feat(run): Propagates `model_name` from the gateway request through the runtime and persistence stack to the SQLite database. by yitang · Pull Request #2775 · bytedance/deer-flow

yitang · 2026-05-07T18:03:54Z

Changes:

services.py: Extract model_name from RunCreateRequest.context
manager.py: RunRecord dataclass and create_or_reject() accept model_name
base.py: RunStore.put() interface includes model_name parameter
memory.py: MemoryRunStore implementation updated
sql.py: RunRepository.put() writes model_name to database

Verified: model_name correctly persisted in deerflow.db (tested manually with gpt-4o)

Enables users to track which model was used for each run by persisting model_name to the database. This supports auditability, cost tracking, and debugging by making model usage visible in stored run records.

…stence layer Pass model_name through the full run creation pipeline — from RunCreateRequest.context in the gateway, through RunManager, to the RunStore interface and SQL persistence. This enables client-specified model selection to be recorded per-run in the database.

CLAassistant · 2026-05-07T18:04:01Z

All committers have signed the CLA.

ggnnggez · 2026-05-07T20:36:12Z

@yitang
Thanks for the contribution. I think we should avoid persisting context.model_name directly from the request body, even it already process by FastAPI. At least we should do:

DB compatibility check (normalize and enforce the column length limit)
model allowlist check

And there is also an edge case around effective model resolution. This PR currently records the requested model name, not the model that was actually used. If the requested model is invalid and fallback to the default model, the persisted model_name would be inaccurate.

yitang · 2026-05-07T21:15:36Z

@ggnnggez, thanks for looking at it and providing feedback. I think I understand the 1st and 3rd point, I can try to incorporate them into the next PR.

Would you be able to elaborate a bit more on the 2nd point? I can see allowlist about skills, not about models.

Thanks

Copilot

Pull request overview

This PR threads a model_name value from the gateway run request (body.context) through the run lifecycle so it can be persisted on run records (in-memory store and SQL-backed repository), enabling auditing/cost/debug workflows based on the model used.

Changes:

Extract model_name from the gateway request context and pass it into RunManager.create_or_reject().
Extend the RunRecord / RunStore.put() API surface to carry model_name.
Persist model_name in both the in-memory RunStore and the SQL-backed runs table via RunRepository.put().

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
backend/app/gateway/services.py	Extracts `model_name` from request context and forwards it into run creation.
backend/packages/harness/deerflow/runtime/runs/manager.py	Adds `model_name` to `RunRecord` and persists it via the RunStore.
backend/packages/harness/deerflow/runtime/runs/store/base.py	Extends the RunStore `put()` interface to include `model_name`.
backend/packages/harness/deerflow/runtime/runs/store/memory.py	Stores `model_name` in the in-memory run record dict.
backend/packages/harness/deerflow/persistence/run/sql.py	Writes `model_name` into `RunRow` when persisting runs to SQL.

+    body_context = getattr(body, "context", None) or {}
+    model_name = body_context.get("model_name")
+


        row = RunRow(
            run_id=run_id,
            thread_id=thread_id,
            assistant_id=assistant_id,
            user_id=resolved_user_id,
+            model_name=model_name,
            status=status,


        *,
        on_disconnect: DisconnectMode = DisconnectMode.cancel,
        metadata: dict | None = None,
        kwargs: dict | None = None,
        multitask_strategy: str = "reject",
+        model_name: str | None = None,
    ) -> RunRecord:


…pture - Validate model_name against allowlist in gateway services.py using get_app_config().get_model_config() - Truncate model_name to 128 chars to match DB column constraint - In worker.py, capture effective model name from agent.metadata after agent creation and persist if resolved differently than requested

…ip persistence tests - Add _normalize_model_name() to RunRepository for whitespace stripping and 128-char truncation before DB writes. - Add round-trip unit tests for model_name creation and default None in test_run_manager.py.

…n _normalize_model_name

yitang · 2026-05-08T05:57:25Z

@ggnnggez @WillemJiang updated the PR to address your comments and Copilot's comments. please have a look when you have time.

WillemJiang · 2026-05-09T12:05:55Z

@yitang here are some additional comments

No Alembic migration for the model_name column (High)

The SQLAlchemy model already declares model_name (model.py:23), but there is no migration file — the migrations/versions/ directory only has a .gitkeep. If this column doesn't exist in production databases, the
INSERT will fail. If it does exist because it was added in a prior migration that I'm not seeing, this is fine — but the PR description should clarify the migration status.

