feat(automations): implement model eligibility checks for automation creation

- Added model eligibility checks to ensure automations can only use billable models (premium or BYOK).
- Introduced new API endpoint to report model eligibility status for search spaces.
- Updated frontend components to display eligibility alerts and disable creation options when models are not billable.
- Enhanced automation creation forms to reflect model eligibility, preventing users from submitting invalid configurations.
- Implemented server-side logic to capture and preserve model preferences across automation edits, ensuring consistent behavior during execution.

surfsense_backend/app/automations/services/__init__.py

@@ -3,14 +3,26 @@
 from __future__ import annotations
 
 from .automation import AutomationService, get_automation_service
+from .model_policy import (
+    AutomationModelPolicyError,
+    assert_automation_models_billable,
+    assert_models_billable,
+    get_automation_model_eligibility,
+    get_model_eligibility,
+)
 from .run import RunService, get_run_service
 from .trigger import TriggerService, get_trigger_service
 
 __all__ = [
+    "AutomationModelPolicyError",
     "AutomationService",
     "RunService",
     "TriggerService",
+    "assert_automation_models_billable",
+    "assert_models_billable",
+    "get_automation_model_eligibility",
     "get_automation_service",
+    "get_model_eligibility",
     "get_run_service",
     "get_trigger_service",
 ]

surfsense_backend/app/automations/services/automation.py

@@ -18,9 +18,15 @@ from app.automations.schemas.api import (
     AutomationUpdate,
     TriggerCreate,
 )
+from app.automations.schemas.definition.envelope import AutomationModels
+from app.automations.services.model_policy import (
+    AutomationModelPolicyError,
+    assert_automation_models_billable,
+    get_automation_model_eligibility,
+)
 from app.automations.triggers import get_trigger
 from app.automations.triggers.schedule import compute_next_fire_at
-from app.db import Permission, User, get_async_session
+from app.db import Permission, SearchSpace, User, get_async_session
 from app.users import current_active_user
 from app.utils.rbac import check_permission
 
@@ -37,6 +43,16 @@ class AutomationService:
         await self._authorize(
             payload.search_space_id, Permission.AUTOMATIONS_CREATE.value
         )
+        search_space = await self._assert_models_billable(payload.search_space_id)
+
+        # Snapshot the search space's current (already-validated) model prefs onto
+        # the definition so runs are insulated from later chat/search-space model
+        # changes. Captured ids are guaranteed billable by the check above.
+        payload.definition.models = AutomationModels(
+            agent_llm_id=search_space.agent_llm_id or 0,
+            image_generation_config_id=search_space.image_generation_config_id or 0,
+            vision_llm_config_id=search_space.vision_llm_config_id or 0,
+        )
 
         automation = Automation(
             search_space_id=payload.search_space_id,
@@ -105,9 +121,15 @@ class AutomationService:
         if "status" in data:
             automation.status = data["status"]
         if "definition" in data:
-            automation.definition = patch.definition.model_dump(
-                mode="json", by_alias=True
-            )
+            new_def = patch.definition.model_dump(mode="json", by_alias=True)
+            # Preserve the captured model snapshot across edits so a definition
+            # change never silently re-binds the automation to the current chat
+            # model selection. Backend-managed; survives whether or not the
+            # client round-trips ``models``.
+            existing_models = (automation.definition or {}).get("models")
+            if existing_models is not None:
+                new_def["models"] = existing_models
+            automation.definition = new_def
             automation.version += 1
 
         await self.session.commit()
@@ -143,6 +165,40 @@ class AutomationService:
             )
         return automation
 
+    async def model_eligibility(self, *, search_space_id: int) -> dict:
+        """Return whether a search space's models are billable for automations.
+
+        ``{"allowed": bool, "violations": [{kind, config_id, reason}, ...]}``.
+        """
+        await self._authorize(search_space_id, Permission.AUTOMATIONS_READ.value)
+        search_space = await self.session.get(SearchSpace, search_space_id)
+        if search_space is None:
+            raise HTTPException(
+                status_code=404, detail=f"search space {search_space_id} not found"
+            )
+        return get_automation_model_eligibility(search_space)
+
+    async def _assert_models_billable(self, search_space_id: int) -> SearchSpace:
+        """Reject creation when the search space's models aren't billable.
+
+        Automations may only use premium global models or user BYOK models; free
+        global models and Auto mode are blocked. Mirrors the runtime backstop in
+        ``agent_task`` so users can't save an automation that would fail to run.
+
+        Returns the loaded :class:`SearchSpace` so the caller can capture its
+        model prefs without a second DB read.
+        """
+        search_space = await self.session.get(SearchSpace, search_space_id)
+        if search_space is None:
+            raise HTTPException(
+                status_code=404, detail=f"search space {search_space_id} not found"
+            )
+        try:
+            assert_automation_models_billable(search_space)
+        except AutomationModelPolicyError as exc:
+            raise HTTPException(status_code=422, detail=str(exc)) from exc
+        return search_space
+
     async def _authorize(self, search_space_id: int, permission: str) -> None:
         await check_permission(
             self.session,

