To register each no-code agent as a separate Phoenix project, you can use the OpenInferenceTracer and LangChainInstrumentor to dynamically set the project name for each agent. You don't need to extend the LangChain instrumentor; instead, you can configure the tracer provider with different resource attributes for each agent. This can be done by setting the project name in the resource attributes of the tracer provider dynamically as each agent is instantiated.
Here's a general approach:
1. Set Resource Attributes: Define the project name in the resource attributes for each agent dynamically.
2. Configure Tracer Provider: Use the TracerProvider to set up the trace provider with the specified resource attributes for each agent.
3. Instrument Each Agent: Use the LangChainInstrumentor to instrument each agent with its own tracer provider.
Here's a code snippet to illustrate this:
```
from phoenix.trace.langchain import OpenInferenceTracer, LangChainInstrumentor
from opentelemetry.sdk.resources import Resource
from opentelemetry.sdk import trace as trace_sdk
def setup_agent_tracer(agent_name):
resource = Resource(attributes={"openinference.project.name": agent_name})
tracer_provider = trace_sdk.TracerProvider(resource=resource)
tracer = OpenInferenceTracer(tracer_provider=tracer_provider)
LangChainInstrumentor(tracer=tracer).instrument()
agent_names = ["agent1", "agent2", "agent3"] # Dynamically determined
for agent_name in agent_names:
setup_agent_tracer(agent_name)
LBKSThis setup ensures that each agent is registered under a separate project in Phoenix. Would you like more details on any specific part of this setup?
Sources:
- [GitHub Issue #3375](https://github.com/Arize-ai/phoenix/issues/3375)
