fix: register WaitForSubagentsTool in main AgentLoop so it's available in live sessions

Previously wait_for_subagents was only registered inside _run_subagent
(spawned orchestrators). Main conversation agent had no access to it.

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

Dockerfile.oauth

@@ -56,7 +56,7 @@ ENV PATH="/root/.local/bin:${PATH}"
 
 COPY pyproject.toml README.md LICENSE /app/
 COPY nanobot/ /app/nanobot/
-RUN uv pip install --system --no-cache --reinstall /app
+RUN uv pip install --system --no-cache --reinstall /app psycopg2-binary
 
 ENTRYPOINT ["nanobot"]
 CMD ["gateway"]

nanobot/agent/loop.py

@@ -2,6 +2,8 @@
 
 import asyncio
 import json
+import time
+from datetime import datetime
 from pathlib import Path
 from typing import Any
 
@@ -17,6 +19,7 @@ from nanobot.agent.tools.shell import ExecTool
 from nanobot.agent.tools.web import WebSearchTool, WebFetchTool
 from nanobot.agent.tools.message import MessageTool
 from nanobot.agent.tools.spawn import SpawnTool
+from nanobot.agent.tools.wait import WaitForSubagentsTool
 from nanobot.agent.tools.cron import CronTool
 from nanobot.agent.memory import MemoryStore
 from nanobot.agent.subagent import SubagentManager
@@ -76,6 +79,8 @@ class AgentLoop:
         )
 
         self._running = False
+        self._quota_cache: dict[str, Any] = {}  # {model: str, cached_at: float}
+        self._quota_cache_ttl: float = 300.0  # 5 minutes
         self._register_default_tools()
     
     def _register_default_tools(self) -> None:
@@ -105,6 +110,7 @@ class AgentLoop:
         # Spawn tool (for subagents)
         spawn_tool = SpawnTool(manager=self.subagents)
         self.tools.register(spawn_tool)
+        self.tools.register(WaitForSubagentsTool(manager=self.subagents))
         
         # Cron tool (for scheduling)
         if self.cron_service:
@@ -144,6 +150,94 @@ class AgentLoop:
         self._running = False
         logger.info("Agent loop stopping")
 
