> ## Documentation Index
> Fetch the complete documentation index at: https://docs.responsibleailabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Providers and Reference

> LLM provider wrappers (OpenAI, Anthropic, Gemini), observability, error handling, utility functions, और TypeScript types।

## LLM provider wrappers

Drop-in wrappers जो आपके existing LLM calls में automatic RAIL scoring add कर देते हैं। सभी wrappers `{ response, content, railScore, evaluation }` return करते हैं।

### OpenAI

```typescript theme={null}
import { RAILOpenAI } from '@responsible-ai-labs/rail-score';
import OpenAI from 'openai';

const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });
const railOpenAI = new RAILOpenAI(client, openai, {
  thresholds: { safety: 7.0 },
});

const result = await railOpenAI.chat({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Explain quantum computing simply." }],
});

console.log(result.content);            // LLM response text
console.log(result.railScore.score);    // RAIL score
console.log(result.evaluation);         // Full EvalResult
```

### Anthropic

```typescript theme={null}
import { RAILAnthropic } from '@responsible-ai-labs/rail-score';
import Anthropic from '@anthropic-ai/sdk';

const anthropic = new Anthropic();
const railAnthropic = new RAILAnthropic(client, anthropic, {
  thresholds: { safety: 7.0 },
});

const result = await railAnthropic.message({
  model: "claude-sonnet-4-6",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Explain quantum computing simply." }],
});

console.log(result.content);
console.log(result.railScore.score);
```

### Google Gemini

```typescript theme={null}
import { RAILGemini } from '@responsible-ai-labs/rail-score';
import { GoogleGenerativeAI } from '@google/generative-ai';

const genAI = new GoogleGenerativeAI(process.env.GOOGLE_API_KEY);
const model = genAI.getGenerativeModel({ model: "gemini-2.0-flash" });
const railGemini = new RAILGemini(client, model, {
  thresholds: { safety: 7.0 },
});

const result = await railGemini.generate("Explain quantum computing simply.");
console.log(result.content);
console.log(result.railScore.score);
```

## Observability

### Langfuse

```typescript theme={null}
import { RAILLangfuse } from '@responsible-ai-labs/rail-score';
import { Langfuse } from 'langfuse';

const langfuse = new Langfuse({ publicKey: "...", secretKey: "..." });
const railLangfuse = new RAILLangfuse(client, langfuse);

// Content evaluate करें और scores को Langfuse trace में push करें
const result = await railLangfuse.traceEvaluation("trace-id", "Content to evaluate");

// Existing evaluation result को trace में push करें
await railLangfuse.scoreTrace("trace-id", existingResult);
```

### Guardrail handler

```typescript theme={null}
import { RAILGuardrail } from '@responsible-ai-labs/rail-score';

const guardrail = new RAILGuardrail(client, {
  inputThresholds:  { safety: 7.0 },
  outputThresholds: { safety: 7.0, fairness: 7.0 },
});

const preResult = await guardrail.preCall("User message");
if (!preResult.allowed) {
  console.log("Input blocked:", preResult.failedDimensions);
}

const postResult = await guardrail.postCall("LLM response");
if (!postResult.allowed) {
  console.log("Output blocked:", postResult.failedDimensions);
}
```

## Error handling

```typescript theme={null}
import {
  AuthenticationError,
  InsufficientCreditsError,
  InsufficientTierError,
  ValidationError,
  ContentTooLongError,
  SessionExpiredError,
  ContentTooHarmfulError,
  RateLimitError,
  RAILBlockedError
} from '@responsible-ai-labs/rail-score';

try {
  const result = await client.eval({ content: "Content to evaluate" });
} catch (error) {
  if (error instanceof AuthenticationError) {
    console.error("Invalid API key");
  } else if (error instanceof InsufficientCreditsError) {
    console.error(`Need ${error.required} credits, have ${error.balance}`);
  } else if (error instanceof RateLimitError) {
    console.error(`Rate limited. Retry after ${error.retryAfter}s`);
  } else if (error instanceof ContentTooHarmfulError) {
    console.error("Content too harmful to regenerate (avg score < 3.0)");
  } else if (error instanceof SessionExpiredError) {
    console.error("Safe-regenerate session expired (15 min TTL)");
  } else if (error instanceof RAILBlockedError) {
    console.error(`Blocked by policy: ${error.policyMode}`);
  }
}
```

