Help

FAQ

What TrustFix is, prove-it vs enforce-it, agentless ingestion, how fixes are proven safe, whether the proofs are real, evidence, and the open-source CLI.

What is TrustFix?

The verifiable authority layer for non-human and AI-agent identity. Where most tools detect agents or broker and rotate their credentials, TrustFix issues, signs, and proves what an agent is allowed to do — as Ed25519-signed, hash-chained, Merkle-anchored entries in a Trust Ledger anyone can verify independently. See what TrustFix is.

What's the difference between "prove it" and "enforce it"?

Today, TrustFix issues, verifies, revokes, and proves authority — that's the control plane, and it ships now. Inline-blocking a live tool call that exceeds scope is the optional Gate proxy runtime tier, which we label as roadmap rather than imply it's always on. What ships is the governance and proof; per-request interception is a separate deployment. That discipline is the point: we'd rather attest provably than assert.

Does TrustFix work with my AI stack?

Yes — by construction, not by a connector per vendor. TrustFix governs identities, grants, MCP tools, and delegation hops, which every agent runtime has. Whatever you run — bespoke LangChain/AutoGen, any frontier model, any MCP server, on AWS/GCP/Azure — flows through the same capture → prove → attest → evidence pipeline with no provider branching. We don't need to "support" your AI platform; we govern the authority it exercises.

Does TrustFix have write access to my cloud account?

No. TrustFix reads through least-privilege, read-only connectors and never writes to your infrastructure during connection or scanning. For AWS we recommend a curated least-privilege template that grants only the specific read actions needed and never reads secret, parameter, or object values (the broader managed ReadOnlyAccess policy also works as a fallback). All fix recommendations are delivered as pull requests that require human review — TrustFix never auto-applies changes.

Does TrustFix store my cloud credentials?

No. AWS access is via temporary STS credentials from assuming your cross-account role, and GCP and Azure use Workload Identity Federation — short-lived federation over stored secrets. These credentials expire automatically and are never written to disk. Any long-lived token that must be held (OAuth, PATs) is KMS-envelope-encrypted and bound to your org. See the security model.

How are fixes proven safe?

When you generate a fix, the Policy Intelligence Engine validates it before opening a PR — ending in a formal proof that the new policy only narrows access. If the engine can't prove it, you get the reason and the score instead of a PR: the product will not ship a fix it can't verify. You can see this for any finding in the Remediation Dry-Run, which runs the same engine with no side effects on preview.

Are the proofs actually real?

Yes — you can verify them yourself. Agent passports and Trust Receipts have public pages (/agent-passport/[id], /trust-receipt/[id]) that re-verify the Ed25519 signature and Merkle inclusion proof with no access to your tenant; a tampered entry shows does not verify. Evidence items each carry a real ledger entry id checkable via /api/ledger/[id]/proof, and delegation verdicts are derived live by running the engine at read time, not read from a stored value. See Trust Ledger & proofs.

What's the honest limit on what you attest?

We attest mediated delegation hops; the human→agent root and unmediated hops show as observed, never proven. We don't sign agent outcomes, spend, or intent — those need the runtime Gate, which is roadmap. Per-segment delegation-attestation signing is currently gated, so those segments read "attestation pending (signing gated)" rather than show a fabricated ledger id. And evidence is Ed25519- and Merkle-anchored but not yet independently third-party-timestamped. We say all of this in-product rather than paint an unproven thing green.

What compliance evidence can I get?

A control-mapped, signed evidence pack that maps real artifacts onto controls across SOC 2, ISO 27001, NIST 800-53, OWASP ASI, and the AI-governance frameworks (ISO 42001, NIST AI RMF, EU AI Act Annex IV). Export a signed NIST OSCAL Assessment Results document, JSON, or CSV. Controls with no backing artifact are shown as EMPTY, never fabricated, and become open POA&M items in the OSCAL pack.

How does the AI fix generation work?

When you request a fix, TrustFix generates a least-privilege Terraform patch (template-first, with an AI fallback) targeting only the affected part of the policy — it does not rewrite unrelated parts of your role. Every generated fix is then validated by the Policy Intelligence Engine before it is offered, and the engine refuses rather than ship a fix it cannot prove safe.

Is there a free version?

Yes — the open-source oidc-audit CLI scans your AWS IAM roles for GitHub Actions OIDC misconfigurations from your terminal or CI, with no account required. It's a focused slice of one detector family; the platform is the full verifiable authority layer.