Job Management: Preserve llm_call details

backend/app/alembic/versions/058_make_llm_call_fields_nullable.py

@@ -0,0 +1,28 @@
+"""make llm_call input_type, provider, model nullable
+
+Revision ID: 058
+Revises: 057
+Create Date: 2026-05-11 00:00:00.000000
+
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "058"
+down_revision = "057"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.alter_column("llm_call", "input_type", nullable=True)
+    op.alter_column("llm_call", "provider", nullable=True)
+    op.alter_column("llm_call", "model", nullable=True)
+
+
+def downgrade() -> None:
+    op.alter_column("llm_call", "model", nullable=False)
+    op.alter_column("llm_call", "provider", nullable=False)
+    op.alter_column("llm_call", "input_type", nullable=False)

backend/app/crud/llm.py

@@ -19,7 +19,7 @@
 logger = logging.getLogger(__name__)
 
 
-def serialize_input(query_input: QueryInput | str) -> str:
+def serialize_input(query_input: QueryInput | str | list) -> str:
     """Serialize query input for database storage.
 
     For text: stores the actual content value
@@ -279,3 +279,100 @@ def get_llm_calls_by_job_id(
     )
 
     return list(session.exec(statement).all())
+
+
+def get_llm_call_by_job_id(session: Session, job_id: UUID) -> LlmCall | None:
+    """Return the single active LlmCall for a standalone job (no chain_id)."""
+    statement = (
+        select(LlmCall)
+        .where(
+            LlmCall.job_id == job_id,
+            LlmCall.chain_id.is_(None),
+            LlmCall.deleted_at.is_(None),
+        )
+        .order_by(LlmCall.created_at.desc())
+    )
+    return session.exec(statement).first()
+
+
+def create_llm_call_pending(
+    session: Session,
+    *,
+    job_id: UUID,
+    project_id: int,
+    organization_id: int,
+    request: LLMCallRequest,
+    chain_id: UUID | None = None,
+) -> LlmCall:
+    """Create a minimal LlmCall row at job-creation time.
+
+    Only fields available before config resolution are populated.
+    input_type, output_type, provider, model, content, usage stay NULL
+    and are filled in by the Celery task via update_llm_call_resolved_fields().
+    """
+    config_dict: dict[str, Any] | None = None
+    if request.config.is_stored_config:
+        config_dict = {
+            "config_id": str(request.config.id),
+            "config_version": request.config.version,
+        }
+
+    conversation_id = None
+    auto_create = None
+    if request.query.conversation:
+        conversation_id = request.query.conversation.id
+        auto_create = request.query.conversation.auto_create
+
+    db_llm_call = LlmCall(
+        job_id=job_id,
+        project_id=project_id,
+        organization_id=organization_id,
+        chain_id=chain_id,
+        input=serialize_input(request.query.input),
+        conversation_id=conversation_id,
+        auto_create=auto_create,
+        config=config_dict,
+    )
+
+    session.add(db_llm_call)
+    session.commit()
+    session.refresh(db_llm_call)
+
+    logger.info(
+        f"[create_llm_call_pending] Created pending LLM call id={db_llm_call.id}, job_id={job_id}"
+    )
+
+    return db_llm_call
+
+
+def update_llm_call_resolved_fields(
+    session: Session,
+    *,
+    llm_call_id: UUID,
+    input_type: str,
+    output_type: str | None,
+    provider: str | None,
+    model: str | None,
+    config: dict[str, Any],
+) -> LlmCall:
+    """Populate config-resolved fields on a pending LlmCall row."""
+    db_llm_call = session.get(LlmCall, llm_call_id)
+    if not db_llm_call:
+        raise ValueError(f"LLM call not found with id={llm_call_id}")
+
+    db_llm_call.input_type = input_type
+    db_llm_call.output_type = output_type
+    db_llm_call.provider = provider
+    db_llm_call.model = model
+    db_llm_call.config = config
+    db_llm_call.updated_at = now()
+
+    session.add(db_llm_call)
+    session.commit()
+    session.refresh(db_llm_call)
+
+    logger.info(
+        f"[update_llm_call_resolved_fields] Updated resolved fields | llm_call_id={llm_call_id}, provider={provider}, model={model}"
+    )
+
+    return db_llm_call

backend/app/crud/llm_chain.py

@@ -2,7 +2,7 @@
 from typing import Any
 from uuid import UUID
 
-from sqlmodel import Session
+from sqlmodel import Session, select
 
 from app.core.util import now
 from app.models.llm.request import ChainStatus, LlmChain
@@ -144,3 +144,9 @@ def update_llm_chain_block_completed(
         f"llm_call_id={llm_call_id}"
     )
     return db_chain
+
+
+def get_llm_chain_by_job_id(session: Session, job_id: UUID) -> LlmChain | None:
+    """Return the LlmChain record associated with the given job."""
+    statement = select(LlmChain).where(LlmChain.job_id == job_id)
+    return session.exec(statement).first()

backend/app/models/llm/request.py

@@ -537,11 +537,11 @@ class LlmCall(SQLModel, table=True):
     )
 
     # NOTE: image, pdf, multimodal are internal labels stored in the table not user facing.
-    input_type: Literal["text", "audio", "image", "pdf", "multimodal"] = Field(
-        ...,
+    input_type: Literal["text", "audio", "image", "pdf", "multimodal"] | None = Field(
+        default=None,
         sa_column=sa.Column(
             sa.String,
-            nullable=False,
+            nullable=True,
             comment="Input type: text, audio, image, pdf, multimodal",
         ),
     )
@@ -556,20 +556,22 @@ class LlmCall(SQLModel, table=True):
     )
 
     # Provider and model info
-    provider: str = Field(
-        ...,
+    provider: str | None = Field(
+        default=None,
         sa_column=sa.Column(
             sa.String,
-            nullable=False,
+            nullable=True,
             comment="AI provider as sent by user (e.g openai, -native, google)",
         ),
     )
 
-    model: str = Field(
-        ...,
-        sa_column_kwargs={
-            "comment": "Specific model used e.g. 'gpt-4o', 'gemini-2.5-pro'"
-        },
+    model: str | None = Field(
+        default=None,
+        sa_column=sa.Column(
+            sa.String,
+            nullable=True,
+            comment="Specific model used e.g. 'gpt-4o', 'gemini-2.5-pro'",
+        ),
     )
 
     # Response fields