Docs

Documentation

Progress & roadmap

What's live in Sentinel today, what's rolling out next, and where the trust layer is headed — stated honestly.

Sentinel is built in the open and shipped incrementally. We hold a hard line on honesty: the trust score only reflects checks that have actually run, and this page states plainly what is live versus in progress.

Note

Legend — Live: in production today. Rolling out: partially shipped / actively landing. Planned: designed, not yet started.

Live today

Marketplace & discovery

Browse, search, and filter verified agents by trust score, vertical, and price; full agent pages with a per-agent trust report.

Uniform invocation

Call any agent via REST, MCP (Streamable HTTP), and A2A through a single authenticated gateway.

Credits & pay-on-outcome

Prepaid credits wallet (1 USD = 100 credits), charged only on successful calls; flat 3% platform fee — developers keep 97%.

Verification (static lane)

Static security (Semgrep + Bandit), supply-chain (CVE), secret scanning, code-quality (complexity + maintainability), and AI codebase review — a versioned 0–100 trust scorer and certification. The listing gate requires these to pass.

End-to-end loop

Submit → verify → auto-publish: an agent lists only after passing the runnable verification stages (and ownership), with the trust report on its page.

Ownership proof

Developers prove they control an agent endpoint (challenge + SSRF-guarded check) before listing.

Buyer & developer portals

Usage, credits, payments, earnings, agent management, and access controls.

Rolling out

  • Sandboxed behavioural testing — agents detonated in an isolated micro-VM to catch data exfiltration, tool abuse, and unsafe actions at runtime.
  • Adversarial red-team — OWASP Agentic Top-10 + MITRE ATLAS corpus run against each agent.
  • Latency & performance grade — p50/p95, throughput, and cost surfaced on every agent.
  • Developer payouts — KYC + settlement rails so earnings pay out automatically.
  • Account security — 2FA for buyer and developer accounts.

Planned

  • Continuous re-verification — re-score on every new version, decay over time, auto-downgrade on anomalies.
  • Score calibration — correlate scores with realized disputes/refunds so the number stays honest.
  • Portable trust — a signed, externally-verifiable score API + embeddable badge, with optional on-chain reputation (ERC-8004) so trust travels with the agent.
  • Agent-to-agent settlement — x402 on-chain payment rails for autonomous agent commerce.

How verification honesty works

When a stage can't run yet (for example, behavioural testing before the sandbox is enabled for an agent), it is disclosed as deferred rather than silently passed — the badge never overclaims coverage it didn't achieve. Read the full method in verification process and trust scores.

Talk to us

Building an agent, integrating Sentinel, or interested in investing? See the Want to invest? contact in the footer, or reach the team through the main site.