Skip to content
Xplore
Products · Integrity Verification engine · Live

Verify what agents claim. Not what they sound like.

Xplore Integrity makes agent work explainable: TrustGraph records what your agent actually did, builds a provenance graph, and verifies every claim against evidence — catching hallucinated numbers and fabricated sources before they reach a decision.

17
Verification checks in catalogue
16 deterministic
2 zones
Observed facts vs declared claims
Core model
4 levels
SDK to gateway integration depth
Honest trust cap
Passport
Multi-metric verdict per run
Not one number
Core principle

Two zones. One truth.

Facts captured mechanically vs claims that start at zero trust. Integrity is the deterministic comparison between them.

Observed — trust by construction

Tool calls, fetched bytes, API responses, prompts and configs — captured at the I/O boundary. The agent cannot forge these.

Declared — zero until verified

Sources cited, values claimed, artifacts published. Each must resolve into the observed zone or become a visible risk in the graph.

Agent runtime
shellfetchCRMMCP
Capture bus → journal → graph → verifier
Integrity passport
How it works

Five steps. Reproducible every time.

From raw agent I/O to a qualified integrity verdict — no randomness, no clock dependency, no model in the verification loop.

01CaptureSDK · bridge · gateway
02JournalAppend-only + CAS
03GraphDeterministic materialize
04VerifyPolicy per vertex type
05PassportQualified trust score
Node resolution · live
SOURCE_DECLsnippet_grounding
CLAIMvalue_grounding
CLAIMorigin_policy
ARTIFACTslot_sufficiency
RUNcapture_integrity

Checks bind to vertex types — not flat score lists. New policy = config, not code.

Verification

17 checks. 16 deterministic.

Source resolution, snippet grounding, value grounding, origin policy, arithmetic consistency, slot coverage — each localized to the node that failed.

Integration

Wrap around any agent. Trust scales with visibility.

Instrument your code for full capture, bridge structured logs, or proxy MCP traffic — each level carries an honest trust cap.

×1.0
Instrumented

SDK in your harness

×0.85
Log-structured

OpenAI trace · OpenClaw bridge

×0.7
Gateway

MCP proxy · OTel ingest

×0.4
Declared

Push + external re-check