| Error                      | Status | कब होता है                            |
| -------------------------- | ------ | ------------------------------------- |
| `AuthenticationError`      | 401    | Invalid या missing API key            |
| `InsufficientCreditsError` | 402    | Credits कम हैं                        |
| `InsufficientTierError`    | 403    | Feature के लिए higher plan चाहिए      |
| `ValidationError`          | 400    | Invalid parameters                    |
| `ContentTooLongError`      | 400    | Content max length से ज़्यादा है      |
| `SessionExpiredError`      | 410    | Safe-regenerate session expire हो गया |
| `ContentTooHarmfulError`   | 422    | Content का avg score 3.0 से नीचे है   |
| `RateLimitError`           | 429    | Rate limit exceed हो गया              |
| `RAILBlockedError`         | —      | Policy engine ने content block किया   |

## Utility functions

```typescript theme={null}
import {
  getScoreLabel, getScoreColor, getScoreGrade, formatScore,
  formatDimensionName, normalizeDimensionName, resolveFrameworkAlias,
  validateWeights, normalizeWeights, calculateWeightedScore,
  isPassing, getDimensionsBelowThreshold, getLowestScoringDimension,
  getHighestScoringDimension, aggregateScores
} from '@responsible-ai-labs/rail-score';

getScoreLabel(8.5);                      // "Excellent"
getScoreColor(8.5);                      // "green"
getScoreGrade(8.5);                      // "A-"
formatScore(8.567, 2);                   // "8.57"
formatDimensionName("user_impact");      // "User Impact"
normalizeDimensionName("legal_compliance"); // "inclusivity"
resolveFrameworkAlias("ai_act");         // "eu_ai_act"

const weakAreas = getDimensionsBelowThreshold(result, 7.0);
const lowest = getLowestScoringDimension(result);
const stats = aggregateScores([result1, result2, result3]);
console.log(stats.averageScore, stats.minScore, stats.maxScore);
```

## TypeScript types

```typescript theme={null}
import type {
  // Client
  RailScoreConfig,

  // Evaluation
  EvalParams, EvalResult, EvalIssue, DimensionScore,
  Dimension, EvaluationMode, ContentDomain, ScoreLabel,

  // Safe Regeneration
  SafeRegenerateParams, SafeRegenerateResult, SafeRegenerateContinueParams,

  // Compliance
  ComplianceCheckSingleParams, ComplianceCheckMultiParams,
  ComplianceResult, MultiComplianceResult, ComplianceFramework,

  // Session & Policy
  SessionConfig, SessionMetrics, PolicyMode, PolicyConfig, MiddlewareConfig,

  // Observability
  GuardResult, RAILGuardrailConfig,
} from '@responsible-ai-labs/rail-score';
```

## आगे क्या देखें

<CardGroup cols={2}>
  <Card title="Evaluation API Reference" icon="code" href="/api-reference/evaluation">
    Full HTTP parameters और response schema।
  </Card>

  <Card title="Compliance API" icon="shield-check" href="/api-reference/compliance">
    GDPR, HIPAA, EU AI Act और बाकी।
  </Card>

  <Card title="Credits and Pricing" icon="coins" href="/getting-started/credits">
    Per endpoint और mode credit costs।
  </Card>

  <Card title="Integrations Overview" icon="puzzle-piece" href="/integrations/overview">
    सभी supported LLM providers और observability tools।
  </Card>
</CardGroup>