8 AI models. One trusted output.
Run GPT-4o, Claude, Gemini, Groq, Cerebras, Mistral, OpenRouter, and Ollama against your code in parallel. Only findings that hold up across providers reach your report — with explicit confidence scores and disagreement detection.
Built as a second validation layer before release or client handoff.
Windows 10/11 64-bit·Portable .exe·BYOK·No install
If you ship code that AI helped write, single-model review hides risk. Multi-LLM consensus surfaces it.
When Cursor, Copilot, or ChatGPT writes large parts of your code, review speed goes up — and so does uncertainty. NexaVerify adds a cross-check before that uncertainty reaches production.
Add a validation pass when stakes are high — client handoff, production release, or sensitive code paths. Review confidence scores before shipping.
Justify audit fees with a professional multi-LLM consensus report. Concrete deliverable instead of vague "I reviewed it" assurance.
Best used before client delivery
Best used before production release
Best used after heavy AI-generated code
Best used on auth, shell, or runtime-sensitive logic
96% of developers don't fully trust AI-generated code. Only 48% verify systematically. NexaVerify is for the other 52% who know they should — but don't have time.
One LLM returns findings. Some are real. Some are hallucinations. You can't tell which is which — so you either trust everything (waste time on false positives) or ignore everything (miss real bugs).
Multiple LLMs analyze the same code independently. NexaVerify compares outputs — issues confirmed across providers get higher confidence. Isolated findings get lower. Disagreements are surfaced, not hidden.
Fastest sanity pass. Small checks, quick signal.
Best default. Speed, signal quality, and cost balanced.
Pre-delivery audits, client work, sensitive code.
Select project. NexaVerify builds a local view: files, chunks, stack, hotspots.
Multiple providers analyze in parallel. Consensus engine filters weak or isolated signals.
Truncated or malformed LLM responses are automatically repaired — bracket matching, fallback extraction, and structural validation before parsing.
HTML for humans. JSON for automation, archiving, run-to-run diffing.
Findings ranked by severity, weighted by how many providers agreed.
Single-provider: lower confidence. Multi-provider agreement: higher confidence.
When providers disagree on severity, it's surfaced instead of hidden.
HTML report for human review. JSON export for automation and diffing.
See which providers succeeded, which failed, and why — in every report.
Targeted review angles: bugs, security, performance, QA.
Bring your own API keys. No NEXADiag account. No proxy. Keys stay on your machine.
Gemini's free tier signs up in 30 seconds at aistudio.google.com/apikey — no phone needed. Or use GPT, Claude, Groq — any provider you already have.
Open the app, paste your key in Settings. Done. Key stays on your machine, never sent to NEXADiag servers.
Browse to your project, click Analyze. HTML report opens in your browser. That's it.
A real scan of the NexaVerify v1.6.0 codebase itself — 3 free-tier providers (Gemini, Groq, Cerebras) in parallel. 99 issues found, 0 critical.
Real scan · Balanced mode · 3 providers (Gemini, Groq, Cerebras) · Project: NexaVerify v1.6.0_DEV itself.
→ Open the actual live report ▼ Show full breakdownReal issues found in the NexaVerify source code — click "Open the actual live report" for all 99.
| File | Type | Severity | Detected by | Confidence |
|---|---|---|---|---|
| main.py:118 | bug | MEDIUM | gemini | 36% |
| _init_settings() not wrapped in try-except — crash before logging is initialized | ||||
| constants.py | bug | MEDIUM | cerebras | 36% |
| WINDOW_TITLE references undefined variables — NameError on import | ||||
| analysis_support_v160.py:124 | bug | MEDIUM | cerebras | 36% |
| Missing return value — callers receive None instead of AnalysisPipelineResult | ||||
| arbiter.py | bug | MEDIUM | cerebras | 36% |
| build_consensus_output doesn't return ConsensusOutput — callers receive None | ||||
| consensus.py:165 | bug | MEDIUM | cerebras | 36% |
| Wrong argument passed to _provider_weight — potential TypeError | ||||
This is what you get: real issues, real confidence scores, real provider tracking. Not a demo. Not a simulation.
From the same real v1.6.0 scan: each provider reports independently, the engine tracks who returned what, and any failure is flagged instead of silently dropped.
Multi-provider consensus with 8 AI engines. Quick / Balanced / Deep modes. HTML + JSON + SARIF 2.1.0 reports. Disagreement detection. Local-first with your API keys. Ollama offline support. Proof of Determinism.
Smarter consensus weighting. Better grouping of related issues. Improved performance on large codebases. Scan history and run-to-run comparison.
Early users directly influence priorities. Not a closed SaaS — a tool that grows with real workflows. Buyers today shape what ships next.
No hype roadmap. Only what is actively being built.
No subscription. No lock-in. No "starter plan" upsell traps.
Free tier: limited analyses/day · resets at local midnight · fewer files per scan · reduced provider slots.
| Feature | Free | Pro (€19) |
|---|---|---|
| AI providers | 3 (Gemini + Groq + Cerebras) | All 8 |
| Analyses per day | 3 | Unlimited |
| Files per scan | 10 | 200 |
| Consensus Engine | ✓ | ✓ |
| HTML + JSON reports | ✓ | ✓ |
| SARIF 2.1.0 export | ✓ | ✓ |
| Ollama local support | ✓ | ✓ |
| Proof of Determinism | ✓ | ✓ |
No review yet. Be the first to try it and tell me what works — and what doesn't.
Honest feedback > fake stars. Drop me a note via email or links below.
Real bugs, real fixes, no hype. Published while building the tool.