Run an audit
Run a Standard or Artifact audit against supplied context, schema, code, config, or policy.
Audit AI and structured outputs against the source of truth.
Check whether an output follows the material it was based on before you trust or release it.
- Works with documents, policies, code diffs, tests, schemas, and requirements
- Flags unsupported claims, contradictions, omissions, and concrete mismatches
- Returns ALLOW, REVIEW, or BLOCK with a trust score and clear next step
Use Standard mode for prose and Artifact mode for code, schemas, JSON, SQL, configs, and other technical output.
Ideal for automated workflows that use OpenAI, Anthropic etc. for code review - then use Taplid to audit the review.
What is Taplid?
Taplid is an output validation and auditing platform. Give it the source of truth - such as a document, policy, code diff, test suite, schema, specification, or set of requirements - and the output produced from it. Taplid checks whether the output matches the supplied source, then returns an ALLOW, REVIEW, or BLOCK decision, a 0 to 100 trust score, specific issues, and a next step. Although Taplid is commonly used for AI-generated output, the audit is based on the supplied source and output, not on how the output was produced. Taplid is a decision-support gate; it does not guarantee that an output is universally accurate, safe, or free of contradictions.
Taplid evaluates only the evidence and requirements supplied to the audit. It can be used with AI-generated or non-AI output, but it does not replace testing, qualified professional review, or normal human oversight for critical work.
Taplid video demo
Three things Taplid does
Across the web app, SDK, HTTP API, CLI, MCP, and OpenAI-compatible integrations, Taplid allows the following operations.
Retrieve an audit
Retrieve a persisted public result by auditId. No API key is required; treat the audit ID like a share link.
Verify an audit
Verify the ES256 attestation, issuer, signed fields, and result hash for a saved audit. Token-only verification checks the signature and issuer.
How it works
- 1. Send the request payloadInclude the AI or structured output from your app, API, chatbot, or automated workflow.
- 2. Review trust signalsTaplid returns a trust score, highlights issues, and shows exactly how to fix them in seconds.
- 3. Act before releaseRoute responses to allow, review, or block before users or systems rely on them.
Example Taplid Audit result
REVIEWTrust: 58/100Audit Id: AUD-20260410-0000000123
The review claims the implementation is safe to merge, but the supplied context does not prove the retry handler prevents double-charging.
- Missing source support for a key safety claim.
- No cited idempotency guard or regression test shown.
- The response sounds confident despite weak evidence.
Next step: Add the exact idempotency code or test evidence, or soften the claim before merging.
What the decisions mean
Taplid is conservative by design. REVIEW means “check this before trusting it”, not “the answer is definitely wrong.”
High-risk issue found
Taplid found a high-risk problem such as a contradiction, unsafe instruction, fabricated support, or a claim that should not be used as-is.
Do not ship or rely on the response until the issue is fixed.
Check before relying
Taplid is not confident enough to pass or block it. The response may be missing evidence, making a strong claim, or touching something that needs human judgement.
This is the caution zone: check the highlighted issue before relying on the answer.
No major issue found
The response appears sufficiently supported for the audit performed, and Taplid did not identify a material contradiction or dangerous omission in the evidence checked.
Continue under your normal release policy. Critical or regulated work may still require testing, professional review, or human approval.
Taplid is a risk gate, not a truth oracle. It helps decide whether an AI output should be allowed, reviewed, or blocked before it reaches users or production systems.
Real examples - Taplid in action
Real cases where Taplid saved time or prevented bad decisions.
Confident answer → blocked
Without Taplid
- Looked correct at a glance
- No obvious red flags
With Taplid
- Flags unsupported claims
- Catches overconfidence vs evidence
Outcome: The wrong answer never went live.
Wasted time → instant answer
Without Taplid
- Digging through it manually
- No clear reason why it failed
With Taplid
- REVIEW (79) with diagnosis
- Exact issue and next step surfaced
Outcome: The answer was already in the audit - I just wasn't reading it.
Unclear what to fix → exact next step
Without Taplid
- Vague feedback
- Guessing what to change
With Taplid
- Clear issues identified
- Exact next step given
Outcome: No guesswork. The fix is clear in seconds.
Taplid doesn’t just flag issues - it stops you fixing the wrong thing.
