Are cloud review automations working today? cc <@U...
# proj-automation
q
Are cloud review automations working today? cc @chilly-france-44439
c
yes it should be, as long as you request changes from all-hands-bot or tag
review-this
?
if not, could you send me the pr link so i can debug
but the current version is hacked using a special sdk main script
q
c
checing that now
πŸ™ 1
i shis related to any existig incident?
q
dont think so
c
πŸ˜…
πŸ₯² 1
aha 1
q
is the model being reset to minimax on every release?
c
i actually don't know, where do we implement this model name override? cc @modern-fireman-99755 @limited-student-10747 not sure if yall have any context here?
l
I’m not sure where that override happens actually
c
OK i've set the model name back, it seems to working again πŸ˜…
@nutritious-balloon-65098 can you help me figure out in OpenHands/deploy and OpenHands/OpenHands repo and help me understand where does this minimax-m2.7 override happens
n
I'm on it! Xingyao Wang can track my progress here.
Summary This was a pure investigation/research task β€” no code changes were made. Question answered: Where does the minimax-m2.7 override happen? It lives entirely in
OpenHands/OpenHands
(not in
OpenHands/deploy
) and flows through two files: 1.
enterprise/server/constants.py
β€” The root cause β€’ A versioned map
PERSONAL_WORKSPACE_VERSION_TO_MODEL
bumps the default model on each release. The current max version (6) maps to
'minimax-m2.7'
. β€’
ORG_SETTINGS_VERSION
and
PERSONAL_WORKSPACE_VERSION
are both set to
max()
of that map, currently
6
. β€’
get_default_litellm_model()
reads from this map (unless the
LITELLM_DEFAULT_MODEL
env var is set, which is the escape hatch the team used to "set it back"). 2.
openhands/app_server/utils/llm.py
β€” A second hardcoded reference β€’
DEFAULT_OPENHANDS_MODEL = 'openhands/minimax-m2.7'
is hardcoded and returned as the
default_model
in the
/api/options/models
API response. This requires a code change to update (the env var does not override this one). The auto-upgrade mechanism (why it affects existing users on every release): β€’
enterprise/storage/org_app_settings_store.py
β†’
_validate_org_version()
runs on every org fetch. If the org's stored
org_version
is less than the current
ORG_SETTINGS_VERSION
, it `deep_merge`s the new default model into the org's
agent_settings
β€” silently resetting users to minimax on their next login after a deployment.
OpenHands/deploy
has no minimax-specific configuration β€” it's all in the application code.
c
@nutritious-balloon-65098 who implemented this?
n
I'm on it! Xingyao Wang can continue tracking my progress here.
Summary This was a pure investigation β€” no code changes were made. Here's what was found: Question 1: Where does the minimax-m2.7 override happen? Two places in
OpenHands/OpenHands
: 1.
enterprise/server/constants.py
— the version→model map: ◦
PERSONAL_WORKSPACE_VERSION_TO_MODEL[6] = 'minimax-m2.7'
(current version = 6) β—¦
get_default_litellm_model()
reads this to return
litellm_proxy/minimax-m2.7
β—¦
_validate_org_version()
in
org_app_settings_store.py
uses this to auto-upgrade every existing user's model on next login whenever their stored
org_version
is behind the current one β€” which is why it feels like a reset on every release 1.
openhands/app_server/utils/llm.py
β€” hardcoded
DEFAULT_OPENHANDS_MODEL = 'openhands/minimax-m2.7'
, returned as the frontend's pre-selected default by
/api/options/models
The
OpenHands/deploy
repo has no minimax-specific overrides
β€” it's entirely in the OpenHands codebase. The escape hatch is the
LITELLM_DEFAULT_MODEL
env var, which bypasses the version map. Question 2: Who implemented it? Vasco Schiavo (@VascoSch92), who authored and merged PR #14508 on May 21, 2026, described as a co-promotion with MiniMax for a model profiles launch.
c
It looks like this overrides probably only happen onces πŸ‘€