n8n-nodes-valiqor

Community nodes for n8n that add AI safety checking, failure analysis, and LLM evaluation to your workflows using Valiqor.

n8n is a fair-code licensed workflow automation platform.

Nodes

| Node | Description | |------|-------------| | Valiqor Safety Check | Check LLM input/output pairs for safety violations across 23 categories (prompt injection, PII, violence, hate speech, and more) | | Valiqor Failure Analysis | Diagnose why an LLM interaction failed — root cause, severity, evidence, and recommendations | | Valiqor Eval | Score LLM output quality (hallucination, answer relevance, factual accuracy, coherence, and more) |

Installation

Community Nodes (recommended)

1. Go to Settings → Community Nodes in your n8n instance
2. Click Install a community node
3. Enter n8n-nodes-valiqor
4. Click Install

Manual

npm install n8n-nodes-valiqor

Credentials

You need a Valiqor API key to use these nodes.

1. Get your API key at app.valiqor.com/api-keys
2. In n8n, go to Credentials → New Credential → Valiqor API
3. Paste your API key (starts with vq_)
4. Leave the Base URL as https://api.valiqor.com (default)

Don't have an account? Sign up free at app.valiqor.com — no credit card required.

Node Details

Valiqor Safety Check

Runs a security audit on LLM input/output pairs. Automatically checks all 23 safety violation categories.

Input: Items with user_input and assistant_response fields (field names configurable)

Output: Each item gets a valiqor_safety object with per-item verdicts and category names:

{
  "valiqor_safety": {
    "is_safe": false,
    "status": "unsafe",
    "input_safety": "safe",
    "top_risk": "Violence",
    "unsafe_output_categories": ["Violence", "Criminal Planning/Confessions"],
    "unsafe_input_categories": [],
    "risk_categories": ["Violence", "Criminal Planning/Confessions"],
    "category_verdicts": {
      "S1 - Violence": "fail",
      "S2 - Sexual": "pass",
      "S3 - Criminal Planning/Confessions": "fail",
      "S4 - Guns and Illegal Weapons": "pass",
      "S5 - Controlled/Regulated Substances": "pass",
      "S6 - Suicide and Self Harm": "pass",
      "S7 - Sexual (minor)": "pass",
      "S8 - Hate/Identity Hate": "pass",
      "S9 - PII/Privacy": "pass",
      "S10 - Harassment": "pass",
      "S11 - Threat": "pass",
      "S12 - Profanity": "pass",
      "S13 - Needs Caution": "pass",
      "S14 - Other": "pass",
      "S15 - Manipulation (Prompt Injection/Jailbreak)": "pass",
      "S16 - Fraud/Deception": "pass",
      "S17 - Malware": "pass",
      "S18 - High Risk Gov Decision Making": "pass",
      "S19 - Political/Misinformation/Conspiracy": "pass",
      "S20 - Copyright/Trademark/Plagiarism": "pass",
      "S21 - Unauthorized Advice": "pass",
      "S22 - Illegal Activity": "pass",
      "S23 - Immoral/Unethical": "pass"
    },
    "batch_id": "uuid",
    "total_items": 1,
    "unsafe_items": 1,
    "unsafe_rate": 100,
    "top_risk_category": "Violence"
  },
  "valiqor_links": {
    "dashboard": "https://app.valiqor.com",
    "api_keys": "https://app.valiqor.com/api-keys",
    "docs": "https://docs.valiqor.com",
    "github": "https://github.com/valiqor",
    "python_sdk": "pip install valiqor"
  }
}

Check individual categories via {{ $json.valiqor_safety.category_verdicts["S9 - PII/Privacy"] }} in n8n expressions.

Parameters: | Parameter | Description | Default | |-----------|-------------|---------| | Project Name | Valiqor project (auto-created) | n8n-safety-check | | User Input Field | Field containing user's message | user_input | | Assistant Response Field | Field containing assistant's response | assistant_response |

Valiqor Failure Analysis

This is the key differentiator. Nobody else has automated root-cause analysis in an n8n node. LangSmith/Langfuse tell you something failed; Valiqor tells you it was a wrong_tool_selected failure (severity: critical) with evidence from the conversation.

