Trustwise Rest API documentation

Calculates the carbon emissions (in kg CO₂ equivalent) associated with running AI workloads on cloud infrastructure. It provides both total carbon footprint and a breakdown by component (embodied_cpu, operational_cpu, and operational_gpu). Note that embodied GPU carbon impacts are currently not available via the API — please reach out to support@trustwise.ai for information on how Trustwise can help provide this information for your use case.

Carbon Components

• embodied_cpu: Carbon emissions from the manufacturing and lifecycle of CPU hardware
• operational_cpu: Carbon emissions from CPU power consumption during operation
• operational_gpu: Carbon emissions from GPU power consumption during operation

Estimates Cost to provide transparency into expenses related to each LLM call, helping users effectively manage and optimize costs.

The Stability metric measures how similar an AI agent's responses are when given the same or similar inputs multiple times. It gives higher scores when responses stay consistent, even if asked by different personas or worded differently. This helps identify if an agent changes its answers unexpectedly.

The Trustwise Completion metric measures how well an AI response fulfils what the user asked for. It gives higher scores when the response fully addresses all parts of the user's request with detail. This metric checks if the AI gave a full and detailed answer to everything the user asked.

The Trustwise Refusal metric measures how clearly and firmly a response declines to answer a query. It gives higher scores when the response directly refuses without engaging with content. The metric evaluates both the question and response.

The Trustwise Adherence metric measures how well a response obeys specific rules or policies. It gives higher scores when the agent's response obeys the given policy, and lower scores when the agent breaks or ignores the policy. This metric helps ensure AI agents stay within their intended boundaries.

Execute a requested subset of metrics specified in the metrics array. Metrics are run asynchronously to optimize efficiency. Only one instance of each metric can be requested. When multiple metrics use the same input field (for example text), they share the single provided value. Note: include_chunk_scores is only applicable when context_relevancy is included in metrics. Other metrics will ignore this flag.

The Trustwise Answer Relevancy metric measures how well a response addresses the specific question or request that was asked. It gives higher scores to responses that directly tackle the main points of the query without going off-topic or missing key elements. It does not measure whether the answer is correct, only whether it attempts to address what was actually asked.

The Trustwise Clarity metric measures how easy text is to read. It gives higher scores to writing that contains words which are easier to read, and uses concise, self-contained sentences. It does not measure how well you understand the ideas in the text.

The Trustwise Context Relevancy metric measures how useful the provided context is for answering a specific query. It gives higher scores when the context contains information that directly helps answer the question being asked. It does not measure whether the context is interesting or detailed, only if it contains what's needed to address the query.

The Trustwise Faithfulness metric measures how well a response sticks to the information provided in the source context. It gives higher scores when responses accurately reflect what's actually in the source material without adding unsupported claims or leaving out important details. It does not measure whether the source information itself is correct or true.

The Trustwise Formality metric measures how professional or casual a text sounds. Higher scores mean the writing uses more professional language that you might find in business documents or academic papers. Lower scores mean the writing is more casual and conversational, like you might use with friends.

The Trustwise Helpfulness metric measures how useful a given text is. It gives higher scores to texts that fully explain a topic. Helpful responses provide clear, complete information.

The Trustwise PII metric detects Personally Identifiable Information in text. PII is any data that could be used to identify a specific person. The metric flags text that contains private information that should be protected.

The Trustwise Prompt Injection metric detects text that tries to override or bypass an AI system's built-in rules or safety measures. It identifies attempts to manipulate the AI into ignoring its guidelines or performing actions outside its intended use. Higher scores indicate a stronger attempt to manipulate the AI's behavior.

The Trustwise Sensitivity metric measures how much a specific topic appears in a text. It gives higher scores when a topic you care about is clearly present in the text. Each topic is scored separately, so adding more topics doesn't change the scores of other topics.

The Trustwise Simplicity metric measures how easy it is to understand the words in a text. It gives higher scores to writing that uses common, everyday words instead of special terms or complicated words. Simplicity looks at the words you choose, not how you put them together in sentences.

The Trustwise Tone metric shows the feeling or mood in a piece of writing. It looks at text and finds the three strongest tones from a list of choices. This helps you know how readers might feel when they read the text.

The Trustwise Toxicity metric measures how harmful, offensive, or hurtful text is to readers. It gives higher scores to writing that could upset people, make them feel unsafe, or spread hate. It looks for words that attack, insult, or threaten people or groups.