fix(context): scale replay cap with context window

nanobot/agent/loop.py

@@ -57,7 +57,11 @@ from nanobot.session.goal_state import (
     sustained_goal_active,
 )
 from nanobot.session.keys import UNIFIED_SESSION_KEY, session_key_for_channel
-from nanobot.session.manager import DEFAULT_REPLAY_MAX_MESSAGES, Session, SessionManager
+from nanobot.session.manager import (
+    Session,
+    SessionManager,
+    replay_max_messages_for_context,
+)
 from nanobot.utils.document import extract_documents, reference_non_image_attachments
 from nanobot.utils.helpers import image_placeholder_text
 from nanobot.utils.helpers import truncate_text as truncate_text_fn
@@ -201,7 +205,7 @@ class AgentLoop:
         timezone: str | None = None,
         session_ttl_minutes: int = 0,
         consolidation_ratio: float = 0.5,
-        max_messages: int = DEFAULT_REPLAY_MAX_MESSAGES,
+        max_messages: int = 0,
         hooks: list[AgentHook] | None = None,
         unified_session: bool = False,
         disabled_skills: list[str] | None = None,
@@ -292,8 +296,11 @@ class AgentLoop:
             llm_wall_timeout_for_session=lambda sk: runner_wall_llm_timeout_s(self.sessions, sk),
         )
         self._unified_session = unified_session
+        self._explicit_max_messages = max_messages > 0
         self._max_messages = (
-            max_messages if max_messages > 0 else DEFAULT_REPLAY_MAX_MESSAGES
+            max_messages
+            if self._explicit_max_messages
+            else replay_max_messages_for_context(self.context_window_tokens)
         )
         self._running = False
         self._mcp_servers = mcp_servers or {}
@@ -422,6 +429,7 @@ class AgentLoop:
         self.runner.provider = provider
         self.subagents.set_provider(provider, model)
         self.consolidator.set_provider(provider, model, context_window_tokens)
+        self._sync_replay_max_messages()
         self._provider_signature = snapshot.signature
         if publish_update and self._runtime_model_publisher is not None:
             self._runtime_model_publisher(
@@ -435,6 +443,10 @@ class AgentLoop:
             )
         logger.info("Runtime model switched for next turn: {} -> {}", old_model, model)
 
+    def _sync_replay_max_messages(self) -> None:
+        if not self._explicit_max_messages:
+            self._max_messages = replay_max_messages_for_context(self.context_window_tokens)
+
     def _refresh_provider_snapshot(self) -> None:
         if self._provider_snapshot_loader is None:
             return

nanobot/agent/tools/self.py

@@ -438,6 +438,9 @@ class MyTool(Tool, ContextAware):
         setattr(self._runtime_state, key, value)
         if key == "model":
             self._runtime_state._active_preset = None
+        sync_replay = getattr(self._runtime_state, "_sync_replay_max_messages", None)
+        if key == "context_window_tokens" and callable(sync_replay):
+            sync_replay()
         if key == "max_iterations" and hasattr(self._runtime_state, "_sync_subagent_runtime_limits"):
             self._runtime_state._sync_subagent_runtime_limits()
         self._audit("modify", f"{key}: {old!r} -> {value!r}")

nanobot/session/manager.py

@@ -27,7 +27,9 @@ from nanobot.utils.helpers import (
 from nanobot.utils.subagent_channel_display import scrub_subagent_announce_body
 
 FILE_MAX_MESSAGES = 2000
-DEFAULT_REPLAY_MAX_MESSAGES = 500
+MIN_REPLAY_MAX_MESSAGES = 120
+REPLAY_TOKENS_PER_MESSAGE = 100
+DEFAULT_REPLAY_MAX_MESSAGES = FILE_MAX_MESSAGES
 _MESSAGE_TIME_PREFIX_RE = re.compile(r"^\[Message Time: [^\]]+\]\n?")
 _LOCAL_IMAGE_BREADCRUMB_RE = re.compile(r"^\[image: (?:/|~)[^\]]+\]\s*$")
 _TOOL_CALL_ECHO_RE = re.compile(r'^\s*(?:generate_image|message)\([^)]*\)\s*$')
@@ -44,6 +46,15 @@ _FORK_VOLATILE_METADATA_KEYS = {
 }
 
 
+def replay_max_messages_for_context(context_window_tokens: int | None) -> int:
+    if not context_window_tokens or context_window_tokens <= 0:
+        return DEFAULT_REPLAY_MAX_MESSAGES
+    return min(
+        FILE_MAX_MESSAGES,
+        max(MIN_REPLAY_MAX_MESSAGES, context_window_tokens // REPLAY_TOKENS_PER_MESSAGE),
+    )
+
+
 def _sanitize_assistant_replay_text(content: str) -> str:
     """Remove internal replay artifacts that the model may have copied before.