Input: Items with user_input and assistant_response fields. Optionally include context (for RAG) or tool_calls (for agents).

Output: Each item gets a valiqor_failure_analysis object with failures, warnings, and passes separated:

{
  "valiqor_failure_analysis": {
    "has_failures": true,
    "failure_count": 2,
    "warning_count": 1,
    "pass_count": 8,
    "max_severity": 4,
    "max_severity_label": "Critical",
    "primary_failure": "wrong_tool_selected",
    "primary_failure_name": "Wrong Tool Selected",
    "should_alert": true,
    "should_gate_ci": true,
    "needs_human_review": true,
    "failures": [
      {
        "subcategory": "wrong_tool_selected",
        "subcategory_name": "Wrong Tool Selected",
        "severity": 4,
        "severity_label": "Critical",
        "confidence": 0.92,
        "decision": "fail",
        "evidence": "Agent called web_search instead of database_query...",
        "bucket": "tool_errors",
        "bucket_name": "Tool Errors",
        "item_index": 0
      }
    ],
    "warnings": [
      {
        "subcategory": "partial_task_completion",
        "subcategory_name": "Partial Task Completion",
        "severity": 2,
        "severity_label": "Medium",
        "confidence": 0.78,
        "decision": "unsure",
        "evidence": "Agent completed 2 of 3 requested tasks...",
        "bucket": "task_quality",
        "bucket_name": "Task Quality",
        "item_index": 0
      }
    ],
    "passes": [],
    "run_id": "uuid",
    "duration_ms": 3200,
    "eval_metrics": { "hallucination": 0.05, "coherence": 0.92 },
    "security_flags": { "S1 - Violence": "pass", "S9 - PII/Privacy": "fail" }
  },
  "valiqor_links": {
    "dashboard": "https://app.valiqor.com",
    "api_keys": "https://app.valiqor.com/api-keys",
    "docs": "https://docs.valiqor.com",
    "github": "https://github.com/valiqor",
    "python_sdk": "pip install valiqor"
  }
}

Parameters: | Parameter | Description | Default | |-----------|-------------|---------| | Project Name | Valiqor project (auto-created) | n8n-failure-analysis | | User Input Field | Field containing user's message | user_input | | Agent Output Field | Field containing agent's response | assistant_response | | Context Field | Optional: field containing retrieval context | — | | Tool Calls Field | Optional: field containing tool calls array | — | | Feature Kind | Type of AI app (generic_llm, rag, agent, agentic_rag) | generic_llm | | Run Eval | Also run eval metrics alongside | true | | Run Security | Also run security audit alongside | false |

Valiqor Eval

Score LLM output quality using multiple metrics. Returns per-metric scores, an overall grade, and detailed explanations for each metric.

Input: Items with user_input and assistant_response fields.

Output: Each item gets a valiqor_eval object with scores AND explanations:

{
  "valiqor_eval": {
    "scores": {
      "hallucination": 0.05,
      "answerrelevance": 0.91
    },
    "overall_score": 0.85,
    "pass": true,
    "pass_threshold": 0.7,
    "quality_grade": "A",
    "run_id": "uuid",
    "total_items": 1,
    "evaluated_items": 1,
    "duration_ms": 4200,
    "eval_reasoning": "[hallucination] The response is grounded...\n\n[answerrelevance] Directly addresses...",
    "metric_details": {
      "hallucination": {
        "score": 0.05,
        "verdict": "pass",
        "explanation": "The response is factually grounded in the provided context..."
      },
      "answerrelevance": {
        "score": 0.91,
        "verdict": "pass",
        "explanation": "The response directly addresses the user's question..."
      }
    }
  },
  "valiqor_links": {
    "dashboard": "https://app.valiqor.com",
    "api_keys": "https://app.valiqor.com/api-keys",
    "docs": "https://docs.valiqor.com",
    "github": "https://github.com/valiqor",
    "python_sdk": "pip install valiqor"
  }
}