surfsense_backend/app/automations/services/model_policy.py

@@ -0,0 +1,173 @@
+"""Model-billing policy for automations.
+
+Automations run unattended, so every run must be **billable**: it may only use
+either a premium global model (``billing_tier == "premium"``) or a user-provided
+BYOK model (a positive config id pointing at a per-user/per-space DB row). Free
+global models and Auto mode are blocked, because Auto can dispatch to a free
+deployment and free models aren't metered in premium credits.
+
+Config id conventions (shared across chat / image / vision):
+- ``id == 0``  → Auto mode (``AUTO_MODE_ID`` / ``IMAGE_GEN_AUTO_MODE_ID`` /
+  ``VISION_AUTO_MODE_ID``). Blocked.
+- ``id < 0``   → global YAML/OpenRouter config. Allowed only if premium.
+- ``id > 0``   → user BYOK DB row. Always allowed.
+
+This module is the single source of truth used by both creation-time enforcement
+(``AutomationService.create`` and the ``create_automation`` chat tool) and the
+runtime backstop (``agent_task`` dependencies).
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Literal
+
+if TYPE_CHECKING:
+    from app.db import SearchSpace
+
+ModelKind = Literal["llm", "image", "vision"]
+
+_KIND_LABEL: dict[ModelKind, str] = {
+    "llm": "agent LLM",
+    "image": "image generation model",
+    "vision": "vision model",
+}
+
+
+def _is_premium_global(kind: ModelKind, config_id: int) -> bool:
+    """Return True if a negative (global) config id is a premium tier model."""
+    from app.config import config as app_config
+
+    cfg: dict | None = None
+    if kind == "llm":
+        from app.agents.new_chat.llm_config import load_global_llm_config_by_id
+
+        cfg = load_global_llm_config_by_id(config_id)
+    elif kind == "image":
+        cfg = next(
+            (
+                c
+                for c in app_config.GLOBAL_IMAGE_GEN_CONFIGS
+                if c.get("id") == config_id
+            ),
+            None,
+        )
+    else:  # vision
+        cfg = next(
+            (
+                c
+                for c in app_config.GLOBAL_VISION_LLM_CONFIGS
+                if c.get("id") == config_id
+            ),
+            None,
+        )
+
+    if not cfg:
+        return False
+    return str(cfg.get("billing_tier", "free")).lower() == "premium"
+
+
+def _classify(kind: ModelKind, config_id: int | None) -> tuple[bool, str]:
+    """Classify a resolved config id as allowed or blocked.
+
+    Returns ``(allowed, reason)``; ``reason`` is empty when allowed.
+    """
+    label = _KIND_LABEL[kind]
+
+    if config_id is None or config_id == 0:
+        return (
+            False,
+            f"The {label} is set to Auto mode. Automations require an explicit "
+            "premium model or your own (BYOK) model so every run is billable.",
+        )
+
+    if config_id > 0:
+        # Positive id → user-owned BYOK config. Always allowed.
+        return True, ""
+
+    # Negative id → global config. Allowed only if premium.
+    if _is_premium_global(kind, config_id):
+        return True, ""
+
+    return (
+        False,
+        f"The {label} is a free model. Automations can only use premium models "
+        "or your own (BYOK) models so every run is billable.",
+    )
+
+
+def get_model_eligibility(
+    *,
+    agent_llm_id: int | None,
+    image_generation_config_id: int | None,
+    vision_llm_config_id: int | None,
+) -> dict:
+    """Return ``{"allowed": bool, "violations": [...]}`` for explicit config ids.
+
+    The ID-based core shared by both the search-space path (creation/eligibility)
+    and the captured-snapshot path (runtime backstop). Each violation is
+    ``{"kind", "config_id", "reason"}``.
+    """
+    checks: list[tuple[ModelKind, int | None]] = [
+        ("llm", agent_llm_id),
+        ("image", image_generation_config_id),
+        ("vision", vision_llm_config_id),
+    ]
+
+    violations: list[dict] = []
+    for kind, config_id in checks:
+        allowed, reason = _classify(kind, config_id)
+        if not allowed:
+            violations.append({"kind": kind, "config_id": config_id, "reason": reason})
+
+    return {"allowed": not violations, "violations": violations}
+
+
+def get_automation_model_eligibility(search_space: SearchSpace) -> dict:
+    """Return ``{"allowed": bool, "violations": [...]}`` for a search space.
+
+    Used by the eligibility endpoint and the chat tool's early check. Thin
+    wrapper over :func:`get_model_eligibility`.
+    """
+    return get_model_eligibility(
+        agent_llm_id=search_space.agent_llm_id,
+        image_generation_config_id=search_space.image_generation_config_id,
+        vision_llm_config_id=search_space.vision_llm_config_id,
+    )
+
+
+class AutomationModelPolicyError(Exception):
+    """Raised when a search space's models are not billable for automations."""
+
+    def __init__(self, violations: list[dict]) -> None:
+        self.violations = violations
+        reasons = "; ".join(v["reason"] for v in violations)
+        super().__init__(
+            reasons or "Automations require premium or BYOK models for all model slots."
+        )
+
+
+def assert_models_billable(
+    *,
+    agent_llm_id: int | None,
+    image_generation_config_id: int | None,
+    vision_llm_config_id: int | None,
+) -> None:
+    """Raise :class:`AutomationModelPolicyError` if any explicit id is not billable.
+
+    The ID-based core used by the runtime backstop against an automation's
+    captured model snapshot.
+    """
+    result = get_model_eligibility(
+        agent_llm_id=agent_llm_id,
+        image_generation_config_id=image_generation_config_id,
+        vision_llm_config_id=vision_llm_config_id,
+    )
+    if not result["allowed"]:
+        raise AutomationModelPolicyError(result["violations"])
+
+
+def assert_automation_models_billable(search_space: SearchSpace) -> None:
+    """Raise :class:`AutomationModelPolicyError` if any model slot is not billable."""
+    result = get_automation_model_eligibility(search_space)
+    if not result["allowed"]:
+        raise AutomationModelPolicyError(result["violations"])