Throndar

FAQ

Questions, answered

Everything about the governed council, security, credits, and the studio.

What is Throndar?

Throndar is a governed AI portal: it sends your prompt to the seats of a thirteen-model council, fuses the drafts of the seats that answered into one synthesized answer, and, in the signed-in console, signs the answer with post-quantum cryptography. You also get a premium media engine suite in the same console.

Why a council of models instead of one?

A single model gives you one guess. Throndar sends your prompt to the seats of the council and has one seat cross-examine and fuse the drafts that arrive. A council answer is normally synthesised from several seats of the thirteen-model roster; when only one seat answered, or the synthesis failed, the signed receipt records it (council-grounded: false). That record — the answer's governance flag, council_synthesised, synthesis_fallback or single_model — rides with the answer and is bound into its signature; no separate check grades the answer.

Learn more: Why one AI model is a single point of failure →

Is it really free to start?

Yes. The Recon tier is $0 with no card: 200 starter credits plus 5 free council answers every day. Upgrade only when you need more throughput or the full apex layer.

What does 'post-quantum signed' mean?

A council answer from the signed-in console is signed with a NIST-standardized post-quantum signature — ML-DSA-87 (FIPS 204, final) — into a receipt you can verify yourself: that's the part you don't have to take on trust. The model-plane bridge also co-signs the answers it produces with Falcon-1024 (Round 3; not FN-DSA), and a saved answer keeps that co-signature only if it verified when it was saved. A console answer whose model plane did not state how it was made is delivered without a receipt, and the free demo and the non-streaming API sign none. Transport, for the browser and for the internal model-plane bridge hop, is whatever TLS the hosting platform negotiates; Throndar runs no ML-KEM key exchange of its own. Measured 2026-11-14 and 2026-11-15, the hosting platform's TLS endpoint for throndar.ai and www.throndar.ai negotiates the hybrid post-quantum key-agreement group X25519MLKEM768 with clients that offer it; offered alone, the classical groups X25519, P-256, P-384 and P-521 each also completed a handshake; ML-KEM-1024 is not offered. The same endpoint accepts TLS 1.3 and TLS 1.2, refuses TLS 1.0 and TLS 1.1, prefers the cipher suite TLS_AES_128_GCM_SHA256 and presents an RSA certificate. The signatures are the post-quantum assurance that travels with a signed answer.

Learn more: What “verified AI” actually means →

Can I verify a Throndar answer myself?

Yes. A signed answer carries an ML-DSA-87 receipt, and you can confirm in your own browser that Throndar produced it, unaltered, at throndar.ai/verify/proof. You can also share a signed answer as a self-proving link, or pin a single quote that anyone can verify against the signed original. Don't trust us — check it yourself.

Learn more: How to verify an AI answer →

What is the public transparency ledger?

Signed answers are appended to a public, append-only log — the same RFC 6962 Merkle structure (Certificate Transparency) that secures the web's TLS certificates. We publish only hashes, never your content. The append is best-effort: an answer whose append failed keeps its receipt but is not in the log. Each checkpoint root is signed and the witness history is checked, so a rewrite of past history is detectable rather than silent, and you can prove an appended answer is in it — verified in your browser at throndar.ai/transparency/ledger.

Which cryptographic algorithms does Throndar use?

All of them are published openly. Our Cryptographic Bill of Materials (CBOM) at throndar.ai/pqc lists every primitive — ML-DSA-87 (NIST FIPS 204), Falcon-1024 (Round 3; not FN-DSA), SLH-DSA-SHA2-256f (FIPS 205: on the model-plane bridge's answers, and on the web layer's receipts when that leg is enabled), SHA-256, and the TLS the hosting platform negotiates — mapped to its standard and role, following the TNO / AIVD / CWI PQC Migration Handbook. Throndar runs no ML-KEM key exchange of its own. It is designed and assessed against these public standards; this is not a certification, and we make no 'quantum-proof' claim.

Learn more: The NIST post-quantum standards, explained →

How does billing work?

Usage is metered in credits. A full council answer is 81 credits; media engines are priced per run. If an action fails, its credits are automatically refunded — you only pay for delivered work. Monthly grant credits roll over for one month; top-up credits never expire.

What payment methods do you accept?

Card (via Stripe's hosted checkout — we never see your card number), crypto (BTC, ETH, SOL, ALGO), and Zelle where enabled (it is switched on per deployment). Switch or cancel anytime: a card subscription is cancelled from your account page, and crypto and Zelle payments don't renew.

What can I create in the Studio?

Image, video, voice and audio from one governed console. FLUX image generation is live now; additional engines are coming online as they're provisioned.

Do you store my keys or data?

Your secrets are never entered into chat, and connector tokens are encrypted server-side and never exposed to the browser; sessions are scoped and revocable. The web layer's receipt keys are derived in-process with HKDF-SHA256 from a server-side environment secret, and the model-plane bridge loads its own signing keys from its environment; signing keys are never sent to the browser. There is no HSM or KMS: only the public keys are published, and you verify answers against them.

Still curious? Start free and ask the council yourself.