Open source · Local-first · MIT

AI customer service that runson your own machine.

Orqiva turns website and Telegram conversations into reliable answers with visual workflows, your own model key, and a human handoff when it matters. No cloud account. No lock-in.

Start in one command

orqiva — zsh
❯npm install -g orqiva

Bundled FrankenPHP · SQLite by default · Node.js 22+

Works with Gemini, OpenAI, and OpenRouter.

7

Workflow nodes

3

AI providers

2

Live channels

1

Command to start

Channels

Meet customers where the conversation already happens.

Start with the website widget. Add Telegram for mobile. Both land in the same inbox, with the same audit trail.

Website widget

Embed

Live

Drop one script into your site, pick the entry workflow, and match the widget to your brand.

Shared inbox · Human takeover

Telegram bot

Long polling

Live

Paste a BotFather token. Orqiva polls for updates, so you never expose a public webhook.

No webhook · No tunnel

WhatsApp

Coming soon

Coming soon

A familiar channel for customers who live in WhatsApp. Planned for the same inbox and workflow engine.

On the roadmap

Human takeover works the same on every channel.

Providers

Your keys. Your models. Your cost.

Connect Gemini, OpenAI, or OpenRouter per user. Keys are encrypted at rest; choose the model inside each AI Agent node.

Bring your own key AES-256 at rest Per-node model

No provider markup. Usage is tracked per step so you can see where tokens go.

Bring your own key

Google Gemini

Fast multimodal models for chat, reasoning, and structured replies.

Gemini 2.x
Bring your own key

OpenAI

Reliable general-purpose models with broad tool and JSON support.

GPT models
Bring your own key

OpenRouter

One key for a wide catalogue of hosted models and fallbacks.

Multi-model routing

Why Orqiva

Built for support work, not demo screenshots.

The pieces you need to ship a real assistant — and the controls to keep it accountable.

7 node types

Visual workflows

Compose triggers, AI agents, conditions, replies, jumps, and handoffs on a canvas. The server validates the graph before every run.

3 providers

Bring your own model

Use Gemini, OpenAI, or OpenRouter with keys encrypted per user. Pick the model per AI Agent node.

Human handoff

Shared inbox

Take over a session, reply as a human, then hand it back. Full contact history stays in one place.

Per-step logs

Auditable by design

Every step records input, output, variables, tokens, duration, and errors. Failed runs resume from the last checkpoint.

Zero webhooks

Telegram without webhooks

Connect a bot and receive updates through long polling. No public URL, tunnel, or certificate.

Bundled runtime

One-command runtime

The CLI ships FrankenPHP, migrations, a queue worker, and the Telegram poller. Install, run, open the browser.

Workflow editor

See the whole conversation before it happens.

Nodes expose only the settings that matter. Variables flow forward automatically, and validation catches missing triggers, unreachable nodes, and invalid branches before a customer ever sees them.

  • Stable node IDs with editable display names
  • Server-side graph validation and cycle-safe execution
  • Token usage and per-step audit logs
  • Draft, test, and publish without a redeploy
TriggerAI AgentConditionHuman handoff
orqiva.local/workflows/qualification
Orqiva workflow editor showing an AI qualification flow
A qualification flow with an AI Agent, a Condition, and a human handoff.

How it works

From install to first reply in three steps.

Useful on a laptop before it becomes a production deployment.

  1. 01

    Install and open

    Install the CLI, start the local service, and create your admin account in the browser. SQLite and the runtime are prepared for you.

    ❯npm install -g orqiva
  2. 02

    Compose a workflow

    Add a Trigger, an AI Agent, and a Send Message. Branch with a Condition when needed. Validate on save.

  3. 03

    Connect a channel

    Embed the website widget or connect a Telegram bot. Conversations land in the same inbox, with the same audit trail.

Get started

Run Orqiva locally in under a minute.

The CLI is the fastest path. Use the source setup when you want to develop or deploy the Laravel API and React app yourself.

npmRecommended
orqiva — install
❯npm install -g orqiva
❯orqiva

Orqiva CLI v0.2.0

✓ Local service ready · http://localhost:8000

✓ SQLite initialized · queue worker online

❯

From sourceFor development
source — zsh
❯cd backend && composer install
❯cp .env.example .env && php artisan key:generate
❯php artisan migrate && php artisan serve
❯cd frontend && npm install && npm run dev

Requirements for source: PHP 8.3+, Composer, Node.js 22+, and npm.

Read the setup guide
Ready when you are

Your customers are already asking.

Install Orqiva, connect a model key, and publish your first workflow today. The data never leaves your machine.

MIT licensed · No account required