Parameters: | Parameter | Description | Default | |-----------|-------------|---------| | Project Name | Valiqor project (auto-created) | n8n-eval | | User Input Field | Field containing user's message | user_input | | Assistant Response Field | Field containing assistant's response | assistant_response | | Context Field | Optional: field for RAG context | — | | Expected Output Field | Optional: field for golden/expected response | — | | Metrics | Eval metrics to compute | hallucination, answerrelevance | | Run Name | Optional name for the eval run | Auto-generated | | Pass Threshold | Score threshold for pass/fail | 0.7 |

Available Metrics:

| Metric | Best For | |--------|----------| | hallucination | Detect fabricated facts in LLM responses | | answerrelevance | Check if the response answers the question | | factualaccuracy | Verify facts against provided context | | coherence | Assess logical flow and consistency | | fluency | Check grammar, readability, and naturalness | | completeness | Verify all parts of the question are answered | | contextprecision | RAG: precision of retrieved context | | contextrecall | RAG: recall of retrieved context | | retrieval | RAG: overall retrieval quality | | intentresolution | Did the agent resolve the user's intent? | | taskadherence | Does the output follow task instructions? | | toolcallaccuracy | Agents: were the right tools called correctly? | | moderation | Content safety and appropriateness |

Workflow Templates

Template 1: AI Agent Safety Pipeline

Automatically check every LLM response for safety violations and diagnose failures.

[Webhook Trigger] → [OpenAI Agent] → [Valiqor Safety Check] → [IF unsafe → Block + Slack Alert]
                                                              → [IF safe → Valiqor Failure Analysis]
                                                                          → [IF critical → Log + Alert]
                                                                          → [IF passed → Return Response]

Setup:

1. Add a Webhook trigger node
2. Connect to your OpenAI/LLM node
3. Add Valiqor Safety Check → check {{ $json.valiqor_safety.is_safe }}
4. For safe items, add Valiqor Failure Analysis → check {{ $json.valiqor_failure_analysis.max_severity }}
5. Route critical failures (severity ≥ 4) to PagerDuty/Slack alerts

Template 2: AI Quality Gate

Run evaluation metrics on a test dataset on a schedule.

[Schedule: daily] → [Fetch test dataset from Sheets] → [Valiqor Eval] → [IF score < threshold → Slack #ai-quality]
                                                                        → [IF passed → Log success]

Setup:

1. Add a Schedule trigger (e.g., daily at 9am)
2. Fetch your test dataset from Google Sheets / Airtable / database
3. Add Valiqor Eval with your quality metrics
4. Check {{ $json.valiqor_eval.pass }} and alert on regressions

Template 3: LLM Output Monitoring

Continuously monitor LLM outputs in production for safety + quality.

[Webhook: LLM output] → [Valiqor Safety Check] → [Valiqor Eval]
                                                 → [IF unsafe OR low quality → Slack + Google Sheets log]
                                                 → [IF OK → Continue]

API Endpoints Used

These nodes call existing Valiqor API endpoints — no custom backend changes needed:

| Node | Endpoint | |------|----------| | Safety Check | POST /v2/security/audit | | Failure Analysis | POST /v2/failure-analysis/analyze | | Eval | POST /v2/evaluate/ |

All endpoints use Authorization: Bearer <api_key> authentication.

Rate Limits & Quotas

Each node handles quota limits automatically. When your free tier limit is reached, you'll see a clear error message with a link to upgrade.

| Resource | Free Limit | |----------|-----------| | Evaluations | 100/month | | Security Audits | 50 items/month | | Failure Analysis | 50 items/month |

Upgrade at app.valiqor.com for higher limits.

Go Further with Valiqor

These n8n nodes cover the most common workflows. For deeper integration, use the Valiqor Python SDK:

pip install valiqor

The SDK gives you:

• Tracing — auto-instrument LangChain, OpenAI, and other LLM frameworks
• Batch evaluation — run evals on thousands of items with parallel processing
• CI/CD integration — add quality gates to your deployment pipeline
• Custom metrics — define your own evaluation criteria

Useful Links

| Resource | URL | |----------|-----| | Get API Key | app.valiqor.com/api-keys | | Dashboard | app.valiqor.com | | Documentation | docs.valiqor.com | | GitHub | github.com/valiqor | | Python SDK | pip install valiqor — PyPI | | Website | valiqor.com | | n8n Community Nodes | docs.n8n.io/integrations/community-nodes/ |

License

MIT