Product updates on Axiomat — Applied AI consulting, run by Andrew Meyer

Old records can now age out on their own

Fri, 28 Aug 2026 00:00:00 +0000

An assistant that’s been running for a while accumulates records — leads it tracked, predictions it scored, items it triaged. Left alone, an older assistant’s default views fill up with things nobody needs to see anymore.

There’s now a place to configure that yourself: a Data & Retention section in workspace settings. Set a default number of days records should stick around before archiving, and override that window for specific collections that should age out faster or slower than the rest.

Steadier market data, and fewer stuck scheduled runs

Fri, 28 Aug 2026 00:00:00 +0000

A handful of fixes, mostly for assistants that watch markets or run on a schedule.

Market data. Stock prices and market overviews could occasionally show up blank instead of a number — usually during a trading session that hadn’t fully settled yet. That’s fixed, with a real fallback instead of a blank. Sector performance for the 1-day view had a similar problem and was returning nothing at all for any sector; it now shows real numbers. When a price or sector genuinely isn’t available, you’ll see that explicitly instead of a silent gap.

Invite links, and runs that waste fewer steps

Wed, 12 Aug 2026 00:00:00 +0000

Two smaller things from this week.

An invite link gets you in

If someone at Axiomat sends you an invite link, opening it and signing in now puts you straight into Ward. No card, no plan to choose first.

This also works if you’ve been here before. Previously, anyone who had ever started a subscription — including people who had already cancelled one — was turned away with a message telling them to cancel a subscription they no longer had. That’s fixed: an invite works whether or not you’ve subscribed in the past. If you’re on a live paid plan, an invite is declined, and you’ll be pointed at the billing portal instead.

Assistants can build dashboards worth looking at

Mon, 10 Aug 2026 00:00:00 +0000

Ward has always been able to draw more than tables. Your assistants didn’t know that.

When an assistant built itself a dashboard, it was choosing from a short list — a table, a list, a couple of others — even though Ward could render far more. So a tracker following a number over six weeks gave you a column of numbers, because a line chart was never on the menu it was reading from.

More accurate market data, and honest failures instead of empty answers

Fri, 31 Jul 2026 00:00:00 +0000

Predictions stay accurate through a stock split

If an assistant tracked a stock prediction against a company that later did a stock split, the comparison could come out wrong — the original price it recorded didn’t reflect the split, but the current price always does, so the two weren’t measuring the same thing anymore.

Assistants that score predictions now re-check the price on a consistent basis at scoring time, so a split no longer throws off the result.

Your assistants can now leave each other notes

Fri, 31 Jul 2026 00:00:00 +0000

If you run more than one assistant, they’ve had no way to talk to each other — a news monitor that spots something a market tracker would care about had no way to say so.

An assistant can now send a short note to another assistant you own. The note shows up automatically the next time that assistant runs, and only once — no digging through history to find it.

Scheduled runs stop stalling, and a tidier connection catalogue

Wed, 29 Jul 2026 00:00:00 +0000

Scheduled assistants no longer stall on read-only work

If you had an assistant that watched a feed, scored trends, or ran a calculation on a schedule, it may have quietly stopped and waited for your approval — even with a policy that granted it those tools and asked for no approval at all.

Ward was being over-cautious. When it wasn’t certain whether an action would change something outside your workspace, it asked for approval — which is the right instinct, but it was applying it to work that only reads. Unattended, there was nobody there to approve, so the run stopped.

Three plans, all with a free trial

Wed, 29 Jul 2026 00:00:00 +0000

Ward now has three plans — Basic at $9.99, Plus at $19.99, and Pro at $29.99 a month — and every one of them starts with a 14-day free trial.

They all include the same features. The only differences are how many runs you get each day (3, 9 and 15) and whether Max mode is available — Basic runs Normal and High, which is the whole range that fits inside a 3-run day anyway.

Some jobs deserve a bigger brain

Sat, 25 Jul 2026 00:00:00 +0000

Not every job needs the same amount of thinking. Watching a feed for a keyword and writing a considered market analysis are different problems, and until now they ran the same way.

Every assistant now has a power level, and you set it per assistant or per run.

Normal — the default, and right for nearly everything. Monitors, trackers, digests, feed pollers.
High — a stronger model when the work needs more care: summarising, classifying, routine analysis.
Max — the most capable model, for research synthesis and multi-step judgment.

The stronger levels draw more from the daily budget your assistants already share, so turning one assistant up means fewer runs elsewhere that day. That’s the whole trade: breadth or depth, on your terms.

A new look, files you can download, and somewhere to start

Fri, 24 Jul 2026 00:00:00 +0000

The biggest visible change since Ward started.

Ward looks like itself now

Ward was built fast and looked it — everything in monospace, the palette of a terminal. It now has a considered design: a typeface meant for reading, monospace kept for the things that genuinely are data, and one calm accent instead of whatever each screen felt like.

Nothing moved. Everything reads more easily, and the eleven themes all inherit it, so switching one still changes the whole app.

Security work you shouldn't have to think about

Mon, 06 Jul 2026 00:00:00 +0000

Most of what shipped this week is invisible if it’s working. Worth saying out loud anyway, because “trust us” is not an argument.

Your connection credentials are encrypted where they’re stored

When you connect an account, Ward holds a scoped token — never your password. Those tokens are now encrypted before they’re stored, so a leaked copy of our storage yields unreadable data rather than working credentials.

The assistants themselves never see the raw credential. They ask Ward to act on a connection; Ward holds the key.

Try an assistant before you trust it

Sun, 05 Jul 2026 00:00:00 +0000

Test mode: a real run that touches nothing

You can now run an assistant in test mode. It thinks through the whole job for real — reads what it would read, reasons over live data — but every action that would change something is held back. Nothing sends, nothing posts, nothing gets written anywhere.

You see what it would have done, listed out, and a comparison against its last real run so you can tell what your edit actually altered.

Follow any feed, trigger from anywhere, and stop a run mid-thought

Thu, 02 Jul 2026 00:00:00 +0000

Point an assistant at any feed

Ask an assistant to follow a blog, a news source, a release page, a regulator’s notices — anything that publishes a feed. You don’t need to find the feed URL yourself; describe the site and Ward goes looking for one.

Feeds turn out to be the cheapest way to make a monitor genuinely useful. A competitor watcher that reads their changelog directly is a different thing from one guessing at a homepage.