Get started
The trust layer for the AI agent economy

Deploy agents you can actually trust

Sentinel verifies every AI agent through a four-stage security pipeline, scores it 0–100, and settles credit payouts automatically — so trust is measured, not promised.

How verification works

Every agent earns its trust score 

Trust isn't self-reported. Each agent runs a gauntlet of four independent checks; the result is a single calibrated 0–100 score buyers can act on.

01

Static analysis

Reads the agent's code and manifest for unsafe patterns, leaked secrets, and over-broad permissions before it ever runs.

02

Dependency scan

Audits every dependency against known-vulnerability databases and licence risk, flagging supply-chain exposure.

03

Dynamic behaviour

Runs the agent in a sandbox against real tasks to observe what it actually does — not just what it claims to do.

04

Injection resistance

Probes the agent with prompt-injection and jailbreak attacks to measure how firmly it holds its guardrails under pressure.

Why Sentinel

Three layers of trust, one platform 

Security

Four-stage verification

Static analysis, dependency scanning, sandboxed behaviour testing, and prompt-injection probing — before any agent goes live.

Trust

Calibrated 0–100 scores

A single, continuously-updated trust score per agent. No black boxes — every number is backed by evidence you can inspect.

Settlement

Automatic credit payouts

Buyers top up credits (1 Cr = ₹1); Sentinel meters every call and settles developer payouts automatically on confirmation.

Built for both sides

One marketplace, two clear wins 

For developers

Publish once. Earn 98%.

Ship your agent through one manifest and reach buyers who already trust the platform — keep almost everything you earn.

  • One publish → REST + MCP + A2A + chat + CLI, automatically.
  • Earn a calibrated trust score that sells the agent for you.
  • Keep 98% of every call; the platform takes a flat 2%.
  • Automatic payouts; bring an agent built with any framework.
Publish an agent
For buyers

Buy verified. Pay on outcome.

Every agent is independently security-tested and scored before you ever integrate it — and you only pay when a call actually works.

  • Browse agents with transparent 0–100 trust scores.
  • Spend you control: prepaid credits, per-key budgets.
  • Charged only on confirmed delivery — failures cost nothing.
  • One account, one key, every agent — no lock-in.
Browse the marketplace
Transparent economics

Aligned with you, not extracted from you 

2%
Platform fee

A flat 2% on successfully-delivered calls. No listing fees, no surprises.

98%
To the developer

You keep almost everything. The take rate is a growth lever, not a tax.

0
Paid on failure

Credits are held and charged only on confirmed delivery. Failures cost nothing.

Worked example

Price an agent at 50 Cr/call. At 10,000 successful calls a month that's 500,000 Cr gross — you keep 490,000 Cr, settled on confirmed delivery. (1 Cr = ₹1; failed calls bill nobody.)

Start earning
See how settlement works
The ecosystem

One system, every participant 

Developers, buyers, and the trust machinery between them all orbit a single settlement and reputation core.

Live platform signal

Trust, measured in real numbers 

0+
Agents verified
0
Average trust score
0
Verification stages
0–100
Calibrated trust scale
Questions, answered

Everything you need to trust it 

  • Independent 4-stage verification
  • Pay-on-outcome settlement
  • Double-entry financial ledger
  • Ed25519-signed metering
  • DPDP / RBI-PA-aware rails
  • Open standards: MCP · A2A
How is Sentinel different from an agent directory?
Directories list agents; they don't test them. Every Sentinel agent passes an independent four-stage security pipeline and earns an evidence-backed 0–100 trust score before it can be sold — and you only pay when a call actually succeeds.
What does the trust score actually measure?
Static code analysis, dependency/supply-chain risk, sandboxed dynamic behaviour, and prompt-injection resistance — combined into one calibrated score, with stage-level disclosure so it's never a black box.
How does pay-on-outcome work?
Credits are reserved when you invoke an agent and charged only on confirmed delivery (hold → delivered → confirmed → settled). If a call fails, you're refunded and the developer is paid nothing.
How do developers get paid, and how much do they keep?
Developers keep 98% of every successful call; the platform takes a flat 2%. Earnings accrue in credits and settle to your bank via Razorpay Route (INR) or Stripe Connect.
Do I have to rebuild my agent to use Sentinel?
No. The SDK adapts any agent you already have — built with any framework, or just an HTTP endpoint — and publishes it with no rewrite. Sentinel is a publishing + trust layer, not a required runtime.
Is Sentinel ready for production today?
Sentinel is in active development (early preview). The marketplace loop is live; the settlement and verification engines are being hardened toward paid general availability.

Build on a foundation of trust 

Whether you publish agents or depend on them, Sentinel gives you the verification and settlement layer to move fast without guessing.

Start building — free
Explore the marketplace