Calling private _persist_to_store from worker.py (Medium)

# worker.py:241
record.model_name = effective
await run_manager._persist_to_store(record)

This calls a private method from outside the class, bypassing any coordination logic in RunManager. Consider exposing a public method like update_model_name(run_id, model_name) on RunManager that handles persistence and any necessary locking.

No test for the SQL persistence path (Medium)

Tests only cover the MemoryRunStore path. The RunRepository path has normalization logic (_normalize_model_name) and the actual DB write — this is the critical production path and should have at least a basic integration test.

…ay services Add isinstance check and str() coercion before calling .strip() to prevent AttributeError when non-string types (int, None, etc.) flow through the gateway. Paired with SQL integration test for end-to-end model_name persistence across gateway → langgraph → persistence layer.

…blic update method on RunManager - Drop a1b2c3d4e5f6 migration: model_name already exists in RunRow schema and is auto-created via Base.metadata.create_all() at startup - Add update_model_name() public method to RunManager to replace the private _persist_to_store call in worker.py, preserving internal locking/persistence

yitang · 2026-05-09T13:45:47Z

@WillemJiang thank you for your feedback. Please see the updated PR and below the description. I think it addresses all your comments.

Migration for the model_name column

As far as I can see, the model_name has existed in the codebase and database; it has just never been populated.

I see it in the code, and I see it in my local database. So I think it is not applicable. Please let me know if otherwise.

# backend/packages/harness/deerflow/persistence/engine.py:28-29:
def create_all(engine: Engine) -> None:
    Base.metadata.create_all(engine)
# Base is imported from backend/packages/harness/deerflow/persistence/models/__init__.py:29:
Base = declarative_base()
register_models(Base)  # <-- registers RunRow

calling private function externally

worker.py now calls the new public function update_model_name

additional test on the SQL persistence

The new test uses the production SQL path (RunRepository) , and it covers 4 cases:

Standard string model_name
128-char truncation of long values
Non-string (int) coercion to str
None preservation

WillemJiang requested a review from Copilot May 8, 2026 01:11

Copilot started reviewing on behalf of WillemJiang May 8, 2026 01:12 View session

Copilot AI reviewed May 8, 2026

View reviewed changes

WillemJiang added question Further information is requested reviewing The PR is in reviewing status labels May 8, 2026

yitang added 3 commits May 8, 2026 06:29

fix(run): coerce non-string model_name values before strip/truncate i…

8b05232

…n _normalize_model_name

WillemJiang changed the title ~~Propagates model_name from the gateway request through the runtime and persistence stack to the SQLite database.~~ feat(run): Propagates model_name from the gateway request through the runtime and persistence stack to the SQLite database. May 8, 2026

yitang added 2 commits May 9, 2026 13:58

yitang force-pushed the feat/model-name-propagation branch from f825635 to 4d786eb Compare May 9, 2026 13:21

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(run): Propagates `model_name` from the gateway request through the runtime and persistence stack to the SQLite database.#2775

feat(run): Propagates `model_name` from the gateway request through the runtime and persistence stack to the SQLite database.#2775
yitang wants to merge 6 commits intobytedance:mainfrom
yitang:feat/model-name-propagation

yitang commented May 7, 2026

Uh oh!

CLAassistant commented May 7, 2026 •

edited

Loading

Uh oh!

ggnnggez commented May 7, 2026

Uh oh!

yitang commented May 7, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

yitang commented May 8, 2026 •

edited

Loading

Uh oh!

WillemJiang commented May 9, 2026

Uh oh!

yitang commented May 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

		body_context = getattr(body, "context", None) or {}
		model_name = body_context.get("model_name")

Conversation

yitang commented May 7, 2026

Uh oh!

CLAassistant commented May 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ggnnggez commented May 7, 2026

Uh oh!

yitang commented May 7, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

yitang commented May 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

WillemJiang commented May 9, 2026

Uh oh!

yitang commented May 9, 2026

Migration for the model_name column

calling private function externally

additional test on the SQL persistence

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

CLAassistant commented May 7, 2026 •

edited

Loading

yitang commented May 8, 2026 •

edited

Loading