+    def _select_model_based_on_quota(self) -> str:
+        """Select Opus or Sonnet based on rolling weekly quota burn rate."""
+        # Check cache
+        now = time.time()
+        if self._quota_cache and (now - self._quota_cache.get("cached_at", 0)) < self._quota_cache_ttl:
+            return self._quota_cache["model"]
+
+        # Default models
+        OPUS = "claude-opus-4-6"
+        SONNET = "claude-sonnet-4-6"
+        TOLERANCE = 1.17  # 17% overage triggers downgrade
+
+        # Read rate limits
+        rate_limits_path = self.workspace / "memory" / "rate_limits.json"
+        if not rate_limits_path.exists():
+            logger.warning("rate_limits.json not found, defaulting to Sonnet")
+            return SONNET
+
+        try:
+            with open(rate_limits_path) as f:
+                limits = json.load(f)
+
+            actual_usage = limits.get("weekly_all_models")
+            weekly_reset = limits.get("weekly_reset")
+
+            if actual_usage is None or weekly_reset is None:
+                logger.warning("Rate limit data incomplete, defaulting to Sonnet")
+                return SONNET
+
+            # Calculate expected usage
+            actual_pct = actual_usage * 100
+            week_start = weekly_reset - (168 * 3600)
+            hours_elapsed = max(0, min((now - week_start) / 3600, 168))
+            expected_pct = (hours_elapsed / 168) * 100
+            threshold = expected_pct * TOLERANCE
+
+            # Decision logic
+            if actual_pct > threshold:
+                model = SONNET
+                logger.info(
+                    f"Quota: {actual_pct:.1f}% used, expected {expected_pct:.1f}%, "
+                    f"threshold {threshold:.1f}% → Sonnet"
+                )
+            else:
+                model = OPUS
+                logger.info(
+                    f"Quota: {actual_pct:.1f}% used, expected {expected_pct:.1f}%, "
+                    f"threshold {threshold:.1f}% → Opus"
+                )
+
+            # Cache decision
+            self._quota_cache = {"model": model, "cached_at": now}
+            return model
+
+        except Exception as e:
+            logger.error(f"Error checking quota: {e}, defaulting to Sonnet")
+            return SONNET
+
+    def _get_quota_status(self) -> str:
+        """Return human-readable quota status."""
+        rate_limits_path = self.workspace / "memory" / "rate_limits.json"
+        if not rate_limits_path.exists():
+            return "⚠️ No quota data available yet."
+
+        try:
+            with open(rate_limits_path) as f:
+                limits = json.load(f)
+
+            actual_pct = limits.get("weekly_all_models", 0) * 100
+            reset_ts = limits.get("weekly_reset", 0)
+            now = time.time()
+
+            hours_until_reset = (reset_ts - now) / 3600
+            week_start = reset_ts - (168 * 3600)
+            hours_elapsed = max(0, (now - week_start) / 3600)
+            expected_pct = (hours_elapsed / 168) * 100
+
+            model = self._select_model_based_on_quota()
+
+            return f"""📊 Quota Status:
+• Used: {actual_pct:.1f}% (expected {expected_pct:.1f}%)
+• Resets in: {hours_until_reset:.1f}h
+• Current model: {model}
+• Burn rate: {actual_pct / max(expected_pct, 0.01):.2f}x target"""
+
+        except Exception as e:
+            return f"⚠️ Error reading quota: {e}"
+
     async def _process_message(self, msg: InboundMessage, session_key: str | None = None) -> OutboundMessage | None:
         """
         Process a single inbound message.
@@ -177,7 +271,10 @@ class AgentLoop:
                                   content="🐈 New session started. Memory consolidated.")
         if cmd == "/help":
             return OutboundMessage(channel=msg.channel, chat_id=msg.chat_id,
-                                  content="🐈 nanobot commands:\n/new — Start a new conversation\n/help — Show available commands")
+                                  content="🐈 nanobot commands:\n/new — Start a new conversation\n/help — Show available commands\n/quota — Show quota status")
+        if cmd == "/quota":
+            status = self._get_quota_status()
+            return OutboundMessage(channel=msg.channel, chat_id=msg.chat_id, content=status)
 
         # Consolidate memory before processing if session is too large
         if len(session.messages) > self.memory_window:
@@ -196,15 +293,41 @@ class AgentLoop:
         if isinstance(cron_tool, CronTool):
             cron_tool.set_context(msg.channel, msg.chat_id)
         
+        # Prepend time-gap notice if >5 minutes since last user message
+        current_message = msg.content
+        last_user_ts = None
+        for m in reversed(session.messages):
+            if m.get("role") == "user":
+                last_user_ts = m.get("timestamp")
+                break
+        if last_user_ts:
+            try:
+                last_dt = datetime.fromisoformat(last_user_ts)
+                now_dt = datetime.now()
+                elapsed_seconds = (now_dt - last_dt).total_seconds()
+                if elapsed_seconds > 300:  # 5 minutes
+                    if elapsed_seconds < 3600:
+                        gap_str = f"{int(elapsed_seconds // 60)} minutes"
+                    elif elapsed_seconds < 86400:
+                        gap_str = f"{int(elapsed_seconds // 3600)} hours"
+                    else:
+                        gap_str = f"{int(elapsed_seconds // 86400)} days"
+                    current_message = f"[SYSTEM ANNOUNCEMENT: {gap_str} have elapsed since last user message; take this into account when replying to user]\n\n{msg.content}"
+            except (ValueError, TypeError):
+                pass  # Malformed timestamp — skip silently
+
         # Build initial messages (use get_history for LLM-formatted messages)
         messages = self.context.build_messages(
             history=session.get_history(),
-            current_message=msg.content,
+            current_message=current_message,
             media=msg.media if msg.media else None,
             channel=msg.channel,
             chat_id=msg.chat_id,
         )
 
