AI agents & LLM assistants
Agents built on Claude and other LLMs: tool use, retrieval over your own knowledge base (RAG), Telegram and web bots that do real work instead of just chatting.
We are a small studio. We build systems we run ourselves every day, so we know what breaks in production and how to keep it from breaking.
Services
Agents built on Claude and other LLMs: tool use, retrieval over your own knowledge base (RAG), Telegram and web bots that do real work instead of just chatting.
Self-hosted n8n, integrations with the tools you already use, webhooks, retries and error handling, so a failed step raises an alert instead of silently losing data.
Collection, cleaning and storage of data with Node.js, PostgreSQL/Supabase and Playwright, including sites that need a real browser.
Monitoring, safe deploys with rollback, and human-approval gates for actions that touch money, customers or live sites.
Track record
Not a demo environment: this is our own stack, in daily use.
More than 750 services, workers and scheduled jobs under automated supervision
n8n workflows in our own stack
More than 5,000 production releases shipped through a single deploy gate with rollback
More than 100,000 LLM calls logged
Telegram assistant bots on one shared framework
Two RAG knowledge bases with more than 1,500 curated claims (pgvector)
Figures from our own infrastructure logs, September 2026.
Case studies
B2B equipment distributor
An SEO agent that edits titles, meta tags, schema markup and internal links on a catalog of hundreds of products. Every change passes a gate with an automatic rollback on a failed check, blast-radius limits per run, per day and per share of the catalog, and a regression suite after each publication.
Stack: Node.js, PostgreSQL/Supabase, Playwright, Claude/LLM
B2B equipment distributor
An agent that runs a paid-search account: an anomaly detector, auto-fixes behind an approval gate, and a weekly planned-vs-done report with before/after measurement of every change.
Stack: Node.js, Yandex Direct API, PostgreSQL, Telegram bot
Any business with lead forms
Playwright tests of live lead forms on production sites. Test leads are flagged so they never reach client reports, and a new client is onboarded into all monitoring automatically.
Stack: Playwright, Node.js, PostgreSQL, Telegram
Digital agency & sales team
Several assistant bots on one shared framework: dialogue memory, voice messages, approval buttons for proposed changes and an outgoing message queue.
Stack: Node.js, PostgreSQL, Telegram Bot API, Claude and other LLMs
Marketing expertise
A knowledge base built from open sources with a reliability score per source. Agents consult it before making decisions instead of answering from memory.
Stack: PostgreSQL + pgvector, local embeddings, Node.js
Client names and results are confidential by default — yours will be too.
How we work
We discuss the task, your systems and what "done" looks like.
A written scope, milestones, price and timeline before any work starts.
Each milestone ends with a working demo and your sign-off.
Code, workflow JSON and a runbook, so your team can operate it.
We build and test on our own infrastructure and accounts. You receive code and documentation and deploy on your side following our runbook (or we guide your team). We don't need access to your cloud AI accounts, GPUs or ERP.
Pricing
Hourly
$45 / hour
For open-ended work, audits and small fixes.
Single n8n workflow
from$600
One well-tested workflow with error handling and documentation.
AI agent, turnkey
from$2,500
Design, build, testing and handover of an agent for a defined job.
Monthly support & improvements
from$1,500 / month
Monitoring, fixes and iterative improvements of what we built.
Fixed-price per milestone. Invoiced in USD or EUR; 50% upfront for the first milestone.
Contracting & payments
Contact
A few sentences are enough. We reply by email or Telegram.
Telegram
@aipepden