

Sorry not working.

/.venv/lib/python3.12/site-packages/crewai/agent.py", line 161, in post_init_setup
    if env_var["key_name"] in unnacepted_attributes:
       ~~~~~~~^^^^^^^^^^^^

It is very clear that Ollama provider doesn't have a key called key_name . I wonder how it worked for you. Can you elaborate?

[kspviswa]

Ok I inspected the code and found that I had to upgrade crewai to latest version. Now it is working.