Learn Observyze
Step by Step

Every response includes additional headers for tracking:

npm install @observyze/sdk

Wrap your OpenAI client to automatically capture all requests:

import { ObservyzeClient, wrapOpenAI } from '@observyze/sdk';
import OpenAI from 'openai';

const nw = new ObservyzeClient({ apiKey: process.env.OBSERVYZE_API_KEY });

// Create and wrap your OpenAI client
const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });
const tracedOpenAI = wrapOpenAI(openai, nw);

// Use the traced client - all calls are automatically instrumented
const response = await tracedOpenAI.chat.completions.create({
  model: 'gpt-4o',
  messages: [{ role: 'user', content: 'Hello!' }]
});

Wrap your Anthropic client for Claude integration:

import { ObservyzeClient, wrapAnthropic } from '@observyze/sdk';
import { Anthropic } from '@anthropic-ai/sdk';

const nw = new ObservyzeClient({ apiKey: process.env.OBSERVYZE_API_KEY });

// Create and wrap your Anthropic client
const anthropic = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });
const tracedAnthropic = wrapAnthropic(anthropic, nw);

// Use the traced client
const response = await tracedAnthropic.messages.create({
  model: 'claude-3-5-sonnet-20241022',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Hello!' }]
});

For complete control, create custom spans and traces:

import { ObservyzeClient, SpanType } from '@observyze/sdk';

const nw = new ObservyzeClient({ apiKey: process.env.OBSERVYZE_API_KEY });

// Start a custom trace
const trace = await nw.startTrace({
  name: 'agent-workflow',
  projectId: 'my-project',
  metadata: { userId: 'user-123' }
});

// Add spans for different operations
const llmSpan = await nw.startSpan({
  name: 'llm-call',
  type: SpanType.LLM,
  input: { model: 'gpt-4o', messages: [...] },
  metadata: { temperature: 0.7 }
});

const toolSpan = await nw.startSpan({
  name: 'tool-execution',
  type: SpanType.TOOL,
  input: { tool: 'search', query: '...' }
});

// Complete spans
await llmSpan.complete({
  output: { content: 'LLM response...' },
  metadata: { tokens: 150 }
});

await toolSpan.complete({
  output: { results: [...] }
});

// Complete the trace
await trace.complete({ status: 'success' });

Navigate to any trace detail page and click "Score Hallucination" in the Inspector panel:

curl -X POST "https://api.observyze.com/api/v1/traces/tr_xxx/evaluate" \
  -H "Authorization: Bearer ob_xxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json"

# Response:
{
  "success": true,
  "score": 0.15,
  "details": {
    "reasoning": "The output accurately reflects...",
    "evaluated_by": "mistralai/mistral-7b-instruct:free"
  }
}

# With webhook (re-inject into agent)
curl -X POST "https://api.observyze.com/api/v1/traces/tr_xxx/replay" \
  -H "Authorization: Bearer ob_xxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "webhook_url": "https://your-agent.com/webhook",
    "modified_spans": [
      {
        "span_id": "span_123",
        "input": {"messages": [{"role": "user", "content": "New prompt"}]}
      }
    ]
  }"

# Without webhook (local simulation)
curl -X POST "https://api.observyze.com/api/v1/traces/tr_xxx/replay" \
  -H "Authorization: Bearer ob_xxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{"modified_spans": []}'

Navigate to the Safety page to configure circuit breakers:

# Get current config
curl "https://api.observyze.com/api/v1/optimization/circuit-breaker" \
  -H "Authorization: Bearer ob_xxxxxxxxxxxxxxxx"

# Update config
curl -X POST "https://api.observyze.com/api/v1/optimization/circuit-breaker" \
  -H "Authorization: Bearer ob_xxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "enabled": true,
    "thresholds": {
      "hallucination_score": 0.5,
      "safety_score": 0.5,
      "consecutive_failures": 3,
      "failure_window_minutes": 5
    },
    "action": "halt",
    "cooldown_seconds": 60
  }'

Click "Run Auto-Detect" on the Safety page to manually trigger detection:

# Get bounty history
curl "https://api.observyze.com/api/v1/bug-bounty/history" \
  -H "Authorization: Bearer ob_xxxxxxxxxxxxxxxx"

# Trigger auto-detect
curl -X POST "https://api.observyze.com/api/v1/bug-bounty/auto-detect" \
  -H "Authorization: Bearer ob_xxxxxxxxxxxxxxxx"

# Response:
{
  "detected": 2,
  "bounties": [
    {
      "bounty_id": "bty_xxx",
      "trace_id": "tr_xxx",
      "severity": "critical",
      "hallucination_score": 0.85
    }
  ]
}

Define variables in your prompts using {{variable_name}} syntax:

You are a helpful assistant for a {company_name}.

Your response style should be:
- Tone: {tone}
- Detail level: {detail_level}

User query: {query}

When using the prompt, provide values for company_name, tone, detail_level, and query.

Prompt Hub tracks all changes automatically: