neoly
attesting
Launch console

Private AI agents · NEAR AI Cloud · elizaOS

Agents that can prove nobody was looking.

Every model call runs in a hardware enclave, and is verified before your prompt leaves.

Run a private audit Watch it verify

01Proof

Attest first. Then talk.

  1. 01Fresh nonce32 random bytes, from your browser.
  2. 02Gateway quoteIntel TDX, nonce sealed inside.
  3. 03Model + 8 GPUsTDX and NVIDIA, checked to the root.
  4. 04Signed replyFail any check: nothing is sent.
verify attest --model z-ai/glm-5.3-flash live

02Platform

One plugin. Private by default.

elizaOS · model provider

// agent.ts
const runtime = new AgentRuntime({
  character,
  plugins: [sql, nearAiTee],
});

await runtime.useModel(TEXT_LARGE, { prompt });
// attested → inferred in TEE → signature checked

@neoly/plugin-near-ai-tee

0

checks before a token leaves

0×

H100s attested, every call

0 B

of your prompt on our disk

you ─────────▶ TEE
     neoly

Direct mode: your prompt never touches us

fail-closed

verified? ── yes ──▶ send prompt
    └────── no ───▶ throw · nothing sent

receipts

Every call returns an evidence bundle you can re-verify offline.

$ bun run verify bundle neoly-evidence.json

Agent mode relays through Neoly, which is not yet inside a TEE. Use direct mode to bypass it entirely.

03Verify

Don't trust us. Check.

Open the console Read the source