ADME is an AI-powered platform that turns every small and medium business into a 24/7 operation. We deploy customizable AI agents across messengers (Instagram, Telegram, WhatsApp), web widgets, and voice — so your clients get instant answers, your bookings fill themselves, and your content writes itself in your own voice.
Think of it as hiring a receptionist, a marketing assistant, and an operations manager — except they never sleep, never forget a client’s name, and cost less than a single part-time hire.
Who It’s For
Service businesses drowning in routine:
- Beauty salons, barbershops, nail studios
- Dental clinics, medical practices, physiotherapy
- Psychologists, coaches, consultants
- Marketing agencies and creative studios
- Fitness studios, yoga schools
Two types of users, one platform:
- Non-technical owners — you get a setup wizard, a simple dashboard, and an AI that just works. Plug in Instagram, upload your price list, done.
- Technical teams and power users — you get MCP integrations, custom agent pipelines, webhooks, API access, and hardware control via MCP servers. Build anything.
What You Actually Get
Isolda — Your AI Receptionist
A smart conversational agent that lives inside your client-facing channels. Speaks the client’s language (literally — multilingual out of the box), handles bookings, answers FAQs pulled from your own knowledge base, sends reminders, and hands off to you only when it actually matters. Switchable persona: Isolda or Tristan, your call.
Companion Mode — Your Personal Business Copilot
A full multi-agent system for the business owner. You talk to it like a colleague. It talks to your email, your spreadsheets, your calendar, your Instagram analytics, your Telegram. Example requests that actually work:
- “Find all Nova Poshta invoices from March, extract the totals, put them in a spreadsheet, send Olena the link in Telegram.”
- “Pick my three best-performing photos, caption them in my style, schedule them for next week.”
- “Summarize what happened in my client chats overnight and flag anything urgent.”
Under the hood: chained sub-agents, each with a clear job, orchestrated automatically.
Content Studio — Multi-Agent Content Generation
A full content creation pipeline, not a “generate caption” button. You give it a topic or let it learn from your past posts, and it produces coordinated content plans across Instagram, Facebook, Telegram, Threads, and more.
How it works:
- StyleAnalysisAgent studies your past posts — tone, length, emoji density, hashtag patterns
- InsightsAgent pulls what actually performed well
- CaptionAgent writes new posts that sound like you, not like ChatGPT
- ImageProcessingAgent handles visuals — custom backgrounds, overlays, brand consistency
- InstagramPublishAgent schedules everything through the Meta Graph API
Output: a month of content that reads like you wrote it on a good day.
Silent Collect — The Calm Morning Brief
Your bot goes quiet after hours instead of auto-replying badly. It collects everything — client messages, missed inquiries, complaints — and at 9 a.m. sends you a Telegram digest: summaries per chat, sentiment markers (green calm / yellow attention / red urgent), action items, and alerts. No dashboard required. You open Telegram, you know your day.
The Technical Stack
Core backend: Django 5, PostgreSQL 16 with pgvector, Redis, Celery, Docker Compose on Hetzner VPS, Nginx. Schema-per-tenant isolation in Postgres — full data separation, GDPR-compliant, per-tenant backups.
AI layer:
- Claude API — primary reasoning and generation
- OpenAI — fallback and vision (GPT-4o)
- Cohere Rerank — second-pass ranking over vector search, the single biggest RAG quality win before you even think about fine-tuning
- Qdrant — high-performance vector database for large tenant knowledge bases, with metadata filtering
- pgvector — in-tenant semantic search kept close to the data for smaller workloads
- Google Vision API — OCR on uploaded documents and receipts
- ElevenLabs — voice output for voice channels
- Ollama — local embeddings only; heavy local LLMs aren’t worth the VRAM on a VPS, and we’re honest about that
Frontend: React + Vite on web, React Native (Expo) on mobile.
Payments & comms: Stripe for billing, Resend for transactional email, Calendly OAuth, Telegram Bot API plus userbot for advanced scenarios, Meta Graph API for Instagram.
MCP — Where ADME Gets Really Powerful
MCP (Model Context Protocol) is how ADME agents talk to the outside world without us writing a custom integration for every tool on earth. Instead of brittle API glue, agents speak a standard language and any MCP-compatible service plugs in.
Currently connected MCP servers:
- Google Calendar MCP — agents read and schedule events natively
- Email MCP — IMAP search, SMTP send, attachment parsing
- BrowserBase MCP — agents open real browser sessions and perform actions on sites that have no API
- BrightData MCP — structured web scraping for lead generation and competitor research
- Awesome MCP Hardware — and this is the fun part
Hardware MCP — Yes, ADME Can Control Physical Devices
Through the Hardware MCP layer, ADME agents can reach out to physical devices on your network. The most interesting use case: Raspberry Pi integration.
Imagine a beauty salon where the AI receptionist doesn’t just book appointments, but:
- Turns on the welcome light and plays the client’s preferred playlist when they arrive
- Controls a smart display at the reception showing today’s schedule
- Reads temperature sensors in treatment rooms and adjusts climate before a client walks in
- Triggers a print job for the client’s receipt via a Pi-connected printer
- Manages door access for staff through a Pi-controlled relay
For technical users running Home Assistant, Pi clusters, or custom hardware — ADME becomes the natural-language interface to all of it. You write “turn off the UV lamp in room 2 after the appointment ends” and it happens. No Node-RED flow, no custom script.
For non-technical users — this all sits invisible under “smart automations” in your dashboard. You toggle what you want, we wire it up.
What Makes ADME Different
Not another ChatGPT wrapper. Real multi-agent orchestration with a router that picks the right model for each job — cheap and fast for small talk, deep reasoning for complex workflows. You don’t pay Claude prices for “hello.”
Built on open standards from day one. MCP-ready agents, JSON Schema contracts, stateless sub-agents. When the ecosystem moves, we move with it in days, not months.
Real multi-tenancy. Every business gets a fully isolated Postgres schema. Your client data never touches another salon’s knowledge base. Your Instagram style profile is yours.
Honest about costs. Hybrid pricing — a low monthly base plus a credit system (“bells”, 1 bell = $1) with 10 free on signup. No surprise bills. No pure pay-as-you-go anxiety.
Open to developers. Full API access, webhook events, MCP server compatibility. If you can describe it, an agent can do it.