+        # Select model based on quota
+        selected_model = self._select_model_based_on_quota()
+
         # Agent loop
         iteration = 0
         final_content = None
@@ -214,10 +337,11 @@ class AgentLoop:
             iteration += 1
 
             # Call LLM
+            logger.debug(f"Calling LLM with model={selected_model}, provider.thinking_budget={self.provider.thinking_budget}")
             response = await self.provider.chat(
                 messages=messages,
                 tools=self.tools.get_definitions(),
-                model=self.model
+                model=selected_model
             )
             
             # Handle tool calls
@@ -326,13 +450,16 @@ class AgentLoop:
         iteration = 0
         final_content = None
 
+        # Select model based on quota
+        selected_model = self._select_model_based_on_quota()
+
         while iteration < self.max_iterations:
             iteration += 1
 
             response = await self.provider.chat(
                 messages=messages,
                 tools=self.tools.get_definitions(),
-                model=self.model
+                model=selected_model
             )
             
             if response.has_tool_calls:
@@ -424,7 +551,9 @@ Respond with ONLY valid JSON, no markdown fences."""
                     {"role": "system", "content": "You are a memory consolidation agent. Respond only with valid JSON."},
                     {"role": "user", "content": prompt},
                 ],
-                model=self.model,
+                model="claude-haiku-4-5",
+                thinking_budget=0,
+                max_tokens=16384,
             )
             text = (response.content or "").strip()
             if text.startswith("```"):

nanobot/agent/subagent.py

@@ -15,6 +15,8 @@ from nanobot.agent.tools.registry import ToolRegistry
 from nanobot.agent.tools.filesystem import ReadFileTool, WriteFileTool, EditFileTool, ListDirTool
 from nanobot.agent.tools.shell import ExecTool
 from nanobot.agent.tools.web import WebSearchTool, WebFetchTool
+from nanobot.agent.tools.spawn import SpawnTool
+from nanobot.agent.tools.wait import WaitForSubagentsTool
 
 
 class SubagentManager:
@@ -40,11 +42,15 @@ class SubagentManager:
         self.provider = provider
         self.workspace = workspace
         self.bus = bus
-        self.model = model or provider.get_default_model()
+        # Default to Sonnet, not the provider default (Opus).
+        # Quota switching only affects the main agent's own requests, not SubagentManager.
+        # Explicit model overrides (e.g. Haiku workers) still take precedence.
+        self.model = model or "claude-sonnet-4-6"
         self.brave_api_key = brave_api_key
         self.exec_config = exec_config or ExecToolConfig()
         self.restrict_to_workspace = restrict_to_workspace
         self._running_tasks: dict[str, asyncio.Task[None]] = {}
+        self._task_results: dict[str, str] = {}
     
     async def spawn(
         self,
@@ -84,7 +90,7 @@ class SubagentManager:
         bg_task.add_done_callback(lambda _: self._running_tasks.pop(task_id, None))
         
         logger.info(f"Spawned subagent [{task_id}]: {display_label}")
-        return f"Subagent [{display_label}] started (id: {task_id}). I'll notify you when it completes."
+        return f"Subagent [{display_label}] started. Task ID: {task_id}"
     
     async def _run_subagent(
         self,
@@ -98,7 +104,7 @@ class SubagentManager:
         logger.info(f"Subagent [{task_id}] starting task: {label}")
         
         try:
-            # Build subagent tools (no message tool, no spawn tool)
+            # Build subagent tools (no message tool)
             tools = ToolRegistry()
             allowed_dir = self.workspace if self.restrict_to_workspace else None
             tools.register(ReadFileTool(allowed_dir=allowed_dir))
@@ -112,6 +118,10 @@ class SubagentManager:
             ))
             tools.register(WebSearchTool(api_key=self.brave_api_key))
             tools.register(WebFetchTool())
+            spawn_tool = SpawnTool(manager=self)
+            spawn_tool.set_context("subagent", origin["chat_id"])
+            tools.register(spawn_tool)
+            tools.register(WaitForSubagentsTool(manager=self))
 
