What it does
The work around the work, handled.
Open it from any page with ⌘J (Ctrl+J on Windows), or press Ask Thames on a company, contact, client, deal, project, meeting, proposal or invoice. It knows which record you're looking at.
Answers from your records
Who a contact is, what's overdue, how many hours are left on a budget. It reads the real records, not a guess.
Does the busywork
Log time, move a deal, create tasks, draft an invoice from the hours. Over 100 actions across the CRM, projects, time, money, email, meetings and planning.
Makes the file
PDF, Word and Excel files and charts, with your workspace's branding, ready to download.
Researches the web
A research run reads several sources and comes back with a short report and its links.
Talks it through
Switch to Voice and ask out loud. Same skills, same rules. Voice runs on your own OpenAI key.
Remembers
Tell it how you like things done, just for you, or for the whole workspace when an admin saves it.
Briefs you each morning
Switch on a daily brief at the hour you choose: in the app, and by email if you like. It uses no AI requests.
Summarises meetings
Once an admin switches it on, recorded calls are summarised onto the deal, project or contact, with suggested tasks for you to approve.
Reads what you attach
Drop in a PDF, image, Word, Excel, CSV or text file - up to five at a time - and ask about it.
You stay in charge
People do the work. Thames does the rest, with a yes.
Built for a team that wants help, not a second boss. Four rules, enforced in the product rather than promised in a prompt.
- 01
It asks before anything leaves.
Every email, invoice and proposal it sends, every delete, every person it books and every memory it saves waits on an approval card. You approve with a tap, or a yes on a voice call.
- 02
It can only do what you can.
Thames works with the permissions of the person asking. A module your role can't open, it can't open either.
- 03
Most changes can be undone.
For 10 minutes after Thames makes a change, one click puts it back.
- 04
Outside content doesn't get to steer it.
Once a chat has read an email, a form or a web page, every change it makes needs your approval, so text written by someone else can't act through it.
Your data isn’t used to train anything. Requests on your plan go to Anthropic under our account, and Anthropic doesn’t train its models on them. With your own key, that provider’s terms apply.
What it costs
Included in every plan. It never runs up a bill.
One request is one chat answer, one web research run or one meeting summary. Short questions about how to use Nucle.OS, and the daily brief, are free.
200
requests a month on Solo
150
per seat a month on Studio and Agency, shared by the team
£0
for going over. At the limit Thames stops, and nothing is added to your bill
- Need more this month? A pack of 500 requests is £29 + VAT, once. It doesn’t expire.
- Or add your own Anthropic, OpenAI or Google key and choose the model per chat. That use is billed by the provider, with a $50 monthly limit set by default that you can change.
Not yet
What Thames can’t do, yet.
- Build or run workflows. It can list them; you build them.
- Add people to email sequences. It can take them off.
- Edit an invoice that already exists. It drafts new ones and sends them.
- Voice without an OpenAI key of your own.
Getting more useful as we build.
Thames learns a new skill whenever we add one, and what we build next is shaped by what you ask it for. Tell us the job you’d hand it at support.