What session id should I use in FastAPI?

Any stable identifier for the conversation — reuse your existing FastAPI session or thread id so turns are grouped correctly.

Is passing the prompt required for FastAPI?

It is optional but strongly recommended: the prompt is what makes the injected ad contextually relevant to the current turn.

How-to guide

How to Pass Session Context to Ads in FastAPI

This guide shows how to pass session context to ads in a FastAPI app, the goal being to supply session id, prompt, and metadata so injected ads are relevant. It builds on the Monetzly gRPC service (no Python SDK exists for this stack), so the approach is specific to how FastAPI produces and streams responses.

Overview

Ad relevance in a FastAPI app comes from context. When you inject, you pass a session id, the live prompt, and optional metadata — that is what lets Monetzly match a recommendation to what the user is actually asking about in that turn.

The session id groups turns from the same conversation; the prompt drives per-turn relevance; metadata carries anything extra you want to target on. All are supplied per injection call.

FastAPI apps are usually built for LLM API backends, SSE/streaming chat endpoints, and agent microservices, so pass session context to ads typically comes up while a user is mid-conversation, the moment where monetization has to be additive rather than disruptive.

How this works on FastAPI

There is no official Monetzly Python SDK — the package is Node-only. The supported path for a Python stack is the gRPC service the package ships as tps_alter.proto: generate Python stubs and open a ProcessStream bidi call (StartRequest, then a TokenRequest per LLM token, then StopRequest), reading back the ad-injected TokenResponses. Alternatively, run a tiny Node sidecar that uses the SDK and stream tokens to it. Either way, confirm the auth handshake for non-JS clients before publishing.

Since FastAPI is a Python stack with no official SDK, this task runs through the shipped tps_alter.proto gRPC service (or a Node sidecar); the auth handshake for a direct client is still TODO_VERIFY.

Steps

Choose a stable session id for the conversation (reuse your FastAPI session or thread id).
Pass the current user prompt on each injection call.
Add optional metadata (topic, user type) to sharpen targeting.

Integration snippet

TODO_VERIFY: FastAPI uses the gRPC proto path, confirm the auth handshake with Monetzly before relying on this in production.

# 1) Generate stubs from the proto shipped in the SDK package:
#    python -m grpc_tools.protoc -I. --python_out=. --grpc_python_out=. tps_alter.proto
import grpc
import tps_alter_pb2 as pb
import tps_alter_pb2_grpc as rpc

async def inject(llm_tokens, prompt, session_id, api_key):
    # TODO_VERIFY: how is the API key passed for a direct gRPC client?
    # (metadata key name / channel credentials). Confirm with Monetzly.
    creds = grpc.ssl_channel_credentials()
    async with grpc.aio.secure_channel("your-server.com:443", creds) as ch:
        stub = rpc.TPSAlterServiceStub(ch)

        async def requests():
            yield pb.StreamRequest(start=pb.StartRequest(
                prompt=prompt, session_id=session_id, metadata={"api_key": api_key}))
            async for tok in llm_tokens:
                yield pb.StreamRequest(token=pb.TokenRequest(token=tok))
            yield pb.StreamRequest(stop=pb.StopRequest(session_id=session_id))

        async for resp in stub.ProcessStream(requests()):
            if resp.HasField("token"):
                yield resp.token.token  # ad-injected token

# In a FastAPI route, feed your LLM's token generator into inject(...) and
# return the results with StreamingResponse.