@prudra/vault

Prudra vault SDK — Async Agent Workspaces. create vaults, upload files and documents, emit events, seal, persist, and subscribe over SSE.

Installation

npm install @prudra/vault

Prerequisites

initialise() from @prudra/core is required for Vault.create() and Vault.resume(). Vault.open() does not require initialise() — it uses a short-lived access token only.

1 — Creating and writing (server-side)

import { initialise } from '@prudra/core';
import { Vault } from '@prudra/vault';

initialise({ apiKey: process.env.PRUDRA_API_KEY! });

const vault = await Vault.create({ description: 'Analysis job', ttl: '24h' });

// Return 202 to the client immediately, then run background work:
await vault.addDocument({ score: 0.95 }, 'Result');
await vault.emit('step.complete', { step: 1 });
await vault.seal('Finished');

2 — Resuming a background job

const vault = await Vault.resume('vlt_abc123');
await vault.getManifest();

3 — Reading (calling agent)

import { Vault } from '@prudra/vault';

const vault = await Vault.open('vlt_abc', { accessToken: token });

vault.on('vault.sealed', (data) => {
  console.log(data);
});

for await (const ev of vault.stream()) {
  if (ev.type === 'vault.sealed') break;
}

const manifest = await vault.getManifest();

4 — Persisting and quota

import { PROBLEM_TYPES, PrudraConflictError } from '@prudra/core';
import { Vault } from '@prudra/vault';

try {
  await vault.persist();
} catch (e) {
  if (e instanceof PrudraConflictError && e.type === PROBLEM_TYPES.VAULT_QUOTA_EXCEEDED) {
    // handle at-plan limit
  }
}

vaultMiddleware() lives in @prudra/express (Step 4). For background jobs use Vault.create() / Vault.resume() directly.

File and document URLs returned by the API are always under https://assets.prudra.dev/....

Further documentation: docs.prudra.dev/docs/vault