AgentClient

The client for Tecton agent service. The client should always be created using the static methods from_remote or from_local.

Attributes

Name	Data Type	Description
logger	Logger	Get the current logger of the client. The logger can be controlled using the context manager set_logger.
metastore	Dict[str,Any]	Get the metastore of the client. The metastore contains the metadata of the tools, prompts, knowledge and other resources. This function should not be used directly.

Methods

Name	Description
__init__()	Initialize AgentClient
from_local(...)	Create a client for a local agent service. This is for testing
from_remote(...)	Create a client connecting to a deployed agent service.
invoke_agent(...)	Invoke an agent for a specific LLM framework. Compared to make_agent, this
invoke_feature_view(...)	Invoke a feature view in the service.
invoke_prompt(...)	Invoke a prompt in the service.
invoke_tool(...)	Invoke a tool in the service.
make_agent(...)	Make an agent for a specific LLM framework.
search(...)	Search a tool in the service.
set_context(...)	Set the context for the client. This is a context manager. The context
set_logger(...)	Set the logger for the client. This is a context manager.

init(...)

from_local(...)

Create a client for a local agent service. This is for testing and development purposes.

Parameters

• service (AgentService) - The agent service

Returns

AgentClient: The agent client

from_remote(...)

Create a client connecting to a deployed agent service.

Parameters

• url (str) - The Tecton URL


• workspace (str) - The workspace name


• service (str) - The service name


• api_key (str) - The API key

Returns

AgentClient: The agent client

invoke_agent(...)

Invoke an agent for a specific LLM framework. Compared to make_agent, this function is simpler with unified input (str) and output (str), but it is less flexible than getting the agent object of the specific framework and invoking.

Parameters

• llm (Any) - The language model object in a specific framework (e.g. LangChain or LLamaIndex)


• message (str) - The message (question)


• system_prompt (Optional[str]) - The name of the system prompt in the service Default: None


• chat_history (Any) - The chat history Default: None


• context (Optional[Dict[str,Any]]) - The context to run the agent, this will override the context set by set_context Default: None


• kwargs (Any) -

Returns

str: The response

Example

from langchain_openai import ChatOpenAI

openai = ChatOpenAI(model = "gpt-4o", temperature=0)

from tecton_gen_ai.fco import prompt, AgentService, AgentClient
from tecton_gen_ai.utils.tecton_utils import make_request_source

req = make_request_source(age=int)

@prompt(sources=[req])
def sys_prompt(req):
    return "You are talking to a "+str(req["age"])+" years old person."

service = AgentService(
    "app",
    prompts=[sys_prompt],
)

client = AgentClient.from_local(service)
with client.set_context({"age": 3}):
    print(client.invoke_agent(llm, "why sky is blue", "sys_prompt"))
with client.set_context({"age": 30}):
    print(client.invoke_agent(llm, "why sky is blue", "sys_prompt"))

invoke_feature_view(...)

Invoke a feature view in the service.

Parameters

• name (str) - The name of the feature view


• kwargs (Optional[Dict[str,Any]]) - The arguments for the feature view, the keys should match the entity schema of the feature view. Default: None

Returns

Any: The result of the feature view

Example

from tecton_gen_ai.testing import make_local_feature_view, set_dev_mode

set_dev_mode()

bfv = make_local_feature_view(
    "user_age",
    {"user_id": 1, "age": 30},
    ["user_id"],
    description="The age of a user",
)

from tecton_gen_ai.fco import AgentService, AgentClient

service = AgentService(
    "app",
    tools=[bfv],
)

client = AgentClient.from_local(service)
print(client.invoke_feature_view("user_age", {"user_id":1}))

invoke_prompt(...)

Invoke a prompt in the service.

Parameters

• name (str) - The name of the prompt


• kwargs (Optional[Dict[str,Any]]) - The arguments for the prompt, it overrides the context set by set_context Default: None

Returns

str: The result of the prompt

Example

from tecton_gen_ai.fco import AgentService, AgentClient, prompt
from tecton_gen_ai.utils.tecton_utils import make_request_source

req = make_request_source(age=int)

@prompt(sources=[req])
def sys_prompt(req):
    return "You are talking to a "+str(req["age"])+" years old person."

service = AgentService(
    "app",
    prompts=[sys_prompt],
)

client = AgentClient.from_local(service)
print(client.invoke_prompt("sys_prompt", {"age": 3}))

invoke_tool(...)