             # Build messages with subagent-specific prompt
             system_prompt = self._build_subagent_prompt(task)
@@ -121,7 +131,7 @@ class SubagentManager:
             ]
             
             # Run agent loop (limited iterations)
-            max_iterations = 15
+            max_iterations = 50
             iteration = 0
             final_result: str | None = None
             
@@ -191,6 +201,13 @@ class SubagentManager:
         """Announce the subagent result to the main agent via the message bus."""
         status_text = "completed successfully" if status == "ok" else "failed"
 
+        # Child subagents (spawned by other subagents) store results silently.
+        # The parent orchestrator collects them via wait_for_subagents.
+        if origin["channel"] == "subagent":
+            self._task_results[task_id] = result
+            logger.debug(f"Subagent [{task_id}] stored result silently (child subagent)")
+            return
+
         announce_content = f"""[Subagent '{label}' {status_text}]
 
 Task: {task}
@@ -239,7 +256,6 @@ You are a subagent spawned by the main agent to complete a specific task.
 
 ## What You Cannot Do
 - Send messages directly to users (no message tool available)
-- Spawn other subagents
 - Access the main agent's conversation history
 
 ## Workspace
@@ -248,6 +264,25 @@ Skills are available at: {self.workspace}/skills/ (read SKILL.md files as needed
 
 When you have completed the task, provide a clear summary of your findings or actions."""
     
+    async def wait_for(self, task_ids: list[str]) -> str:
+        """Wait for specified child subagents to complete and return their results."""
+        tasks_to_wait = [
+            self._running_tasks[tid]
+            for tid in task_ids
+            if tid in self._running_tasks
+        ]
+        if tasks_to_wait:
+            await asyncio.gather(*tasks_to_wait, return_exceptions=True)
+
+        results = []
+        for tid in task_ids:
+            result = self._task_results.get(tid)
+            if result is not None:
+                results.append(f"[{tid}]:\n{result}")
+            else:
+                results.append(f"[{tid}]: No result found (invalid ID or task failed before storing)")
+        return "\n\n---\n\n".join(results)
+
     def get_running_count(self) -> int:
         """Return the number of currently running subagents."""
         return len(self._running_tasks)

nanobot/agent/tools/spawn.py

@@ -53,7 +53,7 @@ class SpawnTool(Tool):
                 },
                 "model": {
                     "type": "string",
-                    "description": "Optional model override for the subagent (e.g. 'claude-sonnet-4-20250514'). Defaults to the main agent's model.",
+                    "description": "Optional model override for the subagent (e.g. 'claude-haiku-4-5'). Defaults to the main agent's model.",
                 },
             },
             "required": ["task"],

nanobot/agent/tools/wait.py

@@ -0,0 +1,50 @@
+"""Wait-for-subagents tool for orchestrator subagents."""
+
+from typing import Any, TYPE_CHECKING
+
+from nanobot.agent.tools.base import Tool
+
+if TYPE_CHECKING:
+    from nanobot.agent.subagent import SubagentManager
+
+
+class WaitForSubagentsTool(Tool):
+    """
+    Tool to wait for child subagents to complete and collect their results.
+
+    Use this after spawning multiple subagents to wait for all of them
+    and get their results for synthesis.
+    """
+
+    def __init__(self, manager: "SubagentManager"):
+        self._manager = manager
+
+    @property
+    def name(self) -> str:
+        return "wait_for_subagents"
+
+    @property
+    def description(self) -> str:
+        return (
+            "Wait for one or more child subagents to complete and return their results. "
+            "Use this after spawning subagents to collect all results before synthesizing. "
+            "Blocks until all specified subagents finish."
+        )
+
+    @property
+    def parameters(self) -> dict[str, Any]:
+        return {
+            "type": "object",
+            "properties": {
+                "task_ids": {
+                    "type": "array",
+                    "items": {"type": "string"},
+                    "description": "List of task IDs to wait for (from spawn tool responses)",
+                },
+            },
+            "required": ["task_ids"],
+        }
+
+    async def execute(self, task_ids: list[str], **kwargs: Any) -> str:
+        """Wait for the specified subagents and return their results."""
+        return await self._manager.wait_for(task_ids)

