MyHarness
EN FR
Talk to us
Model Example Services Founders Talk to us
AI engineering studio · MCP & agentic tooling

Harness AI inside the systems you already run.

MyHarness builds the connective tissue between modern AI and your existing infrastructure — MCP servers, skills, and custom agents that act on your real data, safely.

Start a conversation → See how it works

Founded by computer-science professors Christian Jauvin & Daniel Lemire.

myharness · agent-bridge
ReasoningAI agent
MyHarnessMCP server + skills
tool inventory.search tool orders.create skill reorder-planner
yourdatabases
yourinternal APIs

Built on open standards

Model Context Protocol Agent skills Your cloud & on-prem No vendor lock-in
The model

Your software already knows the answers. We let AI ask the questions.

Most companies don't need another AI product bolted on from the outside. They need their own systems — the databases, APIs, and workflows that run the business — to become things an AI agent can reason about and act on. That bridge is what we build.

MCP servers

We wrap your data and operations in a Model Context Protocol server — a clean, governed interface any AI agent can connect to. Your systems stay where they are; the agent simply gains a vocabulary to use them.

Skills & agents

On top of those tools we author skills — reusable, expert procedures that encode how work should be done — and assemble agents that follow them reliably, with the guardrails your domain requires.

Integrated, not bolted on

Everything runs in your environment, against your live data, under your access rules. The result feels less like a chatbot and more like a capable new colleague who already knows your tools.

How we work

From scattered systems to a working agent — in four steps.

1

Map

We learn your domain, data, and the workflows people actually repeat by hand every day.

2

Expose

We build an MCP server that turns those systems into safe, well-described tools.

3

Teach

We author skills that capture your experts' judgment so the agent acts the right way.

4

Deploy

We ship it into your environment, measure it, and hand over something your team owns.

Illustrative example A day in the life

A specialty wine importer, suddenly fluent in its own catalogue.

Imagine a mid-sized importer and distributor of fine wines. Decades of knowledge live in an inventory database, a tangle of supplier price lists, and the heads of a few very busy people. Here is what happens once MyHarness wires it together.

Before

  • Stock levels, vintages and allocations live in a system only two people can query confidently.
  • Supplier lead times and minimums are buried in PDFs and email threads.
  • A simple question — "what can I pour for this event, and what should I reorder?" — takes an afternoon.

After MyHarness

  • An MCP server exposes the catalogue, live inventory, supplier terms and order history as governed tools.
  • A reorder-planner skill encodes how the buyer weighs margin, lead time and minimums.
  • Anyone on the team simply asks — in plain language — and gets an answer grounded in real data, plus a draft action.
cellar-assistant · powered by MyHarness
SM
Private tasting for 40 guests next Thursday — Burgundy focus, about $60 a bottle. What can we pull from stock, and what should I reorder to be safe?
MH
Looking at live inventory and your suppliers now.
inventory.search(region:"Burgundy", price≤65, in_stock:true)
suppliers.lead_time(items:[…])
From stock you can pour today: 3 Côte de Beaune whites (52 btl) and 2 village-level reds (38 btl), all within budget.

Heads-up: the Pinot Noir is down to 38 bottles. For a 40-guest tasting I'd reorder 2 cases — your usual supplier ships in 2 days, comfortably before Thursday. Want me to draft the purchase order?
skill: reorder-planner → draft ready
SM
Yes, draft it.
MH
Draft PO #4821 created for 2 cases, pending your approval — nothing is sent until you confirm. (Acts only through the tools and permissions you grant.)
What we deliver

Engagements built around your stack, not ours.

Every project is bespoke, but they tend to fall into a few shapes. We're happy to start small with a focused proof of value.

MCP server development

Production-grade Model Context Protocol servers over your databases, APIs and SaaS tools.

  • Typed, documented, permissioned tools
  • Read-only or action-capable, your call
  • Audit logging & access control baked in

Skills & agent design

We capture how your best people work and turn it into skills agents can run consistently.

  • Domain workflows encoded as reusable skills
  • Guardrails, approvals & human-in-the-loop
  • Evaluation harnesses to keep quality honest

AI strategy & enablement

Where the leverage really is — and how to adopt it without betting the company on hype.

  • Opportunity mapping & feasibility
  • Architecture, security & data governance
  • Team upskilling so you can own it after us
Who we are

Founded by two computer-science professors who build the real thing.

MyHarness is led by university researchers whose work sits at the intersection of machine learning, software performance, and pragmatic engineering — not slideware.

Christian Jauvin

Christian Jauvin

Co-founder · Professor, Université TÉLUQ

Christian's career spans machine learning and large-scale software across public health, video games, geospatial systems and climate modelling. Early on, he co-authored foundational research on neural language models with Yoshua Bengio — work that helped seed the very field today's AI is built on.

Co-author, A Neural Probabilistic Language Model LLMs & ML systems Software architecture Distributed cloud apps
Faculty profile
Daniel Lemire

Daniel Lemire

Co-founder · Full Professor, Université TÉLUQ

Daniel is among the 2% most-cited scientists in the world and one of the most-followed programmers on GitHub. His libraries — simdjson, Roaring Bitmaps, fast number parsing — run inside Google, Microsoft, Node.js, and the browsers on your machine. He makes software fast where it matters.

Creator of simdjson Roaring Bitmaps Top 2% most-cited scientists High-performance systems
Faculty profile
Top 2%
most-cited scientists worldwide
GB/s
throughput from libraries we wrote
Billions
of devices running our code daily
25+ yrs
combined building production systems

Let's find the one workflow worth automating first.

Tell us where your team loses the most time to your own systems. We'll tell you whether an agent can help — honestly — and what a first project would look like.

hello@myharness.xyz → Revisit the model