text
stringlengths
0
59.1k
- `voltOpsClient` - VoltOps client instance (if provided)
- `runtime` - Metadata including `runId`, `startedAt`, `tags`
**Return format:**
```ts
// Full format:
runner: async ({ item }) => {
return {
output: "agent response",
metadata: { tokens: 150 },
traceIds: ["trace-id-1"], // Optional: trace IDs for observability
};
};
// Short format (just the output):
runner: async ({ item }) => {
return "agent response";
};
```
The runner can return the output directly or wrap it in an object with metadata and trace IDs.
### Optional Fields
#### `scorers`
Array of scoring functions to evaluate outputs. Each scorer compares the runner output against the expected value or applies custom logic.
**Basic usage with heuristic scorers:**
```ts
// These scorers don't require LLM/API keys
scorers: [scorers.exactMatch, scorers.levenshtein, scorers.numericDiff];
```
**With thresholds and custom IDs:**
```ts
scorers: [
{
id: "similarity-check",
scorer: scorers.levenshtein,
threshold: 0.8,
},
];
```
**With LLM-based scorers:**
```ts
import { createAnswerCorrectnessScorer } from "@voltagent/scorers";
import { openai } from "@ai-sdk/openai";
scorers: [
{
scorer: scorers.levenshtein, // Heuristic scorer
threshold: 0.8,
},
{
scorer: createAnswerCorrectnessScorer({
model: openai("gpt-4o-mini"), // LLM scorer requires model
}),
threshold: 0.9,
},
];
```
When a threshold is set, the item fails if `score < threshold`. Scorers without thresholds contribute to metrics but don't affect pass/fail status.
#### `passCriteria`
Defines overall experiment success. Can be a single criterion or an array.
**Mean score:**
```ts
passCriteria: {
type: "meanScore",
min: 0.9,
scorerId: "exactMatch", // optional - defaults to all scorers
severity: "error", // optional - "error" or "warn"
label: "Accuracy check", // optional - for reporting
}
```
**Pass rate:**
```ts
passCriteria: {
type: "passRate",
min: 0.95,
scorerId: "exactMatch",
}
```
**Multiple criteria:**
```ts
passCriteria: [