What we build
AI agents are making decisions and executing actions autonomously — trading, writing to memory, handing off to other agents, advising on compliance. The question is no longer whether agents can act, but whether their actions are defensible.
ThoughtProof runs four separate validation APIs. They are distinct products with their own endpoints, not tiers of one service:
Reasoning Verification (RV)
Adversarial multi-model verification. Independent models evaluate a claim, a dedicated red-team critic challenges the verdicts, and a synthesizer produces the final decision. For settlement conditions and high-stakes gating.
Plan-Level Verification (PLV)
Enterprise cascade verification. Checks agent output against a plan or evidence source. Cost-efficient cascade with optional EAS on-chain attestation. For compliance, audit trails, and enterprise pilots.
Sentinel
Lightweight pre-execution checkpoint. Verifies agent reasoning before irreversible actions. Four verification modes, lowest cost per call. The typical agent exit in a high-frequency loop.
Decision Quality Layer (DQL)
Five-axis decision quality scoring at POST /dql/verify. A separate product from Plan-Level Verification, for evaluating the quality of a decision rather than the faithfulness of a plan.
Standards and ecosystem
We believe verification infrastructure should be interoperable. ThoughtProof participates in and builds on open standards:
- ERC-8004 — Agent Registry
- ERC-8183 — Agentic Commerce. Active contributor to evaluator composition and settlement verification.
- ERC-8210 — Verification Evidence. Standards work for on-chain verification evidence.
- x402 — HTTP-native micropayments. RV, Sentinel and DQL support x402 payment in USDC.
- HOL — Hashgraph Online Partner Program (Registries track). Registries/trust scores complement decision validation; they do not replace it.
- HCS-25 — AI Trust Score. ThoughtProof contributed the output-verification adapter (per-decision independent verification at decision time) to the Hiero consensus specifications.
Hashgraph Online Partner Program
ThoughtProof is a member organization in the HOL Partner Program and participates in the Registries track. Registries and trust scores answer identity and standing questions; decision validation answers whether a specific action is justified — complementary layers, not a score product.
Open source
Our developer-facing tools are MIT-licensed and open:
- pot-cli — Local verification from the terminal
- pot-sdk — Integrate verification into agent code (npm)
- thoughtproof-mcp@0.3.2 — MCP connector; hero tools
verify_before_action/verify_decision(npm, listed on Glama). Setup on the DQL page.
Company
ThoughtProof LLC is a small, focused team. The four APIs are live and the developer tooling is auditable.
Raul Jäger is the founder and Managing Member. He set the company's direction on decision validation as a boundary outside the agent loop, and is the point of contact for pilots and technical questions — raul@thoughtproof.ai.
Registration details and the data protection notice are in the Impressum and Datenschutz.
Get in touch
Enterprise pilots, integration questions, or partnership inquiries: