TTS/STT: speech-to-text using gemini in unified API

backend/app/alembic/versions/041_add_llm_call_table.py

@@ -0,0 +1,185 @@
+"""add llm call table
+
+Revision ID: 9ca4b67c5226
+Revises: 040
+Create Date: 2026-01-23 13:36:23.023444
+
+"""
+from alembic import op
+import sqlalchemy as sa
+import sqlmodel.sql.sqltypes
+from sqlalchemy.dialects import postgresql
+
+# revision identifiers, used by Alembic.
+revision = "041"
+down_revision = "040"
+branch_labels = None
+depends_on = None
+
+
+def upgrade():
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.create_table(
+        "llm_call",
+        sa.Column(
+            "id",
+            sa.Uuid(),
+            nullable=False,
+            comment="Unique identifier for the LLM call record",
+        ),
+        sa.Column(
+            "job_id",
+            sa.Uuid(),
+            nullable=False,
+            comment="Reference to the parent job (status tracked in job table)",
+        ),
+        sa.Column(
+            "project_id",
+            sa.Integer(),
+            nullable=False,
+            comment="Reference to the project this LLM call belongs to",
+        ),
+        sa.Column(
+            "organization_id",
+            sa.Integer(),
+            nullable=False,
+            comment="Reference to the organization this LLM call belongs to",
+        ),
+        sa.Column(
+            "input",
+            sqlmodel.sql.sqltypes.AutoString(),
+            nullable=False,
+            comment="User input - text string, binary data, or file path for multimodal",
+        ),
+        sa.Column(
+            "input_type",
+            sa.String(),
+            nullable=False,
+            comment="Input type: text, audio, image",
+        ),
+        sa.Column(
+            "output_type",
+            sa.String(),
+            nullable=True,
+            comment="Expected output type: text, audio, image",
+        ),
+        sa.Column(
+            "provider",
+            sa.String(),
+            nullable=False,
+            comment="AI provider: openai, google, anthropic",
+        ),
+        sa.Column(
+            "model",
+            sqlmodel.sql.sqltypes.AutoString(),
+            nullable=False,
+            comment="Specific model used e.g. 'gpt-4o', 'gemini-2.5-pro'",
+        ),
+        sa.Column(
+            "provider_response_id",
+            sqlmodel.sql.sqltypes.AutoString(),
+            nullable=True,
+            comment="Original response ID from the provider (e.g., OpenAI's response ID)",
+        ),
+        sa.Column(
+            "content",
+            postgresql.JSONB(astext_type=sa.Text()),
+            nullable=True,
+            comment="Response content: {text: '...'}, {audio_bytes: '...'}, or {image: '...'}",
+        ),
+        sa.Column(
+            "usage",
+            postgresql.JSONB(astext_type=sa.Text()),
+            nullable=True,
+            comment="Token usage: {input_tokens, output_tokens, reasoning_tokens}",
+        ),
+        sa.Column(
+            "conversation_id",
+            sqlmodel.sql.sqltypes.AutoString(),
+            nullable=True,
+            comment="Identifier linking this response to its conversation thread",
+        ),
+        sa.Column(
+            "auto_create",
+            sa.Boolean(),
+            nullable=True,
+            comment="Whether to auto-create conversation if conversation_id doesn't exist (OpenAI specific)",
+        ),
+        sa.Column(
+            "config",
+            postgresql.JSONB(astext_type=sa.Text()),
+            nullable=True,
+            comment="Configuration: {config_id, config_version} for stored config OR {config_blob} for ad-hoc config",
+        ),
+        sa.Column(
+            "created_at",
+            sa.DateTime(),
+            nullable=False,
+            comment="Timestamp when the LLM call was created",
+        ),
+        sa.Column(
+            "updated_at",
+            sa.DateTime(),
+            nullable=False,
+            comment="Timestamp when the LLM call was last updated",
+        ),
+        sa.Column(
+            "deleted_at",
+            sa.DateTime(),
+            nullable=True,
+            comment="Timestamp when the record was soft-deleted",
+        ),
+        sa.ForeignKeyConstraint(["job_id"], ["job.id"], ondelete="CASCADE"),
+        sa.ForeignKeyConstraint(
+            ["organization_id"], ["organization.id"], ondelete="CASCADE"
+        ),
+        sa.ForeignKeyConstraint(["project_id"], ["project.id"], ondelete="CASCADE"),
+        sa.PrimaryKeyConstraint("id"),
+    )
+    op.create_index(
+        "idx_llm_call_conversation_id",
+        "llm_call",
+        ["conversation_id"],
+        unique=False,
+        postgresql_where=sa.text("conversation_id IS NOT NULL AND deleted_at IS NULL"),
+    )
+    op.create_index(
+        "idx_llm_call_job_id",
+        "llm_call",
+        ["job_id"],
+        unique=False,
+        postgresql_where=sa.text("deleted_at IS NULL"),
+    )
+    op.alter_column(
+        "collection",
+        "llm_service_name",
+        existing_type=sa.VARCHAR(),
+        comment="Name of the LLM service",
+        existing_comment="Name of the LLM service provider",
+        existing_nullable=False,
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade():
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.alter_column(
+        "collection",
+        "llm_service_name",
+        existing_type=sa.VARCHAR(),
+        comment="Name of the LLM service provider",
+        existing_comment="Name of the LLM service",
+        existing_nullable=False,
+    )
+    op.drop_index(
+        "idx_llm_call_job_id",
+        table_name="llm_call",
+        postgresql_where=sa.text("deleted_at IS NULL"),
+    )
+    op.drop_index(
+        "idx_llm_call_conversation_id",
+        table_name="llm_call",
+        postgresql_where=sa.text("conversation_id IS NOT NULL AND deleted_at IS NULL"),
+    )
+    op.drop_table("llm_call")
+    # ### end Alembic commands ###

backend/app/celery/beat.py

@@ -1,6 +1,7 @@
 """
 Celery beat scheduler for cron jobs.
 """
+
 import logging
 from celery import Celery
 from app.celery.celery_app import celery_app

backend/app/celery/utils.py

@@ -2,6 +2,7 @@
 Utility functions for easy Celery integration across the application.
 Business logic modules can use these functions without knowing Celery internals.
 """
+
 import logging
 from typing import Any, Dict, Optional
 from celery.result import AsyncResult

backend/app/celery/worker.py

@@ -1,6 +1,7 @@
 """
 Celery worker management script.
 """
+
 import logging
 import multiprocessing
 from celery.bin import worker

backend/app/cli/bench/commands.py

@@ -210,7 +210,7 @@ def send_benchmark_request(
         )
     else:
         typer.echo(response.text)
-        typer.echo(f"[{i+1}/{total}] FAILED - Status: {response.status_code}")
+        typer.echo(f"[{i + 1}/{total}] FAILED - Status: {response.status_code}")
         raise Exception(f"Request failed with status code {response.status_code}")
 
 

backend/app/core/providers.py

@@ -12,6 +12,7 @@ class Provider(str, Enum):
     OPENAI = "openai"
     AWS = "aws"
     LANGFUSE = "langfuse"
+    GOOGLE = "google"
 
 
 @dataclass