Invoke a tool in the service.

Parameters

• name (str) - The name of the tool


• kwargs (Optional[Dict[str,Any]]) - The arguments for the tool Default: None

Returns

Any: The result of the tool

Example

from tecton_gen_ai.fco import AgentService, AgentClient, tool

@tool
def get_price_of_fruit(fruit:str) -> int:
    '''
    Get the price of a fruit

    Args:

        fruit: The name of the fruit

    Returns:

        int: The price of the fruit
    '''
    return 10 if fruit == "apple" else 5

service = AgentService(
    "app",
    tools=[get_price_of_fruit],
)

client = AgentClient.from_local(service)
print(client.invoke_tool("get_price_of_fruit", {"fruit":"apple"}))

make_agent(...)

Make an agent for a specific LLM framework.

Parameters

• llm (Any) - The language model object in a specific framework (e.g. LangChain or LLamaIndex)


• system_prompt (Optional[str]) - The name of the system prompt in the service Default: None


• kwargs (Any) -

Returns

Any: The agent object

Example

from langchain_openai import ChatOpenAI

openai = ChatOpenAI(model = "gpt-4o", temperature=0)

from tecton_gen_ai.fco import prompt, AgentService, AgentClient
from tecton_gen_ai.utils.tecton_utils import make_request_source

req = make_request_source(age=int)

@prompt(sources=[req])
def sys_prompt(req):
    return "You are talking to a "+str(req["age"])+" years old person."

service = AgentService(
    "app",
    prompts=[sys_prompt],
)

client = AgentClient.from_local(service)
agent = client.make_agent(openai, system_prompt="sys_prompt")
with client.set_context({"age": 3}):
    print(agent.invoke({"input":"why sky is blue"}))
with client.set_context({"age": 30}):
    print(agent.invoke({"input":"why sky is blue"}))

search(...)

Search a tool in the service.

Parameters

• name (str) - The name of the search tool


• query (str) - The query string


• top_k (int) - The number of results to return, defaults to 5 Default: 5


• filter (Optional[Dict[str,Any]]) - The filter for the search, default to None (no filter) Default: None

Returns

List[Dict[str,Any]]: The search results

Example

from tecton_gen_ai.testing import make_local_source
from tecton_gen_ai.testing.utils import make_local_vector_db_config

df = [
    {"zip":"98005", "item_id":1, "description":"pencil"},
    {"zip":"98005", "item_id":2, "description":"car"},
    {"zip":"98005", "item_id":3, "description":"paper"},
    {"zip":"10065", "item_id":4, "description":"boat"},
    {"zip":"10065", "item_id":5, "description":"cheese"},
    {"zip":"10065", "item_id":6, "description":"apple"},
]

src = make_local_source(
    "for_sale",
    df,
    description="Items information",  # required for source_as_knowledge
)
vdb_conf = make_local_vector_db_config()

# Create a knowledge base from the source
from tecton_gen_ai.fco import source_as_knowledge

knowledge = source_as_knowledge(
    src,
    vector_db_config=vdb_conf,
    vectorize_column="description",
    filter = [("zip", str, "the zip code of the item for sale")]
)

# Serve the knowledge base
from tecton_gen_ai.fco import AgentService

service = AgentService(
    "app",
    knowledge=[knowledge],
)

# Test locally
from tecton_gen_ai.fco import AgentClient

client = AgentClient.from_local(service)
# search without filter
print(client.search("for_sale", query="fruit"))
# search with filter
print(client.search("for_sale", query="fruit", top_k=3, filter={"zip": "27001"}))
print(client.search("for_sale", query="fruit", top_k=3, filter={"zip": "10065"}))

set_context(...)

Set the context for the client. This is a context manager. The context will be used as the arguments for the prompts, tools and knowledge.

Parameters

• context (Optional[Dict[str,Any]]) - The new context, or None to clear the context

Example

conext = {"a":1, "b":2}
new_args = {"b":3, "c":4}
with client.set_context(context):
    # the context will be used as the arguments of my_tool
    # new_args will override the context
    # the final arguments for my_tool will be {"a":1, "b":3, "c":4}
    client.invoke_tool("my_tool", new_args)

set_logger(...)

Set the logger for the client. This is a context manager.

Parameters

• logger (Optional[logging.Logger]) - The new logger, or None to use the no-op logger

Example

with client.set_logger(logger):
    # do something