Add Execution Cost to Html Reports

docs/05_bring_your_own_model_provider.md

@@ -135,6 +135,46 @@ class MyImageQAProvider(ImageQAProvider):
 ```
 
 
+### Execution Cost Tracking
+
+The built-in VLM providers include default pricing for supported models. You can override the pricing on any provider by passing `input_cost_per_million_tokens` and `output_cost_per_million_tokens`:
+
+```python
+from askui import AgentSettings, ComputerAgent
+from askui.model_providers import AnthropicVlmProvider
+from askui.reporting import SimpleHtmlReporter
+
+with ComputerAgent(
+    reporters=[SimpleHtmlReporter()],
+    settings=AgentSettings(
+        vlm_provider=AnthropicVlmProvider(
+            model_id="claude-sonnet-4-6",
+            input_cost_per_million_tokens=3.0,
+            output_cost_per_million_tokens=15.0,
+        ),
+    ),
+) as agent:
+    agent.act("Open settings")
+```
+
+If you implement a fully custom `VlmProvider`, override the `pricing` property to enable cost tracking:
+
+```python
+from askui.model_providers import VlmProvider, ModelPricing
+
+class MyVlmProvider(VlmProvider):
+    @property
+    def pricing(self) -> ModelPricing | None:
+        return ModelPricing(
+            input_cost_per_million_tokens=1.0,
+            output_cost_per_million_tokens=5.0,
+        )
+
+    # ... rest of implementation
+```
+
+---
+
 ## Advanced: Injecting a Custom Client
 
 For full control over HTTP settings (timeouts, proxies, retries), you can inject a pre-configured client:

docs/08_reporting.md

@@ -32,6 +32,15 @@ This generates an HTML file (typically in the current directory) showing:
 SimpleHtmlReporter(output_dir="./execution_reports", filename="agent_run.html")
 ```
 
+### Execution Cost Tracking
+
+The HTML report automatically shows the estimated API cost when using a `VlmProvider` with pricing information. The built-in Anthropic and AskUI providers include default pricing for supported Claude models.
+
+The report will display:
+- Total estimated cost
+- Per-token rates used for the calculation
+- Input and output token breakdowns (as before)
+
 ### Custom Reporters
 
 Create custom reporters by implementing the `Reporter` interface:

src/askui/agent_base.py

@@ -75,7 +75,12 @@ def __init__(
         # Create conversation with speakers and model providers
         speakers = Speakers()
         _callbacks = list(callbacks or [])
-        _callbacks.append(UsageTrackingCallback(reporter=self._reporter))
+        _callbacks.append(
+            UsageTrackingCallback(
+                reporter=self._reporter,
+                pricing=self._vlm_provider.pricing,
+            )
+        )
         self._conversation = Conversation(
             speakers=speakers,
             vlm_provider=self._vlm_provider,

src/askui/model_providers/__init__.py

@@ -23,6 +23,7 @@
 from askui.model_providers.google_image_qa_provider import GoogleImageQAProvider
 from askui.model_providers.image_qa_provider import ImageQAProvider
 from askui.model_providers.vlm_provider import VlmProvider
+from askui.utils.model_pricing import ModelPricing
 
 __all__ = [
     "AnthropicImageQAProvider",
@@ -33,5 +34,6 @@
     "DetectionProvider",
     "GoogleImageQAProvider",
     "ImageQAProvider",
+    "ModelPricing",
     "VlmProvider",
 ]

src/askui/model_providers/anthropic_vlm_provider.py

@@ -16,6 +16,7 @@
 )
 from askui.models.shared.prompts import SystemPrompt
 from askui.models.shared.tools import ToolCollection
+from askui.utils.model_pricing import ModelPricing
 
 _DEFAULT_MODEL_ID = "claude-sonnet-4-6"
 
@@ -38,6 +39,11 @@ class AnthropicVlmProvider(VlmProvider):
             `\"claude-sonnet-4-6\"`.
         client (Anthropic | None, optional): Pre-configured Anthropic client.
             If provided, other connection parameters are ignored.
+        input_cost_per_million_tokens (float | None, optional): Override
+            cost in USD per 1M input tokens. Both cost params must be set
+            to override the built-in defaults.
+        output_cost_per_million_tokens (float | None, optional): Override
+            cost in USD per 1M output tokens.
 
