Independent Trust Layer

Is your AI provider honest?
Find out in one line of code.

The independent audit layer for LLM transit providers. Detect model substitution, compare by price & trust, verify what you actually pay for.

# npm install -g truthprobe-mcp
# Add to your AI tool's MCP config:

{
  "mcpServers": {
    "truthprobe": {
      "command": "truthprobe-mcp"
    }
  }
}
# Done. Ask your AI: "Is my provider honest?"
Audit: ONBackground: ON
Claude CodeClaude Code
Codex CLICodex CLI
CursorCursor
WindsurfWindsurf
HermesHermes
85-97%
Detection Accuracy
<50ms
Added Latency
18+
Providers Ranked
30s
Integration Time

Audit. Compare. Verify.

Three capabilities that keep your LLM provider accountable.

Free

Trust Audit

Passively verify your provider actually serves the model you pay for. Multi-signal detection catches downgrades with zero extra cost to you.

85-97% detection accuracy
Free

Provider Rankings

Compare transit providers by trust score, speed, and price — per model. Data from real probes, price crawlers, and community signals.

Updated daily with real data
Free

Full Transparency

See which providers are trustworthy, who has the best prices, and where the market stands — all backed by verifiable data.

Open methodology, public data
PREVIEW

Provider Reliability Index

Who's honest? We test so you don't have to.

#PROVIDERHONESTSPEEDSCOREStatus
1
OR
OpenRouter
US / Multi-provider
948891Trusted
2
CA
CloseAI Asia
CN / Enterprise
828684Acceptable
3
A2
API2D
CN / Aggregator
768279Acceptable
4
YW
云雾 API
CN / Budget
627668Flagged
5
V3
V-API (V36)
CN / Reseller
418248Avoid

Scores based on 100+ automated probes per provider. Updated weekly. Tests funded by TruthProbe.

PASSIVE DETECTION

Trust Audit

Passively verify your provider actually serves the model you pay for. Multi-signal detection catches downgrades with zero extra cost to you.

PASS — Model Verified
Requestedclaude-opus-4-7
Text complexity9.2/10 — matches Opus tier
Timing varianceNormal pace — consistent with Opus
Reasoning depth7.8/10 — top tier
Confidence96%
ALERT — Model Mismatch Detected
Requestedclaude-opus-4-7
Text complexity5.1/10 — below Opus baseline
Timing varianceAbnormal pace — too fast for Opus
Likely actual modelclaude-sonnet-4-6 (78% conf.)
ActionFlagged for review

85-97% detection accuracyMulti-signal detection runs on every request at zero cost to you.

EVERY REQUEST

What happens on each request

One integration. Both trust verification and cost optimization happen automatically.

Your Request
OpenAI-compatible call
TruthProbe
Trust auditFree
Cost optimizationPro
Response
Verified + metadata

What you see after every request:

Free user — audit only
pip install truthprobe
$ python app.py > Requesting claude-sonnet-4-6 via API2D... ╭─ TruthProbe ─────────────────────────╮ 🔍 AUDIT ✓ PASS (94%) requested: claude-sonnet-4-6 detected: claude-sonnet-4-6 provider: API2D — trust 92/100 ╰──────────────────────────────────────╯
Pro user — audit + savings
base_url = truthprobe.com/v1
$ python app.py > Requesting claude-sonnet-4-6 via API2D... ╭─ TruthProbe ─────────────────────────╮ 🔍 AUDIT ✓ PASS (94%) requested: claude-sonnet-4-6 detected: claude-sonnet-4-6 OPTIMIZED simple → routed route: sonnethaiku cost: $0.003 saved: $0.041 (−93%) ╰──────────────────────────────────────╯
This request saved you−93%

Two ways to use TruthProbe

Both free. MCP for AI coding tools, SDK for Python scripts.

RECOMMENDED

MCP Server — Recommended

Integrates directly into your AI tool. Zero code changes, continuous monitoring.

Claude CodeClaude CodeCodex CLICodex CLICursorCursorWindsurfWindsurfHermesHermes
5 built-in tools (verify, status, balance, report, ranking)
Background monitoring every 30 min
Zero config — auto-detects your provider
$ npm install -g truthprobe-mcp

Python SDK — For Developers

Audit your provider programmatically. Your API key never leaves your machine.

Passive trust detection on all calls
CLI reports & verification
Key stays local
Community rankings contribution
$ pip install truthprobe

After 47 requests, you'll see this

The SDK automatically logs every request and builds a full audit report. Zero config, zero cost.

$ truthprobe report --week7 days
Trust Score82 → 90 ↑
Daily Costavg ¥0.23/day
Suspicious %12% → 2% ↓
MonTueWedThuFriSatSun
Provider performing well, suspicious rate declining. 60% requests use sonnet — compare pricing on the ranking page.
$ truthprobe report
88
Trust
47
Requests
3
Flagged
sonnet-4-6
60%
gpt-4o
25%
haiku-4-5
15%
Real-time audit stream
✓ ¥0.032 │ claude-sonnet-4-6 │ 1.2s ✓ ¥0.018 │ gpt-4o │ 0.8s ⚠ ¥0.041 │ opus-4-6 → suspected swap │ 2.1s ✓ ¥0.005 │ claude-haiku-4-5 │ 0.4s ✓ ¥0.028 │ claude-sonnet-4-6 │ 1.1s
Anomaly signal distribution
Timing50%
Text quality50%

Get started

1

Install

npm install -g truthprobe-mcp
2

Add to MCP config

Add truthprobe to your AI tool's MCP settings (Claude Code, Codex CLI, Cursor, Windsurf, Hermes, etc.)

3

Restart your AI tool

Done! Ask your AI: "Is my provider honest?" — all tools are now available.

Start auditing your provider today.

Free forever. Install in 30 seconds, see results on first request.

Help us improve

Submit feedback or feature suggestions