feat: add fallback_max_context_tokens config for context compression (#7942)

- New config item fallback_max_context_tokens (default 128k)
- When max_context_tokens is 0 and model not in LLM_METADATAS,
  use fallback_max_context_tokens as the context window limit
- Unified global config under provider_settings, in truncate_and_compress section
- i18n: zh-CN, en-US, ru-RU

Co-authored-by: AstrBot <astrbot@container>

Author: Pleiades1726

astrbot/core/astr_main_agent.py

@@ -152,6 +152,8 @@ class MainAgentBuildConfig:
     This enforce max turns before compression"""
     dequeue_context_length: int = 1
     """The number of oldest turns to remove when context length limit is reached."""
+    fallback_max_context_tokens: int = 128000
+    """Fallback max context tokens. When max_context_tokens is 0 and the model is not in LLM_METADATAS, use this value."""
     llm_safety_mode: bool = True
     """This will inject healthy and safe system prompt into the main agent,
     to prevent LLM output harmful information"""
@@ -1367,6 +1369,11 @@ async def build_main_agent(
             provider.provider_config["max_context_tokens"] = model_info["limit"][
                 "context"
             ]
+        else:
+            # fallback: default to configured fallback value
+            provider.provider_config["max_context_tokens"] = (
+                config.fallback_max_context_tokens
+            )
 
     if event.get_platform_name() == "webchat":
         asyncio.create_task(_handle_webchat(event, req, provider))

astrbot/core/config/default.py

@@ -309,7 +309,7 @@ class ChatProviderTemplate(TypedDict):
     "model": "",
     "modalities": [],
     "custom_extra_body": {},
-    "max_context_tokens": 0,
+    "max_context_tokens": 128000,
 }
 
 """
@@ -3544,6 +3544,14 @@ class ChatProviderTemplate(TypedDict):
                             "provider_settings.agent_runner_type": "local",
                         },
                     },
+                    "provider_settings.fallback_max_context_tokens": {
+                        "description": "上下文窗口兜底值",
+                        "type": "int",
+                        "hint": "当 max_context_tokens 为 0 且模型不在内置元数据中时，使用此值作为上下文窗口大小。默认 128000。",
+                        "condition": {
+                            "provider_settings.agent_runner_type": "local",
+                        },
+                    },
                 },
                 "condition": {
                     "provider_settings.agent_runner_type": "local",

astrbot/core/pipeline/process_stage/method/agent_sub_stages/internal.py

@@ -107,6 +107,9 @@ async def initialize(self, ctx: PipelineContext) -> None:
         )
         if self.dequeue_context_length <= 0:
             self.dequeue_context_length = 1
+        self.fallback_max_context_tokens: int = settings.get(
+            "fallback_max_context_tokens", 128000
+        )
 
         self.llm_safety_mode = settings.get("llm_safety_mode", True)
         self.safety_mode_strategy = settings.get(
@@ -136,6 +139,7 @@ async def initialize(self, ctx: PipelineContext) -> None:
             llm_compress_provider_id=self.llm_compress_provider_id,
             max_context_length=self.max_context_length,
             dequeue_context_length=self.dequeue_context_length,
+            fallback_max_context_tokens=self.fallback_max_context_tokens,
             llm_safety_mode=self.llm_safety_mode,
             safety_mode_strategy=self.safety_mode_strategy,
             computer_use_runtime=self.computer_use_runtime,

dashboard/src/i18n/locales/en-US/features/config-metadata.json

@@ -272,6 +272,10 @@
         "llm_compress_provider_id": {
           "description": "Model Provider ID for Context Compression",
           "hint": "When left empty, will fall back to the 'Truncate by Turns' strategy."
+        },
+        "fallback_max_context_tokens": {
+          "description": "Fallback context window size",
+          "hint": "When max_context_tokens is 0 and the model is not in built-in metadata, use this value as the context window size. Default: 128000."
         }
       }
     },

dashboard/src/i18n/locales/ru-RU/features/config-metadata.json

@@ -272,6 +272,10 @@
                 "llm_compress_provider_id": {
                     "description": "Модель для сжатия контекста",
                     "hint": "Если не выбрано, произойдет откат к стратегии удаления сообщений."
+                },
+                "fallback_max_context_tokens": {
+                    "description": "Запасной размер окна контекста",
+                    "hint": "Если max_context_tokens равен 0 и модель отсутствует во встроенных метаданных, используется это значение. По умолчанию: 128000."
                 }
             }
         },

dashboard/src/i18n/locales/zh-CN/features/config-metadata.json

@@ -274,6 +274,10 @@
         "llm_compress_provider_id": {
           "description": "用于上下文压缩的模型提供商 ID",
           "hint": "留空时将降级为\"按对话轮数截断\"的策略。"
+        },
+        "fallback_max_context_tokens": {
+          "description": "上下文窗口兜底值",
+          "hint": "当 max_context_tokens 为 0 且模型不在内置元数据中时，使用此值作为上下文窗口大小。默认 128000。"
         }
       }
     },