May update: refreshed chat, in-app support, memory, and one simpler plan

A lot landed this month. Here is a quick walk through what changed, why we changed it, and how we are working now.

A tighter chat

The daily-use surfaces feel more consistent. Mobile chrome is now unified across the native shell and the web, so the app behaves the same whether you opened it from your home screen or a browser. The chat drawer drags from anywhere, the streak strip lines up with the day labels, and the caret stops doing odd things mid-message. Small fixes that add up to less friction in the moments you actually use the app: kit bag, train, after a workout.

In-app support

You can now message support from inside the chat. Replies land in the same thread you sent them from, so there is no inbox to chase and no email lag. If something feels off or a tool call goes sideways, we want to hear it in seconds, not days.

Persistent memory

Intervals Pro now remembers context across conversations. Your goals, your training philosophy, the constraints you have told it about, your race calendar, the things you said you do not want to hear about again. You stop having to re-explain yourself every Monday. The memory is yours: you can read it, edit it, and clear it from your settings.

Smarter AI, sharper tools

The AI has gotten better at picking the right tool for the right job, and we route across frontier models so each request lands on the one that is strongest for it. Simple questions get a fast answer. Deep analysis and multi-step planning get a deeper model. None of this is something you have to configure. We benchmark new model candidates against real traffic every day and roll in the winners on canary first, then everyone, when they win on quality, latency, or cost without giving up either of the other two.

One Pro tier, fair-use

We collapsed the old multi-tier ladder into a single paid plan: Pro, at £5.99 a month or £59.99 a year. Intervals Pro now starts with a 30-day trial, then Pro. No power-user tier, no per-feature gating, no usage meters in the corner of the screen.

Pro is unlimited with a fair-use safety net behind the scenes. The cap exists to catch runaway scripts or genuine abuse. In normal training use we do not expect anyone to come close to it. If you are an actual athlete using the chat to plan your week, get your daily check, analyse a workout, and import a plan, you will not notice it.

This is also a direction change. We used to lean on per-user budgets you had to manage, and that framing put a meter in front of the conversation. We think the conversation is the product. The new shape: you pay for Pro, you ask what you need, we worry about the costs on our side.

How we work now

We are iterating on the system daily. In practice that means three things, in order:

1. Best model for the job. We bench candidate models against real traffic replays every day, project the quality and cost impact, and roll in the winners on canary first, then everyone. Where a frontier model regresses on a workload (rare but it happens), we roll back.
2. Customer feedback addressed fast. The new in-app support feeds straight into our triage. If you flag something that is costing you training time, that goes to the top of the queue.
3. Features shipped carefully but quickly. Staging first, prod second, with the same checks both sides. Most changes are out within a day of merge. Bigger work goes through a staging soak and a small group of athletes before everyone sees it.

What is next

More of the same, with the dial turned up. Better memory recall, deeper plan-building, a few rough edges on mobile that are already under the microscope. And, well, coming to an app store near you?

If something is bugging you in the meantime, send a note from inside the chat. We read everything.