diff --git a/gateway/builtin_hooks/boot_md.py b/gateway/builtin_hooks/boot_md.py
index c2868a1e6..332a28821 100644
--- a/gateway/builtin_hooks/boot_md.py
+++ b/gateway/builtin_hooks/boot_md.py
@@ -18,6 +18,7 @@ suppress delivery.
 """
 
 import logging
+import os
 import threading
 
 logger = logging.getLogger("hooks.boot-md")
@@ -42,17 +43,74 @@ def _build_boot_prompt(content: str) -> str:
     )
 
 
+def _load_boot_model() -> str:
+    """Read the configured default model for boot agent runs."""
+    try:
+        from hermes_cli.config import load_config
+
+        cfg = load_config() or {}
+        model_cfg = cfg.get("model", {})
+        if isinstance(model_cfg, str):
+            return model_cfg
+        if isinstance(model_cfg, dict):
+            return model_cfg.get("default") or model_cfg.get("model") or ""
+    except Exception as exc:
+        logger.debug("boot-md could not load configured model: %s", exc)
+    return ""
+
+
+def _resolve_boot_agent_kwargs() -> dict:
+    """Resolve model/provider credentials for the boot agent.
+
+    Gateway-created agents resolve runtime provider settings before constructing
+    AIAgent. The boot hook runs outside normal message handling, so it must do
+    the same explicitly instead of relying on AIAgent's legacy constructor
+    defaults.
+    """
+    model = _load_boot_model()
+    runtime = {}
+    try:
+        from hermes_cli.runtime_provider import resolve_runtime_provider
+
+        runtime = resolve_runtime_provider(
+            requested=os.getenv("HERMES_INFERENCE_PROVIDER"),
+        ) or {}
+    except Exception as exc:
+        logger.warning("boot-md could not resolve runtime provider: %s", exc)
+
+    if not model and runtime.get("provider"):
+        try:
+            from hermes_cli.models import get_default_model_for_provider
+
+            model = get_default_model_for_provider(runtime["provider"]) or ""
+        except Exception:
+            pass
+
+    return {
+        "model": model,
+        "api_key": runtime.get("api_key"),
+        "base_url": runtime.get("base_url"),
+        "provider": runtime.get("provider"),
+        "api_mode": runtime.get("api_mode"),
+        "command": runtime.get("command"),
+        "args": list(runtime.get("args") or []),
+        "credential_pool": runtime.get("credential_pool"),
+    }
+
+
 def _run_boot_agent(content: str) -> None:
     """Spawn a one-shot agent session to execute the boot instructions."""
     try:
         from run_agent import AIAgent
 
         prompt = _build_boot_prompt(content)
+        runtime_kwargs = _resolve_boot_agent_kwargs()
         agent = AIAgent(
             quiet_mode=True,
             skip_context_files=True,
             skip_memory=True,
             max_iterations=20,
+            **runtime_kwargs,
         )
         result = agent.run_conversation(prompt)
         response = result.get("final_response", "")
diff --git a/tests/gateway/test_boot_md.py b/tests/gateway/test_boot_md.py
new file mode 100644
index 000000000..3253ce3b2
--- /dev/null
+++ b/tests/gateway/test_boot_md.py
@@ -0,0 +1,91 @@
+"""Tests for the built-in BOOT.md gateway hook."""
+
+from unittest.mock import patch
+
+from gateway.builtin_hooks import boot_md
+
+
+def test_resolve_boot_agent_kwargs_uses_configured_runtime_provider(monkeypatch):
+    """BOOT.md agents should use the same configured model/provider as gateway chat.
+
+    The hook runs outside normal message handling, so it must explicitly resolve
+    runtime provider settings before constructing AIAgent.
+    """
+    from hermes_cli import config as hermes_config
+    from hermes_cli import runtime_provider
+
+    monkeypatch.setenv("HERMES_INFERENCE_PROVIDER", "openai-codex")
+    monkeypatch.setattr(
+        hermes_config,
+        "load_config",
+        lambda: {"model": {"default": "gpt-5.5"}},
+    )
+
+    def fake_resolve_runtime_provider(requested=None):
+        assert requested == "openai-codex"
+        return {
+            "provider": "openai-codex",
+            "api_key": "test-key",
+            "base_url": "https://chatgpt.com/backend-api/codex",
+            "api_mode": "codex_responses",
+            "command": None,
+            "args": (),
+            "credential_pool": object(),
+        }
+
+    monkeypatch.setattr(
+        runtime_provider,
+        "resolve_runtime_provider",
+        fake_resolve_runtime_provider,
+    )
+
+    kwargs = boot_md._resolve_boot_agent_kwargs()
+
+    assert kwargs["model"] == "gpt-5.5"
+    assert kwargs["provider"] == "openai-codex"
+    assert kwargs["api_key"] == "test-key"
+    assert kwargs["base_url"] == "https://chatgpt.com/backend-api/codex"
+    assert kwargs["api_mode"] == "codex_responses"
+    assert kwargs["args"] == []
+    assert kwargs["credential_pool"] is not None
+
+
+def test_run_boot_agent_passes_runtime_kwargs_to_ai_agent(monkeypatch):
+    """The boot hook should not rely on AIAgent constructor defaults."""
+    captured = {}
+
+    class FakeAgent:
+        def __init__(self, **kwargs):
+            captured.update(kwargs)
+
+        def run_conversation(self, prompt):
+            captured["prompt"] = prompt
+            return {"final_response": "[SILENT]"}
+
+    monkeypatch.setattr(
+        boot_md,
+        "_resolve_boot_agent_kwargs",
+        lambda: {
+            "model": "gpt-5.5",
+            "provider": "openai-codex",
+            "api_key": "test-key",
+            "base_url": "https://chatgpt.com/backend-api/codex",
+            "api_mode": "codex_responses",
+            "command": None,
+            "args": [],
+            "credential_pool": "pool",
+        },
+    )
+
+    with patch("run_agent.AIAgent", FakeAgent):
+        boot_md._run_boot_agent("Send a startup report.")
+
+    assert captured["quiet_mode"] is True
+    assert captured["skip_context_files"] is True
+    assert captured["skip_memory"] is True
+    assert captured["max_iterations"] == 20
+    assert captured["model"] == "gpt-5.5"
+    assert captured["provider"] == "openai-codex"
+    assert captured["api_mode"] == "codex_responses"
+    assert captured["credential_pool"] == "pool"
+    assert "Send a startup report." in captured["prompt"]