Throndar fuses a 12-model AI council into one answer signed with ML-DSA-87. Free tier, paid plans from $49/mo. See features, pricing and verdict.
Researched and fact-checked by Bartu Cavusoglu · Updated October 26, 2026
Throndar is a governed AI portal. It sends your prompt to the seats of a twelve-model council, fuses the drafts from the seats that answered into one synthesized answer, and, in the signed-in console, signs that answer with a post-quantum signature.
The central claim is provenance rather than raw model quality. A signed answer comes with a receipt that records how it was made, and you can check that receipt yourself in a browser. Throndar says this is the part you do not have to take on trust.
Alongside the council, the same console includes a metered media studio for image, voice and video generation, and connectors that let the council read context from tools such as mail, calendars, files and code.
A prompt is routed to every configured seat of the council, the same prompt to each, or to the first three on a fast Live-Voice turn. The seats answer in parallel. Any seat that has not answered within the draft window, at most 30 seconds, is dropped.
One seat then cross-examines and fuses the drafts into a single answer. The answer's governance flag records how it was made: council_synthesised, synthesis_fallback or single_model. The receipt signs that flag together with the answer. Throndar states that no separate check grades or approves an answer before it ships.
Signed-in council answers receive an ML-DSA-87 (FIPS 204) receipt. The model-plane bridge adds Falcon-1024 and SLH-DSA-SHA2-256f co-signatures. An answer whose model plane did not state how it was made is delivered without a receipt.
The security layer is the most developed part of the product. Receipts can be verified at /verify/proof against embedded public keys, answers can be shared as self-proving links, and a single quote can be pinned for verification against the signed original. Signed answers are also appended, best-effort, to a public append-only ledger that publishes hashes rather than content.
The media studio offers FLUX image generation by default. ElevenLabs voice, Topaz upscale and Runway, Kling and HeyGen video are built and switched on per deployment. All of it is metered in credits.
Connectors only read. Throndar does not send, post or edit as you, though GitHub and Google Ads offer no read-only grant, and their cards say so. Connector tokens are encrypted server-side and never exposed to the browser.
Throndar is a governed AI portal that sends a prompt to a twelve-model council and fuses the drafts into one answer. In the signed-in console the answer is signed with an ML-DSA-87 post-quantum receipt. It also includes a metered media studio and read-only connectors.
Recon is free. Paid plans are Operator at $49/mo, Studio at $149/mo, Throne at $399/mo and Ultra at $999/mo, with yearly billing saving about 17%. Sovereign is custom, from $2,500/mo on an annual contract.
Yes. The Recon tier costs $0 with no card required. It includes 200 one-time starter credits and up to 5 free council answers a day with a verified email.
A signed-in council answer receives an ML-DSA-87 (FIPS 204) signature in a receipt you can verify yourself. The model-plane bridge also adds Falcon-1024 and SLH-DSA co-signatures. Throndar runs no key exchange of its own, so transport uses the TLS the hosting platform negotiates.
Yes. You can check a signed answer in your own browser at throndar.ai/verify/proof against the public keys. A sample proof is published at /demo-proof.json.
Connectors cover email, calendar, drive and storage, and code and repos, including GitHub and Google Ads. They only read, although GitHub and Google Ads offer no read-only grant. Availability depends on each provider registering Throndar's app.
Usage is metered in credits. A full council answer is 81 credits and media engines are priced per run. Failed actions are refunded, monthly grant credits roll over for one month, and top-up credits never expire.
No. Throndar states that it is designed and assessed against public standards, that this is not a certification, and that it makes no quantum-proof claim. Its algorithm choice is only partly in common with CNSA 2.0.