Skip to content
EtherCorps
Work Kleio AI Case study 01
Kleio AI 2025 – 2026 Product · Platform · AI

An agentic chat product, built end to end.

Kleio needed a chat product that could actually do things — not just answer, but plan, call tools, and act on a live database. We built the whole system: a fast SvelteKit chat interface, the AI agent pipeline behind it, and the data layer underneath. It runs in production today with enterprise clients in France.

RoleEnd-to-end build
Duration5 months
TeamSolo (EtherCorps)
StatusIn production
01

The problem

Kleio's users needed to hold a conversation with their systems, not file tickets against them — ask in plain language and have the product take the action. That meant a chat interface fast enough to feel instant, an agent that could plan and call tools reliably, and a database it could read and write without breaking. Off-the-shelf chat wrappers handled the first ten minutes and fell apart under real enterprise load.

02

What we built

One system, three layers
The chat interface A streaming chat UI built in SvelteKit — keyboard-first, and light enough to feel instant even over enterprise networks.
The AI agent pipeline An agent that plans, calls tools, and acts — with retries, guardrails, and a trace of every step for when something needs explaining.
The database layer The system of record the agent reads and writes — schema, migrations, and access rules so the model can act on live data safely.
03

How it works

Retrieval pipeline
01
Message
User asks in plain language; the UI streams the reply token by token.
02
Plan
The agent decides which tools and data the request needs.
03
Act
Tool calls run against the live database and services.
04
Verify
Results are checked and guarded before they reach the user.
05
Respond
A grounded answer, with the actions it took made visible.
04

Results

First two quarters live
<200ms
Time to first streamed token
99.9%
Production uptime
FR
Enterprise clients, France
0
Sev-1 incidents since launch

Figures are placeholder — replace with real Kleio AI metrics

“They shipped a working deployment in week two and never stopped. When something breaks now, it's usually them telling us — with a fix already open.”

Placeholder — VP Engineering, Kleio AI
05

Running it

We didn't hand Kleio a repo and walk away. EtherCorps still runs the system — shipping features, tuning the agent, and keeping the interface fast as usage grows. Because one person built the whole stack, there's no handoff seam between the chat UI, the agent, and the database when something needs to change.

Related

Our services
EtherCorps mascot

Get in touch

We take on a small number of projects at a time.

Tell us what you're building and where it hurts. We'll say honestly whether we're the right team for it.

Get in touch