Workflows

This page shows how the individual pieces fit together in real applications. Each workflow is a complete, copy-paste-ready scenario.

Workflow 1: Full provisioning pipeline

The one-time setup every integration goes through: sanity-check auth → create a Provider → verify it → create supporting resources → create the Agent → test it via chat.

import { PersonaClient } from '@personaai/sdk';

const persona = new PersonaClient({
  baseUrl: process.env.PERSONA_BASE_URL!,
  credential: process.env.PERSONA_CREDENTIAL!,
});

async function provision() {
  // 1. Sanity-check the credential.
  const who = await persona.whoami();
  console.log(`Provisioning for ${who.domain}`);

  // 2. Provider — the LLM endpoint the Agent will use.
  const provider = await persona.providers.create({
    label: 'OpenAI (prod)',
    baseURL: 'https://api.openai.com/v1',
    apiKey: process.env.OPENAI_API_KEY!,
    defaultModel: 'gpt-4o-mini',
  });

  // 3. Verify it works before wiring it anywhere.
  const test = await persona.providers.testConnection(provider.id);
  if (!test.success) throw new Error(`Provider rejected: ${test.message}`);

  // 4. A Skill the Agent can use.
  const skill = await persona.skills.create({
    name: 'Resume Reviewer',
    description: 'Reviews resumes against a rubric.',
    instructions: 'You are an expert resume reviewer. Use the attached rubric.',
    files: [{ path: 'rubric.md', content: '# Scoring rubric\n- Impact: 30%\n- Clarity: 30%...' }],
  });

  // 5. A Knowledge base with company FAQs.
  const kb = await persona.knowledge.create({
    name: 'Placement Office FAQ',
    providerId: provider.id,
  });

  // 6. The Agent, wired to all of it.
  const agent = await persona.agents.create({
    name: 'Career Launchpad',
    systemPrompt: 'You help students find internships. Answer from the FAQ when relevant.',
    providerId: provider.id,
    skills: [skill._id],
    knowledgeBases: [kb._id],
    category: 'productivity',
    visibility: 'unlisted',
  });

  console.log('Provisioned agent:', agent._id);
  return agent;
}

Workflow 2: Per-user chat app (the typical production shape)

The standard server pattern: a control-plane client for admin work + a runtime-plane client per incoming request. Your own auth confirms who the user is; then you scope the SDK client to them.

// persona.ts — shared client factory
import { PersonaClient } from '@personaai/sdk';

export function personaFor(externalUserId?: string) {
  return new PersonaClient({
    baseUrl: process.env.PERSONA_BASE_URL!,
    credential: process.env.PERSONA_CREDENTIAL!,
    externalUserId,
  });
}

// routes/chat.ts — Express example
import { personaFor } from '../persona.js';

app.post('/api/chat', async (req, res) => {
  try {
    const userId = await authenticate(req); // YOUR OWN auth
    const persona = personaFor(userId);

    // Resolve any pending human-in-the-loop interrupt from a previous turn.
    const pending = await loadPendingInterrupt(userId); // your app state
    const result = await persona.chat.sendMessage(req.body.agentId, {
      threadId: req.body.threadId, // resume the named conversation
      messages: pending ? [] : [{ role: 'user', content: req.body.message }],
      resume: pending,
    });

    if (result.interrupt) {
      // Surface an approval/clarification prompt to your frontend.
      res.json({ needsInput: true, kind: result.interrupt.kind, detail: result.interrupt.value });
    } else if (result.error) {
      res.status(502).json({ error: result.error });
    } else {
      res.json({ text: result.text });
    }
  } catch (err) {
    handlePersonaError(err, res); // see Errors page
  }
});

Workflow 3: RAG pipeline — ingest, search, attach

Build a searchable knowledge base and let an Agent use it.

import { readFile } from 'node:fs/promises';
import { PersonaClient } from '@personaai/sdk';

const persona = new PersonaClient({
  baseUrl: process.env.PERSONA_BASE_URL!,
  credential: process.env.PERSONA_CREDENTIAL!,
});

// 1. Create the knowledge base (providerId required — it supplies the embedding key).
const kb = await persona.knowledge.create({
  name: 'Product Handbook',
  providerId: 'prov_123',
  chunkSize: 800,
  chunkOverlap: 100,
  topK: 5,
});

// 2. Ingest a batch of documents.
const files = await Promise.all(
  ['handbook.md', 'faq.md', 'pricing.csv'].map(async (f) => ({
    filename: f,
    content: await readFile(`./docs/${f}`),
    contentType: f.endsWith('.csv') ? 'text/csv' : 'text/markdown',
  })),
);
const upload = await persona.knowledge.uploadDocuments(kb._id, files);
console.log(`${upload.documentCount} documents, ${upload.chunkCount} chunks`);

// 3. Sanity-check retrieval before attaching to an Agent.
const hits = await persona.knowledge.search(kb._id, 'how do refunds work?', { topK: 3 });
console.log(hits.map((h) => ({ source: h.source, score: h.score })));

// 4. Attach and go live.
await persona.agents.update('agent_123', { knowledgeBases: [kb._id] });

