From zero to validated agent output in five steps. ValGuard sits between your app and LLM providers — rules run on every completion in microseconds, not seconds. For curl, SDKs, and webhooks see Quickstart.
Recommended first templates: Invoice extraction, RAG Q&A baseline, PII detection. Each pack is production-ready and runs deterministically — no LLM-as-judge latency.
Loading
In Account → API keys, create a live key. Store it securely — it is shown once. The in-dashboard launch wizard walks you through keys, optional provider vault (Growth+), playbooks, and a sandbox validation run.
Apply one of the three recommended starter packs below, or browse all system templates in the marketplace. Templates copy curated rules into your agent — no YAML required.
extraction
Structured invoice JSON with line items, tax math, ISO currency codes, and dates — the fastest path for finance workflows.
Best for: AP automation · ERP ingestion · OCR post-processing
valid_format · required_fields · numeric_relations · valid_iso_currency · valid_iso_date
rag
Catch truncation markers, unsafe URLs, repetition, and malformed citations before answers reach users.
Best for: Knowledge-base chat · Support copilots · Internal search assistants
no_truncation_markers · repetition_ratio_max · url_safety · markdown_link_well_formed · no_lorem_ipsum
compliance
Block common personal-data and secret patterns in model output — a quick compliance win with zero LLM judges.
Best for: Customer-facing chatbots · Support replies · Data-handling guardrails
pii_detection · pii_ssn_us · pii_pesel · secrets_detection · forbidden_patterns
Point your existing OpenAI-compatible client at the ValGuard validation layer. Set X-VG-Agent to the agent slug (usually default or a template mirror like invoice_extraction_agent).
export VG_PROXY="https://api.valguard.ai"
export VG_API_KEY="vg_live_YOUR_KEY"
export VG_AGENT="default"
# Required when PROXY_MOCK_UPSTREAM=false (BYOK):
export VG_PROVIDER_KEY="sk-..."
curl -s "$VG_PROXY/v1/chat/completions" \
-H "Authorization: Bearer $VG_API_KEY" \
-H "X-VG-Agent: $VG_AGENT" \
-H "X-Provider-Key: $VG_PROVIDER_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"openai/gpt-4o-mini","messages":[{"role":"user","content":"Return JSON with name and email"}]}'Start in shadow mode to log validation failures without blocking traffic. When rules look right, switch individual validators to block or reask in your selected agent's Validatorstab, or turn off shadow mode in the same agent's Settings tab. Open the agents list in Dashboard → Agents, then select the target agent.. See shadow mode docs.