Before and after Taplid
Before Taplid
- Fluent output that still contains reliability risk
- No consistent release decision for uncertain responses
- Manual guesswork by developers and reviewers
After Taplid
- Clear allow, review, or block decision
- Trust score with actionable issues
- Defined next action before release
Audit trail: Every run is saved with a unique audit ID for compliance, reporting, and historical review.
Where Taplid helps
Unsupported artifact claims
Flag claims not supported by the supplied code, document, policy, or source context.
AI hallucination checkerAI code reviews
Check whether AI review comments are actually supported by the supplied code or artifact.
AI code review auditAudit trail
Every audit generates a persistent record with decision trace, score, and context.
Fast deterministic artifact checks
Artifact mode is deterministic-first, using structured checks for code, SQL, configs, schemas, docs, JSON, and technical AI responses, with AI review added only where it improves confidence.
Schema checkerMCP server
Run Taplid directly inside Claude, Cursor, and any MCP client, so agents can verify implementation plans, code reviews, and other artifacts before acting.
Release gates
Apply consistent ALLOW, REVIEW, or BLOCK decisions before outputs are trusted.
Validate AI responsesPolicy-grounded answers
Check generated HR, support, finance, legal-style, and policy answers against the approved source material that governs them.
AI policy checkerOpenAI SDK drop-in
Works with OpenAI-compatible SDKs. Change one base URL. Verify AI responses before your app uses them.
CI/CD pipeline eval
Run Taplid checks in CI/CD and gate releases on trust scores. Every result returns clean JSON for agents, humans, and automated workflows.
Internal deployment
An internal deployment option keeps fast code reviews, schema checks, CI validation, and audit evidence secure inside your own network.
False-block control
Route uncertain blockers to REVIEW instead of stopping developers with weak evidence.
Schema checker, config validator, and AI output validator
Artifact mode checks structured outputs against the supplied spec, schema, policy, or source context. Use Taplid to validate AI-generated code, SQL, configs, schemas, API contracts, Docker, Kubernetes YAML, Terraform, security rules, and technical review claims before they reach production.
JSON Schema validator
Check whether generated JSON, object shapes, required fields, enum values, and closed-schema constraints match the supplied schema.
SQL schema checker
Validate SQL columns, types, NOT NULL rules, VARCHAR lengths, NUMERIC precision, filters, and query claims against the stated database contract.
Docker and Kubernetes checks
Check Dockerfiles, Kubernetes YAML, manifests, container users, ports, image tags, replicas, secrets, and deployment policy claims.
Code and framework validation
Validate TypeScript, JavaScript, React, Next.js, Angular, C#/.NET, and frontend claims against the supplied code or implementation plan.
Security and secrets rules
Catch unsupported claims about credentials, API keys, auth headers, public buckets, root containers, CORS, encryption, and secret handling.
API and policy contract checks
Validate API responses, HTTP/CORS rules, refund policies, pricing rules, entitlement logic, configuration values, and release criteria.
Common use cases
AI code reviews
Verify reviewer claims against the supplied diff, code, or implementation context.
Implementation plans
Check AI-generated plans before Claude, Codex, Cursor, or a developer acts on them.
Agent workflows
Use API, CLI, or MCP to audit artifacts before agents trigger downstream actions.
Release gates
Block unsupported claims before generated reports, docs, or review artifacts reach production.
Add Taplid to your AI workflow
Start in the web audit page, then use the same audit, retrieval, and verification flow from your SDK, API, CLI, CI pipeline, MCP agent, or OpenAI-compatible integration.
Try itNo signup required for first-use testing.
Developer setup
Use Taplid from your backend, terminal, CI pipeline, or MCP-capable agent.
FAQ
- What is Taplid?
- Taplid checks an output against the source material or requirements it was meant to follow and returns an ALLOW, REVIEW, or BLOCK decision with the issues found.
- What happens if you don't validate AI output?
- Unchecked AI can produce confident but incorrect, outdated, unsupported, or context-unsafe answers. Sending those answers to users can damage trust and lead to poor decisions.
- Is this only for hallucination detection?
- No. Taplid also helps with output validation, reliability checks, and safer review workflows before shipping AI output.
- Do I need an account to try it?
- No signup is required for first-use testing. You can run an audit and see results immediately.
Common AI reliability questions
What is AI hallucination?
AI hallucination is when a model gives a confident answer that is not reliably supported by facts or evidence.
How do you verify AI responses?
AI response verification means comparing material claims with the approved source of truth and routing unsupported or contradicted output before release.