nanobot/channels/telegram.py

@@ -92,6 +92,7 @@ class TelegramChannel(BaseChannel):
         BotCommand("start", "Start the bot"),
         BotCommand("new", "Start a new conversation"),
         BotCommand("help", "Show available commands"),
+        BotCommand("quota", "Show current quota status"),
     ]
     
     def __init__(
@@ -127,6 +128,7 @@ class TelegramChannel(BaseChannel):
         self._app.add_handler(CommandHandler("start", self._on_start))
         self._app.add_handler(CommandHandler("new", self._forward_command))
         self._app.add_handler(CommandHandler("help", self._forward_command))
+        self._app.add_handler(CommandHandler("quota", self._forward_command))
         
         # Add message handler for text, photos, voice, documents
         self._app.add_handler(

nanobot/providers/anthropic_oauth.py

@@ -225,6 +225,7 @@ class AnthropicOAuthProvider(LLMProvider):
         max_tokens: int = 4096,
         temperature: float = 0.7,
         tools: list[dict[str, Any]] | None = None,
+        thinking_budget_override: int | None = None,
     ) -> dict[str, Any]:
         """Make request to Anthropic API."""
         client = await self._get_client()
@@ -236,14 +237,15 @@ class AnthropicOAuthProvider(LLMProvider):
         }
 
         # Extended thinking: temperature must be 1 when enabled
-        if self.thinking_budget > 0:
+        effective_thinking = thinking_budget_override if thinking_budget_override is not None else self.thinking_budget
+        if effective_thinking > 0:
             payload["temperature"] = 1
             # max_tokens must exceed budget_tokens
-            if max_tokens <= self.thinking_budget:
-                payload["max_tokens"] = self.thinking_budget + 4096
+            if max_tokens <= effective_thinking:
+                payload["max_tokens"] = effective_thinking + 4096
             payload["thinking"] = {
                 "type": "enabled",
-                "budget_tokens": self.thinking_budget,
+                "budget_tokens": effective_thinking,
             }
         else:
             payload["temperature"] = temperature
@@ -255,9 +257,10 @@ class AnthropicOAuthProvider(LLMProvider):
             payload["tools"] = tools
 
         logger.info(
-            "Anthropic request: model=%s max_tokens=%d thinking=%s",
+            "Anthropic request: model={} max_tokens={} thinking={} tools={}",
             payload.get("model"), payload.get("max_tokens"),
             payload.get("thinking", "disabled"),
+            len(payload.get("tools", [])),
         )
 
         response = await client.post(
@@ -266,6 +269,39 @@ class AnthropicOAuthProvider(LLMProvider):
             json=payload,
         )
 
