Gate · agent exit

Sentinel

Pre-execution checkpoint for AI agents. Call it before irreversible actions — transfers, publishes, deletes, tool calls with side effects — and get a machine-readable ALLOW, BLOCK, or UNCERTAIN with structured objections.

API root Docs Pricing Endpoint reference

Host sentinel.thoughtproof.ai
Endpoint POST /sentinel/verify
Auth X-API-Key or x402 (USDC)
Outcomes ALLOW · BLOCK · UNCERTAIN — fail-closed on non-ALLOW

Call example

curl -X POST https://sentinel.thoughtproof.ai/sentinel/verify \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{"claim":"Publish the weekly report","evidence":"Draft matches the approved outline.","mode":"output_synthesis"}'

Pay-per-call tiers: Sentinel on /pricing/. x402 (USDC) is accepted on the same path.

When to use it

Default agent exit Gate every step that can change the world. Most integrations start here and only escalate consequential decisions to RV or DQL.
High volume, low latency path Checkpoint and Standard tiers are built for frequent calls, not one-off audits.
Lifecycle-aware Verify before irreversible actions — not after the damage is done.

What you get

A bounded decision package goes in (mandate, action, evidence, provenance). A native verdict comes out — not a score that you still have to interpret into a stop/go.

BLOCK and UNCERTAIN both stop execution. ALLOW alone is never a settlement guarantee; compose with your own authorization and execution layers.

Escalate heavy reasoning checks to RV, plan/trace checks to PLV, or five-axis scoring to DQL. Category framing: what decision validation is.

Sentinel

Call example

When to use it

What you get

Related