Acts on what keeps showing up in your data.
Upshot is an AI agent platform for operations teams. It watches the systems you already run, finds the pattern behind each repeat problem, and does the follow-up work.
In pilots with teams in freight, software, healthcare, finance and manufacturing
Your team fixes the same problems every month. Upshot spots the pattern, drafts the fix and asks before it acts.
Everything it does, you can check.
Figures from five pilot customers. You see your own from the first week.
median model and tool spend on a run, shown on the run itself
median time from a trigger to a proposed fix
Never trained on your data.
- Model providers contracted to zero retention
- Runs in your own cloud on Scale
- Delete any run and everything it read
Connects to the tools you run now.
Read-only to start. It writes only through the actions you switch on.
Live in four weeks, with a founder beside you.
Ines Carvalho and Kwame Osei run every pilot themselves, in a shared channel with a call each week.
- Week 1Connect two or three systems with read access
- Weeks 2 and 3Upshot proposes, nothing changes
- Week 4Your team switches on what it trusts
- AfterReview every run, then decide
Free to pilot. Then pay for the runs you use.
A run is one problem taken from trigger to finished action. Reading, retries and sign-offs inside it are included.
$0.30 a run after the first 1,000. Above 20,000 runs a month, Scale pricing is usually cheaper.
Pilot
One workflow, set up with you, on one problem your team is tired of fixing.
- Read-only for the first three weeks
- Weekly review call with a founder
- Your run log and evals from day one
Team
For teams running Upshot on a few workflows every day.
- 1,000 runs included, then $0.30 a run
- Up to 10 workflows
- Sign-off rules per action and per amount
- Run log kept for a year
Scale
For high volume, your own cloud, or a security review before you start.
- Runs in your cloud account
- Single sign-on and audit log streaming
- Data kept in your region
Questions before a pilot.
What does a run actually cost us?
On Team, $0.30 a run after the first 1,000 each month. Model and tool calls are included, so there is no second bill from a model provider. Model and tool spend in our pilots has a median of $0.14 a run, and every run shows its own cost.
What happens when it gets something wrong?
Anything that changes a system, or spends more than the limit you set, waits for a sign-off, so a wrong proposal gets rejected rather than carried out. Rejections and edits feed the evals, and you can see the approval rate per workflow.
Which models does it use?
We pick the model per step, a larger one for diagnosis and smaller ones for reading and formatting. On Scale you can bring your own provider keys.
Can it act without approval?
Only for actions you mark as safe, such as posting a summary to a channel. Anything else needs a person, and you decide which person.
What happens after the pilot?
We review every run with you. If it earned its place, you move to Team and keep your workflows and run log. If not, we delete your data and you owe nothing.
What does your team keep fixing by hand?
Tell us about it in a sentence or two. We will tell you on the first call whether Upshot can take it on.
Pilot request received.
Ines or Kwame will reply within one working day with two times for a first call.