Workflow 4: MCP OAuth — user-mode connection flow

Connect an OAuth-protected MCP for a specific end user (a common CRM/calendar integration).

import { PersonaClient } from '@personaai/sdk';

const persona = new PersonaClient({
  baseUrl: process.env.PERSONA_BASE_URL!,
  credential: process.env.PERSONA_CREDENTIAL!,
});

// 1. Register the MCP (OAuth, user-mode). Note: create() probes OAuth discovery synchronously.
const mcp = await persona.mcps.create({
  name: 'Company CRM',
  transport: 'http',
  url: 'https://crm.example.com/mcp',
  authType: 'oauth',
  authMode: 'user',
  oauth: {
    clientId: process.env.CRM_CLIENT_ID!,
    clientSecret: process.env.CRM_CLIENT_SECRET!,
    scopes: ['contacts:read', 'deals:write'],
  },
});

// 2. Attach to the shared Agent (Project-level).
await persona.agents.update('agent_123', { mcps: [mcp._id] });

// 3. When a user needs to connect, get THEIR authorize URL and redirect them.
async function connectUser(externalUserId: string, returnToUrl: string) {
  const userClient = new PersonaClient({
    baseUrl: process.env.PERSONA_BASE_URL!,
    credential: process.env.PERSONA_CREDENTIAL!,
    externalUserId,
  });
  const { url } = await userClient.mcps.oauth.getUserAuthorizeUrl(mcp._id, returnToUrl);
  return url; // redirect the end user's browser here
}

// 4. Later, check connection status / disconnect.
async function status(externalUserId: string) {
  const userClient = new PersonaClient({ /* ... */ externalUserId });
  const { connected } = await userClient.mcps.oauth.getUserConnectionStatus(mcp._id);
  if (connected) await userClient.mcps.oauth.disconnectUserConnection(mcp._id);
}

Workflow 5: Seeding memory before first contact

Give an Agent useful context about a new user before their first message — indistinguishable from memory the Agent wrote itself.

import { PersonaClient } from '@personaai/sdk';

const userClient = new PersonaClient({
  baseUrl: process.env.PERSONA_BASE_URL!,
  credential: process.env.PERSONA_CREDENTIAL!,
  externalUserId: 'user_123',
});

await userClient.memory.writeFile({
  path: '/memories/user/preferences.md',
  content: [
    '- Prefers concise, bullet-point answers',
    '- Works in fintech',
    '- Looking for summer internships in NYC',
    '- Already has a strong resume; needs interview practice',
  ].join('\n'),
});

// Then the user's first chat call automatically sees this context.

Workflow 6: Named conversation history with files

Threads + Files + Chat working together — a user uploads a document, chats about it in a named thread, and can come back later.

import { readFile } from 'node:fs/promises';
import { PersonaClient } from '@personaai/sdk';

const userClient = new PersonaClient({
  baseUrl: process.env.PERSONA_BASE_URL!,
  credential: process.env.PERSONA_CREDENTIAL!,
  externalUserId: 'user_123',
});

// 1. Create a named thread so the conversation is listable.
const thread = await userClient.threads.create({ agentId: 'agent_123' });

// 2. Upload a document and associate it with the thread.
const resume = await readFile('./resume.pdf');
const file = await userClient.files.upload({
  filename: 'resume.pdf',
  content: resume,
  contentType: 'application/pdf',
  threadId: thread._id,
});

// 3. Chat within that thread, streaming to the client.
for await (const event of userClient.chat.stream('agent_123', {
  threadId: thread.threadId,
  messages: [{ role: 'user', content: 'Review my resume for the internship role.' }],
})) {
  if (event.type === 'TEXT_MESSAGE_CHUNK' && event.delta) process.stdout.write(event.delta);
}

// 4. Let the user rename it; surface it in a "history" screen later.
await userClient.threads.updateTitle(thread._id, 'Resume review');

// 5. History screen:
const { items } = await userClient.threads.list({ limit: 20 });
console.log(items.map((t) => ({ title: t.title, agent: t.agentId.name })));

Workflow 7: Stores as read-only reference material

Large reference docs mounted on an Agent, populated by you — the Agent reads them on demand.

import { readFile } from 'node:fs/promises';
import { PersonaClient } from '@personaai/sdk';

const persona = new PersonaClient({
  baseUrl: process.env.PERSONA_BASE_URL!,
  credential: process.env.PERSONA_CREDENTIAL!,
});

// 1. Create a domain-scoped, READONLY store (Agents can read, never write).
const store = await persona.stores.create({
  name: 'company-policies',
  scope: 'domain',
  accessMode: 'readonly',
});

// 2. Populate it through the API (writeFile is never gated by accessMode).
for (const [path, file] of [
  ['/policies/hiring.md', './policies/hiring.md'],
  ['/policies/expenses.md', './policies/expenses.md'],
]) {
  await persona.stores.writeFile(store._id, {
    path,
    content: await readFile(file, 'utf8'),
  });
}

// 3. Mount it on every Agent that should consult the policies.
await persona.agents.update('agent_123', { storeMounts: [store._id] });
// The Agent now sees the files at /stores/company-policies/policies/... via read_file.

