refactor(model-routing): centralize explicit foreground fallback policy (#6020)

* refactor(model-routing): centralize explicit foreground fallback policy

Make foreground fallback an explicit per-user, availability-only policy shared by streaming Chat, non-stream Chat, and Agent runs.

Preserve strict defaults, owner/model and credential boundaries, pinned Agent routes, and truthful per-round provenance/accounting. Carry provider-reported model identifiers through native streaming adapters, non-stream responses, and caches, and keep legacy default_model_fallbacks as tombstoned raw storage that generic settings APIs and agent tools cannot expose or mutate.

* fix(agent-loop): restore rebase-dropped qwen routing, workspace prompt, and temperature clamp

* fix(model-routing): thread selected endpoint identity, fix cost classification and fallback eligibility

* fix(chat): restore stream helpers and harden run stop lifecycle

* fix(model-routing): let numeric provider codes win over symbolic rate-limit statuses

* fix(agent-loop): apply qwen temperature and notes-tool clamps per fallback candidate

* fix(chat): honor queued stop across resend and reload canonical terminal on EOF

* fix(chat): track stop queue and cleanup ownership by per-send generation

* fix(agent-loop): preserve requested temperature for non-qwen fallback candidates

* fix(chat): reserve send ownership before any await and scope stop to the current send

* fix(chat): clear the previous run identity at send reservation

---------

Co-authored-by: RaresKeY <158580472+RaresKeY@users.noreply.github.com>
Co-authored-by: StressTestor <212606152+StressTestor@users.noreply.github.com>
This commit is contained in:
Alexandre Teixeira
2026-08-14 08:10:30 +01:00
committed by GitHub
co-authored by RaresKeY StressTestor
parent b52296471b
commit c4369305f0
46 changed files with 11418 additions and 1006 deletions
+18 -6
View File
@@ -510,7 +510,12 @@ async def do_manage_settings(content: str, owner: Optional[str] = None) -> Dict:
# set/get/list/delete operate on the REAL app settings (the same store
# the Settings panel writes), so changing a model / voice / search
# engine / reminder channel from chat actually takes effect.
from src.settings import load_settings, save_settings, DEFAULT_SETTINGS
from src.settings import (
DEFAULT_SETTINGS,
RETIRED_SETTING_KEYS,
load_settings,
save_settings,
)
# Secrets/credentials the agent must NOT write: kept read-only (masked)
# so API keys never flow through chat. User sets these in the panel.
@@ -562,6 +567,9 @@ async def do_manage_settings(content: str, owner: Optional[str] = None) -> Dict:
return k2
return _ALIASES_SET.get(k2, (k or "").strip())
def _is_managed_key(key):
return key in DEFAULT_SETTINGS and key not in RETIRED_SETTING_KEYS
_ENUMS = {
"image_quality": ["low", "medium", "high"],
"reminder_channel": ["browser", "email", "ntfy", "webhook"],
@@ -624,14 +632,18 @@ async def do_manage_settings(content: str, owner: Optional[str] = None) -> Dict:
if action == "list":
s = load_settings()
shown = {k: _mask(k, v) for k, v in s.items() if k in DEFAULT_SETTINGS and not isinstance(v, dict)}
shown = {
k: _mask(k, v)
for k, v in s.items()
if _is_managed_key(k) and not isinstance(v, dict)
}
return {"response": f"{len(shown)} settings (use get/set with a key)", "settings": shown, "exit_code": 0}
elif action == "get":
key = _resolve(args.get("key", ""))
if not key:
return {"error": "key is required", "exit_code": 1}
if key not in DEFAULT_SETTINGS:
if not _is_managed_key(key):
return {"error": f"Unknown setting '{args.get('key')}'. Use action='list' to see them.", "exit_code": 1}
val = load_settings().get(key, DEFAULT_SETTINGS.get(key))
return {"response": f"{key} = {_mask(key, val)}", "value": _mask(key, val), "exit_code": 0}
@@ -642,11 +654,11 @@ async def do_manage_settings(content: str, owner: Optional[str] = None) -> Dict:
if not raw:
return {"error": "key is required", "exit_code": 1}
key = _resolve(raw)
if key not in DEFAULT_SETTINGS:
if not _is_managed_key(key):
return {"error": f"Unknown setting '{raw}'. Use action='list' to see available settings.", "exit_code": 1}
if _is_secret(key):
return {"response": f"'{key}' is a credential/secret. For security I can't set it from chat. Open Settings and set it there.", "exit_code": 0}
# Structured settings (dicts/lists like keybinds, default_model_fallbacks)
# Structured settings (dicts/lists like keybinds or vision fallbacks)
# have no safe scalar coercion; _coerce would pass a bare string
# straight through and clobber the structure. Refuse them here; they're
# edited in their dedicated panels. (reset/delete still restore the
@@ -675,7 +687,7 @@ async def do_manage_settings(content: str, owner: Optional[str] = None) -> Dict:
elif action == "delete" or action == "reset":
key = _resolve(args.get("key", ""))
if key not in DEFAULT_SETTINGS:
if not _is_managed_key(key):
return {"error": f"Unknown setting '{args.get('key')}'.", "exit_code": 1}
if _is_secret(key):
return {"response": f"'{key}' is a credential. Reset it in the panel.", "exit_code": 0}