The Problem
Your AI agents are making decisions. Executing contracts. Accessing data. Instructing other agents.
And when something goes wrong — a wrong action, a regulatory breach, a cross-organizational dispute — you have logs. Logs tell you what happened. They don't tell you what was authorized. Courts, regulators, and counterparties need the second thing.
How Truss makes every AI action defensible
Every AI action produces a mandate-bound record and chain of custody that holds across organizational boundaries. Three steps, every time.
Map the Regulations
Truss identifies which regulatory frameworks apply to this action in this context, evaluates the action against each obligation, and surfaces conflicts before execution.
Check the Mandate
Truss enforces the mandate against internal policies, what the GC authorized, what the compliance framework allows, and what the counterparty agreed to.
Seal the Record
Truss cryptographically seals the action record, links it to the mandate and the chain of custody. Courts, regulators, and counterparties each get a verifiable evidence package.
A rock-solid audit trail for your AI.
Truss automatically logs every AI decision into a tamper-proof record. The moment your AI tries to step outside its allowed rules, the system catches it and blocks the action.
- ✓Self-hosted or managed securely by us
- ✓Automatic tracking of changing global laws
- ✓Integrates with any AI framework in 5 lines of code
import { TrussClient } from '@tensflare/truss-sdk';
// Initialize with your organization's signing key
const truss = new TrussClient({
apiKey: process.env.TRUSS_API_KEY,
privateKey: process.env.ED25519_PRIVATE_KEY
});
// Wrap consequential agent actions
const result = await truss.recordAction({
mandateId: "man_8f92a1",
actionType: "DATA_EXTRACTION",
context: {
target: "patient_records_db",
purpose: "clinical_trial_matching"
},
execute: async () => {
return await agent.extractData();
}
});
console.log(result.receipt.merkleRoot);
Evidence Packages
Every action. Bound to a mandate. Verifiable.
Truss packages every agent action into a mandate-bound evidence record that holds in court.
Contract Execution
AutoGen agent triggers a Docfide revision under a pre-agreed mandate
Regulatory Disclosure
FCA-mandated data extraction by LangGraph workflow
Cross-border Payment
Agent-to-agent settlement via Restack with jurisdiction-routed mandates
HalluCase Response
Post-incident agent quarantine with chain-of-custody record
Truss Verify
Verify any Truss record. Free. No account required.
Designed for courts, regulators, and counterparties. Drag and drop an evidence package to verify its cryptographic integrity.
Upload Action Record
Drag and drop a Truss evidence JSON file, or click to browse.
Who needs this
Built for developers. Defensible in court.
General Counsel
“When your AI agent caused this outcome, what is your legal exposure? Truss produces the mandate record and chain of custody your defense needs.”
Chief Compliance Officer
“Demonstrate to the FCA, the SEC, or the NDPR that your agents operated within regulatory boundaries. Per framework. Per action. Per jurisdiction.”
Head of Legal Operations
“Every Glawly and Docfide agent is automatically covered. Every action carries a record. Every matter has a trail.”
Counterparty or Dispute Counsel
“Company A's agent instructed Company B's agent. Truss maintains the chain of custody across that boundary. Both sides. One record.”
Ecosystem
Work where your agents already live
Truss integrates into leading agent frameworks, legal AI tools, and agent protocols.
Open Source
The Truss Action Protocol is open
TAP is an open standard for agent action provenance. The protocol specification, SDKs, and verification tooling are MIT-licensed and available on npm, PyPI, and GitHub.
- @tensflare/tap — Core data-model schemas for mandates, actions, and delegations
- truss-sdk — Python SDK for LangGraph, AutoGen, CrewAI
- @tensflare/truss-core — Schema, seal, and verify primitives
- Truss Agent Packages — Pre-built integrations for LangGraph, AutoGen, CrewAI, Dify
// npm install @tensflare/tap
import { tap, seal } from "@tensflare/tap";
// Wrap an agent action with Truss
const result = await tap("agent.execute", {
mandate_id: "mandate-a3821",
input_hash: "sha256:...",
output_hash: "sha256:...",
within_mandate: true,
});
// Seal the evidence package
const pkg = await seal(result);
// -> Verifiable in any jurisdictionLatest
“The government shut down Fable 5 for everyone because it couldn't ask which agents were authorized.”
No mandate records. No precision. Just a blunt shutdown. This is the gap Truss closes.
Early Access
Ready to make your agents provably trustworthy?
Join the waitlist. We're onboarding design partners in regulated industries now.