     Example:
         ```python
@@ -60,6 +66,8 @@ def __init__(
         auth_token: str | None = None,
         model_id: str | None = None,
         client: Anthropic | None = None,
+        input_cost_per_million_tokens: float | None = None,
+        output_cost_per_million_tokens: float | None = None,
     ) -> None:
         self._model_id_value = (
             model_id or os.environ.get("VLM_PROVIDER_MODEL_ID") or _DEFAULT_MODEL_ID
@@ -72,12 +80,22 @@ def __init__(
                 base_url=base_url,
                 auth_token=auth_token,
             )
+        self._pricing = ModelPricing.for_model(
+            self._model_id_value,
+            input_cost_per_million_tokens=input_cost_per_million_tokens,
+            output_cost_per_million_tokens=output_cost_per_million_tokens,
+        )
 
     @property
     @override
     def model_id(self) -> str:
         return self._model_id_value
 
+    @property
+    @override
+    def pricing(self) -> ModelPricing | None:
+        return self._pricing
+
     @cached_property
     def _messages_api(self) -> AnthropicMessagesApi:
         """Lazily initialise the AnthropicMessagesApi on first use."""

src/askui/model_providers/askui_vlm_provider.py

@@ -17,6 +17,7 @@
 )
 from askui.models.shared.prompts import SystemPrompt
 from askui.models.shared.tools import ToolCollection
+from askui.utils.model_pricing import ModelPricing
 
 _DEFAULT_MODEL_ID = "claude-sonnet-4-6"
 
@@ -37,6 +38,11 @@ class AskUIVlmProvider(VlmProvider):
             `"claude-sonnet-4-6"`.
         client (Anthropic | None, optional): Pre-configured Anthropic client.
             If provided, `workspace_id` and `token` are ignored.
+        input_cost_per_million_tokens (float | None, optional): Override
+            cost in USD per 1M input tokens. Both cost params must be set
+            to override the built-in defaults.
+        output_cost_per_million_tokens (float | None, optional): Override
+            cost in USD per 1M output tokens.
 
     Example:
         ```python
@@ -58,18 +64,30 @@ def __init__(
         askui_settings: AskUiInferenceApiSettings | None = None,
         model_id: str | None = None,
         client: Anthropic | None = None,
+        input_cost_per_million_tokens: float | None = None,
+        output_cost_per_million_tokens: float | None = None,
     ) -> None:
         self._askui_settings = askui_settings or AskUiInferenceApiSettings()
         self._model_id_value = (
             model_id or os.environ.get("VLM_PROVIDER_MODEL_ID") or _DEFAULT_MODEL_ID
         )
         self._injected_client = client
+        self._pricing = ModelPricing.for_model(
+            self._model_id_value,
+            input_cost_per_million_tokens=input_cost_per_million_tokens,
+            output_cost_per_million_tokens=output_cost_per_million_tokens,
+        )
 
     @property
     @override
     def model_id(self) -> str:
         return self._model_id_value
 
+    @property
+    @override
+    def pricing(self) -> ModelPricing | None:
+        return self._pricing
+
     @cached_property
     def _messages_api(self) -> AnthropicMessagesApi:
         """Lazily initialise the AnthropicMessagesApi on first use."""

src/askui/model_providers/vlm_provider.py

@@ -10,6 +10,7 @@
 )
 from askui.models.shared.prompts import SystemPrompt
 from askui.models.shared.tools import ToolCollection
+from askui.utils.model_pricing import ModelPricing
 
 
 class VlmProvider(ABC):
@@ -43,6 +44,15 @@ class VlmProvider(ABC):
     def model_id(self) -> str:
         """The model identifier used by this provider."""
 
