Quelm
Work with us
AI Software Studio

This is the future of how software gets built.

Quelm builds AI-powered software — for our clients and for ourselves.

agent.ts
import { Agent } from 'quelm'
const agent = new Agent({
model: 'gemma4:e2b',
offline: true
})
await agent.run()
// ✓ complete
Live signals
streaming
Burnout risk detected9 / 10
API calls blocked2,847
Cost prevented£1,240
Entries analysed14
Model layer · Gemma 4
POST
/v1/analyse
GET/health2ms
POST/v1/score18ms
GET/v1/trend5ms
DEL/v1/entry3ms
Budget limit reached
Agent 'scraper-v2' blocked
limit: £10.00 / day
spent: £10.00 ↑
calls: BLOCKED ✕
TokenCap enforced in 47ms
Local storage
journal.json
offline only✓ private
TokenCap
LIVE
220+ models · sub-50ms enforcement
Daily budget£25.00
Calls blocked today2,847
Cost prevented£1,240
Buildpassed12s
Test suitepassed34s
Deploy to prodrunning...
Production · v1.0.0 · ↑ 100% uptime
Voice · Whisper
🎙
Offline · 72MB · local
99 languages
Burnout risk · 9/10 · High
“exhausted” mentioned this week — up from .
✦ Step outside for 5 minutes today.
Pattern confidence: High · 9 signals
Quelmassembling...
gathering components...

We build software that
didn't exist yesterday.

Custom AI software

We build AI tools for companies that need something specific. No templates, no off-the-shelf — just the tool your team actually needs.

Our own products

We ship our own software under the Quelm name. TokenCapAI is the first — spend enforcement for teams running AI agents. Building products of our own keeps us honest about what's hard.

Shipped under our name.
Used in the real world.

TokenCapAI

AI cost control. Per customer. Actually enforced.

Beta

TokenCapAI caps AI spend per end-customer and blocks calls before they exceed budget — not emails you after the bill is gone. The same per-customer data lets Finance bill customers for their actual AI usage. Works across OpenAI, Anthropic, Gemini and 220+ models with sub-50ms enforcement.

Per-customer capsSub-50ms hard blockBill customers for AI220+ modelsFree tierFrom £25/mo

Quietburn

Burnout detected. Privately.

Open source

An offline AI journaling companion that watches your entries over time and warns you about burnout before you notice it yourself. Runs entirely on your device — no data ever leaves your machine.

Fully offlineGemma 4 via OllamaVoice journalingMultilingual

Our principles.

01

Simple beats clever

We pick small, boring architectures and pay the cost of doing them properly. Atomic where it must be atomic. Observable everywhere. No fashionable abstractions for problems we don’t have. Most software fails because it’s too complex for its purpose — not too simple.

02

A clear no protects a clear yes

We say no to work that isn’t ours, customers who’d be wrong fits, and features that sound right but solve the wrong problem. Narrow focus beats broad approximation — for the product, the customer, and the people building it.

03

Plan for what breaks, not just what works

Every system we ship answers one question: what happens when this fails? Recovery playbooks before we need them. Detection before automation. Audit trails that prove what actually happened, not just what we hoped.

04

Ship something real, then sharpen

We put the minimum credible thing in front of real users, then iterate from what they actually do. Perfect is the enemy of usable. Usable is the start of compounding.

Work with us —
or be first to know.

Whether you need custom AI software or want early access to what we're building, we'd like to hear from you.

Custom projects

Build with Quelm

Tell us what you're building. We'll be honest about whether we can help.

Early access

Product waitlist

Get early access to TokenCapAI and Quietburn. We keep the list small and go deep with early users.

Available products

TokenCapAI

AI spend control

Beta

Quietburn

Offline burnout detection

Open source