XAIP demo

Trust Evidence Before Delegation

Not another agent framework. A portable signed receipt layer underneath them.

Agents often choose tools without inspecting any historical execution evidence.

Three candidate servers can all answer a task. They look interchangeable until you see the receipts behind them.

Without XAIP

memory no evidence
no execution history available
playwright no evidence
no execution history available
unknown-server no evidence
no execution history available
Selection is blind. Each candidate looks the same.

With XAIP

memory evidence-backed
trust 0.996 · 226 signed receipts
playwright low_trust
trust 0.394 · 37 signed receipts · high_error_rate
unknown-server unscored
no execution evidence available · insufficient_data
Selected: memory
Selected using available execution evidence: trust 0.996, 226 signed receipts.

Snapshot Static values shown above. Live API fetch in progress…

XAIP does not make tools safe. It makes execution evidence visible before delegation.

Try it yourself

The Trust API is public and zero-auth. Reproduce the response above:

curl -X POST https://xaip-trust-api.kuma-github.workers.dev/v1/select \
  -H "Content-Type: application/json" \
  -d '{"task":"Store and retrieve user notes","candidates":["memory","playwright","unknown-server"]}'

Receipts come from MCP, LangChain.js, OpenAI-compatible tool-call loops, and other runtimes. The receipt format is provider-neutral.

Next

Live trust graph README & spec Browser playground