+    @property
+    def pricing(self) -> ModelPricing | None:
+        """Pricing information for this provider's model.
+
+        Returns ``None`` if no pricing information is available.
+        Override in subclasses to provide model-specific pricing.
+        """
+        return None
+
     @abstractmethod
     def create_message(
         self,

src/askui/models/shared/usage_tracking_callback.py

@@ -5,26 +5,64 @@
 from typing import TYPE_CHECKING
 
 from opentelemetry import trace
+from pydantic import BaseModel
 from typing_extensions import override
 
 from askui.models.shared.agent_message_param import UsageParam
 from askui.models.shared.conversation_callback import ConversationCallback
-from askui.reporting import NULL_REPORTER, Reporter
+from askui.reporting import NULL_REPORTER
 
 if TYPE_CHECKING:
     from askui.models.shared.conversation import Conversation
+    from askui.reporting import Reporter
     from askui.speaker.speaker import SpeakerResult
+    from askui.utils.model_pricing import ModelPricing
+
+
+class UsageSummary(BaseModel):
+    """Accumulated token usage and optional cost breakdown for a conversation.
+
+    Args:
+        input_tokens (int | None): Total input tokens sent to the API.
+        output_tokens (int | None): Total output tokens generated.
+        cache_creation_input_tokens (int | None): Tokens used for cache creation.
+        cache_read_input_tokens (int | None): Tokens read from cache.
+        input_cost (float | None): Computed input cost in `currency`.
+        output_cost (float | None): Computed output cost in `currency`.
+        total_cost (float | None): Sum of `input_cost` and `output_cost`.
+        currency (str | None): ISO 4217 currency code (e.g. ``"USD"``).
+        input_cost_per_million_tokens (float | None): Rate used to compute `input_cost`.
+        output_cost_per_million_tokens (float|None): Rate used to compute `output_cost`.
+    """
+
+    input_tokens: int | None = None
+    output_tokens: int | None = None
+    cache_creation_input_tokens: int | None = None
+    cache_read_input_tokens: int | None = None
+    input_cost: float | None = None
+    output_cost: float | None = None
+    total_cost: float | None = None
+    currency: str | None = None
+    input_cost_per_million_tokens: float | None = None
+    output_cost_per_million_tokens: float | None = None
 
 
 class UsageTrackingCallback(ConversationCallback):
     """Tracks token usage per step and reports a summary at conversation end.
 
     Args:
         reporter: Reporter to write the final usage summary to.
+        pricing: Pricing information for cost calculation. If ``None``,
+            no cost data is included in the usage summary.
     """
 
-    def __init__(self, reporter: Reporter = NULL_REPORTER) -> None:
+    def __init__(
+        self,
+        reporter: Reporter = NULL_REPORTER,
+        pricing: ModelPricing | None = None,
+    ) -> None:
         self._reporter = reporter
+        self._pricing = pricing
         self._accumulated_usage = UsageParam()
 
     @override
@@ -43,7 +81,40 @@ def on_step_end(
 
     @override
     def on_conversation_end(self, conversation: Conversation) -> None:
-        self._reporter.add_usage_summary(self._accumulated_usage.model_dump())
+        input_cost: float | None = None
+        output_cost: float | None = None
+        total_cost: float | None = None
+        currency: str | None = None
+        input_cost_per_million_tokens: float | None = None
+        output_cost_per_million_tokens: float | None = None
+        if self._pricing is not None:
+            input_tokens = self._accumulated_usage.input_tokens or 0
+            output_tokens = self._accumulated_usage.output_tokens or 0
+            input_cost = (
+                input_tokens * self._pricing.input_cost_per_million_tokens / 1e6
+            )
+            output_cost = (
+                output_tokens * self._pricing.output_cost_per_million_tokens / 1e6
+            )
+            total_cost = input_cost + output_cost
+            currency = self._pricing.currency
+            input_cost_per_million_tokens = self._pricing.input_cost_per_million_tokens
+            output_cost_per_million_tokens = (
+                self._pricing.output_cost_per_million_tokens
+            )
+        summary = UsageSummary(
+            input_tokens=self._accumulated_usage.input_tokens,
+            output_tokens=self._accumulated_usage.output_tokens,
+            cache_creation_input_tokens=self._accumulated_usage.cache_creation_input_tokens,
+            cache_read_input_tokens=self._accumulated_usage.cache_read_input_tokens,
+            input_cost=input_cost,
+            output_cost=output_cost,
+            total_cost=total_cost,
+            currency=currency,
+            input_cost_per_million_tokens=input_cost_per_million_tokens,
+            output_cost_per_million_tokens=output_cost_per_million_tokens,
+        )
+        self._reporter.add_usage_summary(summary)
 
     @property
     def accumulated_usage(self) -> UsageParam: