prova-eliza-plugin

Verifiable AI for elizaOS agents. Drop this plugin into any elizaOS agent and every model call runs as confidential inference on Prova — hardware-attested inside a TEE, signed by the enclave, and verified on Solana. Each response comes with an on-chain proof your agent can show anyone: a Solscan link.

Agents are making financial decisions, managing treasuries, and executing trades — with no proof of what happened inside the model. This plugin fixes that. Every inference becomes an un-fakeable, on-chain receipt.

What it does

• Registers Prova as your agent's model provider — TEXT_SMALL, TEXT_LARGE, OBJECT_SMALL, OBJECT_LARGE all route through Prova. No code changes to your agent logic.
• Proves every inference on-chain — each response is hardware-attested (NVIDIA + Intel TDX), ECDSA-signed by the TEE, and anchored on Solana by an on-chain verifier program. A fake signature or unregistered key is rejected by Solana, not by us.
• Returns the Solscan link — a VERIFY_PROOF action and a PROVA_PROOF provider expose the receipt and Solscan URL for the most recent inference, so your agent can prove what it did on demand.

Install

npm install prova-eliza-plugin

@elizaos/core is a peer dependency (your agent already has it).

Configure

Get a Prova API key at prova.chat/dashboard, then set:

PROVA_API_KEY=prova_sk_your_key_here

That's the only required setting. Optional overrides (env var or character settings):

| Setting | Default | Description | | --- | --- | --- | | PROVA_API_KEY | — | Required. Your prova_sk_… key. | | PROVA_BASE_URL | https://api.prova.chat/v1 | OpenAI-compatible base URL. | | PROVA_SMALL_MODEL | deepseek/deepseek-v4-flash | Model for TEXT_SMALL / OBJECT_SMALL. | | PROVA_LARGE_MODEL | deepseek/deepseek-v4-pro | Model for TEXT_LARGE / OBJECT_LARGE. |

Use

Add the plugin to your character:

import { provaPlugin } from "prova-eliza-plugin";

export const character = {
  name: "MyAgent",
  plugins: [provaPlugin],
  settings: {
    secrets: { PROVA_API_KEY: process.env.PROVA_API_KEY },
  },
  // ...
};

That's it. Every runtime.useModel(...) call your agent makes is now confidential and proven on Solana. When a user asks your agent to "prove that" or "show the proof", the VERIFY_PROOF action responds with the on-chain Solscan link.

Reading the proof programmatically

import { provaProofStore } from "prova-eliza-plugin";

const proof = provaProofStore.latest();
// {
//   receiptId, attested, model, teeKey,
//   reqHash, resHash, solanaTx, solscanUrl, status
// }
console.log(proof?.solscanUrl); // https://solscan.io/tx/...

What gets proven on-chain

• Which model ran
• Which TEE hardware signed it
• Hash of the agent's input (req_hash)
• Hash of the model's output (res_hash)
• ECDSA signature from the enclave
• Timestamp, provider
• verifiedSignature / verifiedRegistry — both enforced by Solana before the proof is anchored

Verify any proof at prova.chat/verify, or browse the on-chain verifier program: D4fwqE74azXzC6euWAmDoH6Up1gZEh725odUuZHcCqEB.

How it works

1. Your agent calls a model. The plugin sends the request to Prova's OpenAI-compatible API (stream: false) over your prova_sk_… key.
2. Prova runs the inference inside a confidential TEE and returns the response plus x-prova-* proof headers (receipt id, attestation status, TEE key, request/response hashes).
3. The proof is recorded locally and the Solana anchor — which lands a few seconds later — is resolved in the background via GET /v1/receipts/{id}, attaching the transaction id and Solscan link.
4. The VERIFY_PROOF action / PROVA_PROOF provider surface that link whenever your agent (or its user) wants it.

Compatible with

elizaOS, and any agent framework that consumes elizaOS model providers. Prova's API is OpenAI-compatible, so the same key works from any OpenAI SDK if you're integrating outside eliza.

Links

• Website: prova.chat
• Docs: prova.chat/docs/agents
• Verify a proof: prova.chat/verify
• Get an API key: prova.chat/dashboard

License

MIT © Prova