Workflow 8: Cleanup with dependency checks

Safe teardown that respects the delete-blocking rules.

import { PersonaClient } from '@personaai/sdk';

const persona = new PersonaClient({
  baseUrl: process.env.PERSONA_BASE_URL!,
  credential: process.env.PERSONA_CREDENTIAL!,
});

// Providers, Skills, and MCPs reject delete() while an Agent references them.
// Knowledge bases do NOT block — getUsage() is informational there.
async function safeDelete(kind: 'provider' | 'skill' | 'mcp', id: string) {
  const usage =
    kind === 'provider'
      ? await persona.providers.getUsage(id)
      : kind === 'skill'
        ? await persona.skills.getUsage(id)
        : await persona.mcps.getUsage(id);

  if (usage.agentCount > 0) {
    console.log(`Still referenced by ${usage.agentCount} agents:`, usage.agents);
    return false;
  }
  if (kind === 'provider') await persona.providers.delete(id);
  else if (kind === 'skill') await persona.skills.delete(id);
  else await persona.mcps.delete(id);
  return true;
}

// Batch teardown with best-effort semantics.
async function bulkCleanup() {
  const { deleted, failed } = await persona.agents.bulkDelete(['a1', 'a2', 'a3']);
  console.log(`Deleted ${deleted.length}, failed ${failed.length}`);
  for (const f of failed) console.log(' -', f.id, f.reason);
}

Workflow 9: Architect-driven agent creation

Let the conversational Architect build an Agent (and its Skill) for you, with approval handling:

import { PersonaClient } from '@personaai/sdk';

const persona = new PersonaClient({
  baseUrl: process.env.PERSONA_BASE_URL!,
  credential: process.env.PERSONA_CREDENTIAL!,
});

const result = await persona.architect.sendMessage({
  messages: [
    {
      role: 'user',
      content:
        'Create an agent named "PDF Summarizer" that summarizes uploaded PDFs, uses ' +
        'provider "OpenAI (prod)", and is unlisted. Also create a skill for it called ' +
        '"Chunking" with instructions on splitting long PDFs.',
    },
  ],
});

if (result.interrupt) {
  const approved = await persona.architect.sendMessage({
    messages: [],
    resume: { decisions: [{ action: 'upsert_agent', decision: 'approve' }] },
  });
  console.log(approved.text);
} else {
  console.log(result.text);
}

Workflow 10: Error handling wrapper

A reusable error handler for Express-style apps, covering typed Persona errors and network failures:

import {
  PersonaApiError,
  PersonaAuthError,
  PersonaValidationError,
} from '@personaai/sdk';
import type { Response } from 'express';

export function handlePersonaError(err: unknown, res: Response) {
  if (err instanceof PersonaValidationError) {
    return res.status(400).json({ error: err.message, code: err.code });
  }
  if (err instanceof PersonaAuthError) {
    return res.status(401).json({ error: err.message });
  }
  if (err instanceof PersonaApiError) {
    if (err.statusCode === 404) return res.status(404).json({ error: 'Not found' });
    return res.status(err.statusCode).json({ error: err.message, code: err.code });
  }
  // Network failures / aborts / anything unexpected.
  console.error('Unhandled error calling Persona:', err);
  return res.status(502).json({ error: 'Upstream request failed' });
}

Node.js SDK Reference

React SDK Reference

UI Components Reference

Next.js Adapter

Express Adapter

NestJS Adapter

Python SDK Reference

Runtime Reference

Workflow 1: Full provisioning pipeline

Workflow 2: Per-user chat app (the typical production shape)

Workflow 3: RAG pipeline — ingest, search, attach

Workflow 4: MCP OAuth — user-mode connection flow

Workflow 5: Seeding memory before first contact

Workflow 6: Named conversation history with files

Workflow 7: Stores as read-only reference material

Workflow 8: Cleanup with dependency checks

Workflow 9: Architect-driven agent creation

Workflow 10: Error handling wrapper

​Workflow 1: Full provisioning pipeline

​Workflow 2: Per-user chat app (the typical production shape)

​Workflow 3: RAG pipeline — ingest, search, attach

​Workflow 4: MCP OAuth — user-mode connection flow

​Workflow 5: Seeding memory before first contact

​Workflow 6: Named conversation history with files

​Workflow 7: Stores as read-only reference material

​Workflow 8: Cleanup with dependency checks

​Workflow 9: Architect-driven agent creation

​Workflow 10: Error handling wrapper

Workflow 1: Full provisioning pipeline

Workflow 2: Per-user chat app (the typical production shape)

Workflow 3: RAG pipeline — ingest, search, attach

Workflow 4: MCP OAuth — user-mode connection flow

Workflow 5: Seeding memory before first contact

Workflow 6: Named conversation history with files

Workflow 7: Stores as read-only reference material

Workflow 8: Cleanup with dependency checks

Workflow 9: Architect-driven agent creation

Workflow 10: Error handling wrapper