+        # Dump rate limit headers for analysis
+        try:
+            import datetime
+            import os
+            header_dump = {
+                "timestamp": datetime.datetime.utcnow().isoformat(),
+                "status_code": response.status_code,
+                "model": payload.get("model"),
+                "headers": dict(response.headers),
+            }
+            dump_path = "/root/.nanobot/workspace/api_headers.jsonl"
+            with open(dump_path, "a") as f:
+                f.write(json.dumps(header_dump) + "\n")
+            # Capture rate limit state for quota-based model switching
+            hdrs = response.headers
+            rate_limit_state = {
+                "updated_at": datetime.datetime.utcnow().isoformat(),
+                "model": payload.get("model"),
+                "weekly_all_models": float(hdrs["anthropic-ratelimit-unified-7d-utilization"]) if hdrs.get("anthropic-ratelimit-unified-7d-utilization") else None,
+                "weekly_sonnet": float(hdrs["anthropic-ratelimit-unified-7d_sonnet-utilization"]) if hdrs.get("anthropic-ratelimit-unified-7d_sonnet-utilization") else None,
+                "session_5h": float(hdrs["anthropic-ratelimit-unified-5h-utilization"]) if hdrs.get("anthropic-ratelimit-unified-5h-utilization") else None,
+                "weekly_reset": int(hdrs["anthropic-ratelimit-unified-7d-reset"]) if hdrs.get("anthropic-ratelimit-unified-7d-reset") else None,
+                "session_reset": int(hdrs["anthropic-ratelimit-unified-5h-reset"]) if hdrs.get("anthropic-ratelimit-unified-5h-reset") else None,
+                "binding_limit": hdrs.get("anthropic-ratelimit-unified-representative-claim"),
+                "sonnet_fallback": hdrs.get("anthropic-ratelimit-unified-fallback"),
+            }
+            state_path = "/root/.nanobot/workspace/memory/rate_limits.json"
+            os.makedirs(os.path.dirname(state_path), exist_ok=True)
+            with open(state_path, "w") as f:
+                json.dump(rate_limit_state, f, indent=2)
+        except Exception as e:
+            logger.warning("Rate limit header capture failed: {}", e)
+
         if response.status_code != 200:
             error_text = response.text
             raise Exception(f"Anthropic API error {response.status_code}: {error_text}")
@@ -279,6 +315,7 @@ class AnthropicOAuthProvider(LLMProvider):
         model: str | None = None,
         max_tokens: int = 4096,
         temperature: float = 0.7,
+        thinking_budget: int | None = None,
     ) -> LLMResponse:
         """Send chat completion request to Anthropic API."""
         model = model or self.default_model
@@ -293,6 +330,9 @@ class AnthropicOAuthProvider(LLMProvider):
         system, prepared_messages = self._prepare_messages(messages)
         anthropic_tools = self._convert_tools_to_anthropic(tools)
 
+        # Per-call thinking override (None = use instance default)
+        effective_thinking = self.thinking_budget if thinking_budget is None else thinking_budget
+
         try:
             response = await self._make_request(
                 messages=prepared_messages,
@@ -301,6 +341,7 @@ class AnthropicOAuthProvider(LLMProvider):
                 max_tokens=max_tokens,
                 temperature=temperature,
                 tools=anthropic_tools,
+                thinking_budget_override=effective_thinking,
             )
             return self._parse_response(response)
         except Exception as e:
@@ -341,25 +382,19 @@ class AnthropicOAuthProvider(LLMProvider):
                 ),
             }
 
-        # Log usage and thinking info
-        if thinking_blocks:
-            thinking_chars = sum(len(b.get("thinking", "")) for b in thinking_blocks)
+        stop_reason = response.get("stop_reason", "end_turn")
+        thinking_chars = sum(len(b.get("thinking", "")) for b in thinking_blocks) if thinking_blocks else 0
         logger.info(
-                "Anthropic response: %d thinking block(s) (%d chars), "
-                "input=%d output=%d tokens",
-                len(thinking_blocks), thinking_chars,
-                usage.get("prompt_tokens", 0), usage.get("completion_tokens", 0),
-            )
-        else:
-            logger.info(
-                "Anthropic response: no thinking blocks, input=%d output=%d tokens",
+            "Anthropic response: stop={} tool_calls={} thinking={} chars, "
+            "input={} output={} tokens",
+            stop_reason, len(tool_calls), thinking_chars,
             usage.get("prompt_tokens", 0), usage.get("completion_tokens", 0),
         )
 
         return LLMResponse(
             content=text_content or None,
             tool_calls=tool_calls,
-            finish_reason=response.get("stop_reason", "end_turn"),
+            finish_reason=stop_reason,
             usage=usage,
             reasoning_content=thinking_blocks or None,
         )

nanobot/providers/base.py

@@ -48,6 +48,7 @@ class LLMProvider(ABC):
         model: str | None = None,
         max_tokens: int = 4096,
         temperature: float = 0.7,
+        thinking_budget: int | None = None,
     ) -> LLMResponse:
         """
         Send a chat completion request.

nanobot/providers/litellm_provider.py

@@ -106,6 +106,7 @@ class LiteLLMProvider(LLMProvider):
         model: str | None = None,
         max_tokens: int = 4096,
         temperature: float = 0.7,
+        thinking_budget: int | None = None,
     ) -> LLMResponse:
         """
         Send a chat completion request via LiteLLM.