AI Support Agent for Bicrypto - AI Helpdesk, Ticket Automation, Live Chat & Knowledge Base

AI Support Agent for Bicrypto

An assistant that answers your support tickets and live chat from your own documentation, cites the passage it used, and hands the conversation to a person rather than guessing. It ships switched off and in copilot mode — it drafts, your team presses Send — and is promoted one rung at a time, on evidence from your own tickets.

Please read first: this is an add-on

AI Support Agent is an extension for Bicrypto, a separate product sold separately on CodeCanyon. It does not run on its own. It has no standalone installer, no interface of its own outside the Bicrypto admin panel, and no database of its own. A working, licensed Bicrypto 6.7.2 or newer installation is a hard prerequisite. If you do not already run Bicrypto, buy and install that first — without it there is nothing in this package to run.

Please read: AI usage costs money, and it is not included in this price

The price of this item buys the software. It does not buy AI capacity. Every answer the assistant writes is a paid API call to whichever AI provider you connect it to, billed to you by that provider at their rates, and that bill grows with use for as long as the assistant is switched on. The same applies to any other external service you point it at. There is no configuration in which answering a customer is free.

You choose the provider, and there are three ways to connect one:

• The managed connection — metered, pay per answer, nothing to subscribe to
• Your own vendor key — an Anthropic API key, billed to your own account with that vendor
• Any OpenAI-compatible endpoint — a self-hosted model, or any third-party router or service that speaks the chat-completions shape

On the shipped defaults, one answered conversation turn has measured between $0.02 and $0.30 depending on the provider, the model and how long the conversation runs. Because that is a range and not a number, the add-on makes you set a ceiling and enforces it: $5 a day and $100 a month by default, both editable. When a ceiling is reached the assistant stops answering and hands over to a person instead. It does not degrade quietly and it cannot overspend the limit you set. The cost of every single answer is recorded, so the spend screen shows you where the money went rather than a total.

Two things reduce the bill over time and neither costs anything to switch on. The nightly harvest turns replies your agents wrote by hand into draft articles, so questions you answer twice become questions that answer themselves. And a question the retrieval layer cannot ground is refused before any model call is made — a question the assistant cannot answer is also a question you are not billed for.

What it does

The assistant reads your support conversations and answers them from your knowledge base and the product documentation. Every answer carries the passage it came from, so a customer can check it and your team can audit it afterwards. The model is never asked what it knows about your platform — it is handed passages from your documentation and told to answer from those.

• Support tickets — drafts or sends a grounded reply when a customer replies to their own ticket
• Live chat — answers in the chat widget, with a live agent one click away
• Deflection — offers an answer while the customer is still typing their subject line, so the ticket is never filed
• Escalation and handover — when it stops, the human who picks it up receives a brief: what the customer wants, what was established, and what is blocking it
• Actions — it can offer a customer a next step, but it never acts. It writes a proposal; the customer’s own click is the only thing that runs anything, under their own session and every check that guards the rest of the platform.
• Your own assistant — an admin-side assistant that answers operators from the operator documentation and can walk them to the screen that holds the answer

What it refuses to do

Most of the engineering in this add-on is in the answers it declines to give. An assistant that answers everything is an assistant that invents your refund policy.

• It will not answer below a confidence floor. If retrieval cannot find a good enough source it hands over to a person and records the question as a gap for you to close.
• It will not state your own numbers from somebody else’s documentation. Fees, limits, review times, supported countries and refund policy can only be answered from articles you wrote. That gate applies in every language, not only English.
• It will not name the vendor. Your install is white-labelled and the assistant’s answers are part of your brand. It writes a page’s name and its in-platform path, never an external link.
• It will not act on an account unless you let it. Account tools are read-only by default, and the actions it may take are an allowlist you fill in yourself — empty at install.

Promoted one rung at a time

It ships off. Nothing happens because you installed it.

• Copilot — the assistant drafts, a human reads it and presses Send. This is the shipped default.
• Auto tickets — it answers tickets on its own; live chat still goes to a person.
• Auto everything — it answers tickets and live chat unsupervised.

The console will not let you skip ahead casually: promotion is measured against how much your agents actually changed its drafts, on your own tickets.

The console

Nine admin screens, each with its own permission so you can hand out exactly as much as a support lead needs:

• Overview — spend today and this month, what it did, how much was handled without a person
• Live Inbox — conversations the assistant is handling right now, with take-over, a case brief, the evidence it used and a suggested reply
• Provider — which AI service answers, whether it is connected, and what it charges
• Agents — personas: name, avatar, disclosure line, model, effort and availability
• Sources — what it is allowed to answer from, and the index built out of it
• Gaps — questions your documentation does not answer yet, clustered, with a one-click way to answer them
• Actions — the allowlist of things it may offer to do, and the multi-step processes it may walk a customer through
• Settings — autonomy, budgets, retrieval, channels, office hours, data retention
• Handbook — the operator-side assistant and the procedures it answers from

Technical requirements

Everything except the last two lines is what Bicrypto core already requires. The add-on itself introduces no new npm dependency — the AI provider adapters are plain fetch calls rather than vendor SDKs — so installing it does not change the dependency tree of your installation.

Platform

• Bicrypto core 6.7.2 or newer, installed, licensed and running. No other add-on is required.
• Linux. Ubuntu 24.04 LTS or newer, or another distribution shipping glibc 2.38+ (Debian 13+, RHEL / Rocky / Alma 10+). Ubuntu 22.04 will not run the platform. x86-64 or ARM64, glibc — not Alpine or musl.
• Not shared PHP hosting. A running install is three long-lived Node.js processes behind a reverse proxy.

Runtime

• Node.js 22, 24 or 26. No other major version boots — the platform’s HTTP server is a prebuilt native binary and ships those three ABIs only.
• pnpm — the platform pins pnpm@10.28.0. npm and yarn cannot resolve the workspace.
• 4 GB RAM minimum, 8 GB to build the frontend reliably. 2+ CPU cores. Shell access, and the ability to restart the backend.

Database and services

• MySQL 8.0+ or MariaDB 10.6+, reachable, with the mysql client available on the application server. The add-on adds 17 tables, all prefixed ai_support_, created by the platform’s own update command. No existing schema is modified.
• Redis, reachable and answering PING — a hard dependency of Bicrypto, not a cache.
• No ScyllaDB, no search server, no message broker and no vector database are required.

API dependencies

• Outbound HTTPS from your server to one AI provider, which you choose and configure: the managed connection, the Anthropic API, or any OpenAI-compatible endpoint including a self-hosted one. This is the only external service the add-on calls, and its usage is billed to you by that provider — see the cost section above.
• An AI provider account and credential for whichever of those you pick. Credentials are set in the server’s environment file and never on a screen: the platform’s public settings endpoint would otherwise expose them.
• Optionally, outbound HTTPS to the documentation host, used when the knowledge index is rebuilt. This can be disabled or pointed at your own mirror.

What is in the download

• Full source code. The backend as TypeScript (119 files) and the nine admin screens as TypeScript and React (42 files). No compiled-only code and no obfuscation.
• Documentation — 33 pages of offline HTML covering installation, every console screen, every setting, the API reference, costs, permissions, scheduled jobs, data retention and troubleshooting. Open documentation/index.html in any browser.
• A commented .env.example listing every environment variable the add-on reads, with what each one is for.
• A project manifest, a changelog, a licence file, and a small script that compiles the shipped source into a production installation.

Support

• Documentation: mashdiv.com/docs/ai-support-agent
• Support: mashdiv.com/support