§ 00 / products / apichatbot

APIchatbot. An AI assistant that answers only from your own content.

Upload your price lists, FAQs, manuals and policies; get a chatbot that cites where it read the answer and says "I don't know" when it isn't there. Multi-tenant, bilingual, with a client dashboard for usage, limits and cost. Website widget, internal assistant or API-connected agent — same build, different corpus.

builds from
CA$1,500
delivery
weeks, not months
languages
French · English
hosting
CA$45 /month
status
live · in production
§ 01 / three ways to run it

One build, three deployments.

01/widget

Website widget

A branded chat on your site that answers customers from your own content — hours, prices, policies, availability — around the clock, in their language.

02/internal

Internal assistant

A private desk for your team over manuals, procedures and price lists. Same library as the public widget, different key, different rules, different document access.

03/agent

API-connected agent

Wired into your calendar, CRM, ERP or billing: drafts the quote, looks up the order, books the slot. Every action scoped, logged and within limits you set.

§ 02 / what it does

Answers you can defend.

A documented "I don't know" is cheaper than a fluent error. Everything below exists so an organisation can run that for many teams, customers and applications without mixing secrets or losing control of the bill.

01

Corpus-only answers

Retrieval over your documents, not the open web. Every answer cites the document (and page) it came from.

02

Refuses instead of guessing

When the library does not cover the question it says so — in the language the question was asked — and points to a human.

03

French and English

Follows the question's language or is pinned per channel; cross-lingual retrieval when the corpus mixes both.

04

Real conversations

Follow-ups keep their context, next-question suggestions, typo-tolerant query rewriting, your product names and acronyms learned from your own library.

05

Hard PDFs survive

Tables, dosage schedules and decision charts are converted with a vision model so a table stays a table.

06

A voice per channel

Tone, house rules, disclaimers, answer length and even the model are set per API key — the public widget and the internal desk need not behave alike.

07

Isolated tenants

Separate library, users, keys and invoices per organisation on one installation; keys hashed and shown once; per-key document deny-lists.

08

Provisioned by AI

The platform exposes an MCP server: client, corpus and key can be set up from an assistant like Claude — this very demo was.

§ 03 / client dashboard

It runs as a business, not a science project.

Each client gets a scoped portal — their organisation only — with the tabs you enable per user. Everything they need to run the assistant without opening a ticket with you.

reports
Tokens consumed and amount billed per key and per month, refreshable on demand. Clients see what they were billed — never your cost or margin. The operator view puts cost and revenue side by side.
configuration
Self-service per key: answer language, detail level, citations, refusal, follow-ups, query rewriting, multilingual retrieval, retrieval depth, free-text instructions and extra context. Models, rates and caps stay operator-only.
limits
Requests per minute and a monthly token cap (429 with quota headers), expiry with a reminder e-mail seven days before, instant revoke, regenerate-in-place, and keys narrowed to a subset of documents.
telemetry
Every call logged with model, tokens, cost and revenue; every provider request logged for latency and errors; sensitive actions audited with who and when — without storing what users asked.
cost control
Caps that protect the invoice, the model chosen per key (start economical, upgrade what earns it), sell rates auto-priced from model cost plus your markup or set explicitly, repriceable from any date, and documents assessed — size, chunks, estimated cost — before you index them.
§ try it

Explore the client dashboard — read-only demo

See it for yourself — the demo client's own dashboard, read-only: chat, documents, usage and API docs. No sign-up.

Open the client dashboard user demo · password apichatbot-demo
§ 04 / integrations

It drops into what you already run.

01/api

Public chat API

POST with a Bearer key; streamed tokens or a single JSON answer with citations and follow-ups; 429 with rate-limit and quota headers. Drops into a website, intranet or mobile app.

  • REST
  • Bearer
  • NDJSON
  • 429
02/wp

WordPress plugin

The apigoat-chatbot plugin renders the bubble or an inline panel through a server-side proxy — the key never reaches the browser. Shortcode, floating mode, bilingual, follow-up chips.

  • [apigoat_chatbot]
  • proxy
  • FR/EN
03/mcp

MCP server

Provision a client, upload and embed its corpus, mint a key — from Claude or any MCP client, behind OAuth 2.1. The same server your team can use to operate the platform.

  • MCP
  • OAuth 2.1
  • Claude
04/host

Hosting

Runs on APIgoat's managed application hosting or your own infrastructure. PHP 8, MariaDB with native vector search, OpenAI embeddings, OpenAI or Anthropic chat models per key.

  • PHP 8
  • MariaDB VECTOR
  • OpenAI
  • Anthropic
§ 06 / live demo

Ask it about us_

This is APIchatbot running on our own content — services, rates, products, how we work. Same build we ship to clients; the only difference is the corpus.

Answers come only from our published content; anything else it will decline. Conversations are not stored on this site.

§ 07 / pricing

Transparent, like everything else here.

01/build

Packaged build — from CA$1,500

Scoping call, your content assessed and embedded, tone and rules set, widget or integration installed, your reference questions replayed to prove it answers and refuses correctly. Typically weeks, not months.

02/run

Managed hosting — CA$45 /month

SSL, weekly off-site backup, runtime updates, log retention and uptime monitoring on APIgoat's application hosting. Model usage is metered per key and billed at the rates set on your dashboard.

03/grow

Agents & integrations — CA$85 /hour

Connecting the assistant to your CRM, calendar or billing, custom MCP tools, or a second channel on the same corpus — at the standard hourly rate or a prepaid hour bank. Larger scopes quoted fixed.

Prices in Canadian dollars (CAD). Taxes extra where applicable. Final quote after a short scoping call.

Your documents already have the answers.

Send a brief with the content you have in mind — a price list, a FAQ, a manual — and get a quote and timeline within one business day.