feat: billing and credit management v2 (#429)

* feat: use mps generated correlation ID

* chore: update pipecat submodule

* feat: add credit purchase URL

* feat: carve out billing page and show credit ledger

* feat: deprecate dograh based quota tracking

* fix: remove cost calculation from dograh codebase

* fix: create mps account on migrate to v2

* chore: update pipecat

api/services/pipecat/run_pipeline.py

@@ -162,15 +162,13 @@ async def run_pipeline_telephony(
         workflow_id: Workflow being executed.
         workflow_run_id: Workflow run row.
         user_id: Owner of the workflow.
-        call_id: Provider call identifier (stored in cost_info for billing).
+        call_id: Provider call identifier.
         transport_kwargs: Provider-specific kwargs forwarded to the transport
             factory (e.g. stream_sid + call_sid for Twilio).
     """
     logger.debug(f"Running {provider_name} pipeline for workflow_run {workflow_run_id}")
     set_current_run_id(workflow_run_id)
 
-    await db_client.update_workflow_run(workflow_run_id, cost_info={"call_id": call_id})
-
     workflow = await db_client.get_workflow(workflow_id, user_id)
     if workflow:
         set_current_org_id(workflow.organization_id)
@@ -340,7 +338,7 @@ async def _run_pipeline(
     if workflow_run.is_completed:
         raise HTTPException(status_code=400, detail="Workflow run already completed")
 
-    merged_call_context_vars = workflow_run.initial_context
+    merged_call_context_vars = dict(workflow_run.initial_context or {})
     # If there is some extra call_context_vars, fold them in. Persistence
     # happens once below, after runtime_configuration is also resolved.
     if call_context_vars:
@@ -398,6 +396,19 @@ async def _run_pipeline(
     else:
         user_config = resolved_user_config
 
+    from api.services.managed_model_services import (
+        MPS_CORRELATION_ID_CONTEXT_KEY,
+        ensure_mps_correlation_id,
+    )
+
+    mps_correlation_id = await ensure_mps_correlation_id(
+        ai_model_config=user_config,
+        workflow_run_id=workflow_run_id,
+        initial_context=merged_call_context_vars,
+    )
+    if mps_correlation_id:
+        merged_call_context_vars[MPS_CORRELATION_ID_CONTEXT_KEY] = mps_correlation_id
+
     # Detect realtime mode (speech-to-speech services like OpenAI Realtime, Gemini Live)
     is_realtime = user_config.is_realtime and user_config.realtime is not None
 
@@ -409,11 +420,23 @@ async def _run_pipeline(
         # Realtime services don't implement run_inference, so create a
         # separate text LLM for variable extraction and other out-of-band
         # inference calls.
-        inference_llm = create_llm_service(user_config)
+        inference_llm = create_llm_service(
+            user_config,
+            correlation_id=mps_correlation_id,
+        )
     else:
-        stt = create_stt_service(user_config, audio_config, keyterms=keyterms)
-        tts = create_tts_service(user_config, audio_config)
-        llm = create_llm_service(user_config)
+        stt = create_stt_service(
+            user_config,
+            audio_config,
+            keyterms=keyterms,
+            correlation_id=mps_correlation_id,
+        )
+        tts = create_tts_service(
+            user_config,
+            audio_config,
+            correlation_id=mps_correlation_id,
+        )
+        llm = create_llm_service(user_config, correlation_id=mps_correlation_id)
         inference_llm = None
 
     # Stamp the providers/models actually resolved for this run onto
@@ -695,7 +718,10 @@ async def _run_pipeline(
         # Create a separate LLM instance for the voicemail sub-pipeline
         # (can't share with main pipeline as it would mess up frame linking)
         if voicemail_config.get("use_workflow_llm", True):
-            voicemail_llm = create_llm_service(user_config)
+            voicemail_llm = create_llm_service(
+                user_config,
+                correlation_id=mps_correlation_id,
+            )
         else:
             voicemail_llm = create_llm_service_from_provider(
                 provider=voicemail_config.get("provider", "openai"),

api/services/pipecat/service_factory.py

@@ -78,7 +78,10 @@ def _validate_runtime_service_url(url: str, field_name: str) -> None:
 
 
 def create_stt_service(
-    user_config, audio_config: "AudioConfig", keyterms: list[str] | None = None
+    user_config,
+    audio_config: "AudioConfig",
+    keyterms: list[str] | None = None,
+    correlation_id: str | None = None,
 ):
     """Create and return appropriate STT service based on user configuration
 
@@ -160,6 +163,7 @@ def create_stt_service(
         return DograhSTTService(
             base_url=base_url,
             api_key=user_config.stt.api_key,
+            correlation_id=correlation_id,
             settings=DograhSTTSettings(
                 model=user_config.stt.model,
                 language=language,
@@ -286,7 +290,9 @@ def create_stt_service(
         )
 
 
-def create_tts_service(user_config, audio_config: "AudioConfig"):
+def create_tts_service(
+    user_config, audio_config: "AudioConfig", correlation_id: str | None = None
+):
     """Create and return appropriate TTS service based on user configuration
 
     Args:
@@ -404,6 +410,7 @@ def create_tts_service(user_config, audio_config: "AudioConfig"):
         return DograhTTSService(
             base_url=base_url,
             api_key=user_config.tts.api_key,
+            correlation_id=correlation_id,
             settings=DograhTTSSettings(
                 model=user_config.tts.model,
                 voice=user_config.tts.voice,
@@ -564,6 +571,7 @@ def create_llm_service_from_provider(
     model: str,
     api_key: str | None,
     *,
+    correlation_id: str | None = None,
     base_url: str | None = None,
     endpoint: str | None = None,
     aws_access_key: str | None = None,
@@ -637,6 +645,7 @@ def create_llm_service_from_provider(
         return DograhLLMService(
             base_url=f"{MPS_API_URL}/api/v1/llm",
             api_key=api_key,
+            correlation_id=correlation_id,
             settings=OpenAILLMSettings(model=model),
         )
     elif provider == ServiceProviders.AWS_BEDROCK.value:
@@ -851,7 +860,7 @@ def create_realtime_llm_service(user_config, audio_config: "AudioConfig"):
         )
 
 
-def create_llm_service(user_config):
+def create_llm_service(user_config, correlation_id: str | None = None):
     """Create and return appropriate LLM service based on user configuration."""
     provider = user_config.llm.provider
     model = user_config.llm.model
@@ -880,4 +889,10 @@ def create_llm_service(user_config):
     elif provider == ServiceProviders.SARVAM.value:
         kwargs["temperature"] = user_config.llm.temperature
 
-    return create_llm_service_from_provider(provider, model, api_key, **kwargs)
+    return create_llm_service_from_provider(
+        provider,
+        model,
+        api_key,
+        correlation_id=correlation_id,
+        **kwargs,
+    )