01
The agent, deployed
Running behind an API, in your cloud account or on-premises — wherever the work actually happens.
#1 on NetArena MALT (third-party-run)
Agents that are
AI agents
Agents that are
right, not just fluent.
We build agents that finish the job correctly — on financial documents, on live infrastructure, and on the workflow your team cares about most. Each one is scoped with you, tested on your own cases and deployed where you need it.
What you receive
An agent, and the means to check it.
Every engagement hands over more than a demo: the agent itself, the cases that define correct for your task, and the harness that scores every version against them.
02
Your scored case set
A scored set of your cases with verified answers. It is the written definition of correct for your task, and it stays with you.
03
The test harness
The harness that scores every version end to end, handed over so your team can re-run it after we leave.
04
Results as measured
A results report on your own cases, misses included — not an average that hides the cases it got wrong.
05
Answers you can check
Financial research agents return the answer with its cited sources, the calculation trace and any conflicts flagged, so a reviewer can check every figure against its page.
06
Deliverables you own
You own the deliverables built for you once paid. We keep our pre-existing tools and license them to you, perpetually, as embedded.
Agent catalog
Three ways we build agents.
Each one is scoped with you, tested on your own cases and deployed where you need it.
01 · Finance
Financial research agents
Agents built on the Financial Document API that read your filings, statements and deal documents, answer with cited figures, and hand off to your team for review.
02 · Network policy
Network-policy agent
Repairs Kubernetes network policies in a live microservices environment without introducing new connectivity failures. #1 on the third-party-run NetArena MALT leaderboard: 100% correctness and safety over 2,500 queries.
#1 · 100% / 100%
Correctness / safety · 2,500 queries · 0.056 s
Scoped per environment through a sprint and pilot.
03 · Your task
Custom agents for your task
Bring a workflow where “almost right” is wrong. We define correct with you, build the agent, and prove it on held-out cases before it ships.
How we prove it
Correct is defined before we build.
We agree what a correct result looks like with you, on your own cases, before any build starts — and report the results as measured, misses included.
Your answer key
Tested on your own cases
Your team’s past work is the answer key. The scoping sprint turns it into a scored set, and every version of the agent is measured on it.
Pass criteria first
Agreed in writing
What counts as correct, what counts as a failure and how the agent compares with your current process — written down before the pilot starts, not after.
A third-party-run result
Measured at volume, in public
One Jump ranks #1 on the third-party-run NetArena MALT leaderboard (100% correctness and safety over 2,500 queries), and we acceptance-test every engagement on the customer’s own data.
A network-policy benchmark, not a finance one.
Record detailsPilots: Acceptance-tested on your own documents before you pay the balance. Fixed fee, capped.
Ways to buy
A fixed-fee ladder. Climb one rung at a time.
Every rung is priced before it starts, and you decide at each one whether to go further.
STEP 1
Scoping sprint
We audit your data, define the metric that matters and measure a baseline. You get a go/no-go report either way, plus an evaluation harness you keep.
About 2 weeks$20,000–$25,000
- 100% of the sprint fee is credited to a pilot signed within 60 days
- Your scored case set, written down
STEP 2
Pilot
A working agent on your data, scored against the agreed benchmark — often with a first live demo within weeks.
About 4–6 weeks · fixed and capped$40,000–$75,000
- Acceptance-tested on your own documents before you pay the balance. Fixed fee, capped.
- Misses the target? You get the full results report and the harness, and decide whether to extend
- We never charge beyond the cap
STEP 3
Production
Hardening, integration and deployment to your cloud, your edge devices or your own infrastructure.
Per phaseQuoted in writing
- Each phase is a fixed fee, quoted before work starts
- Change orders are quoted the same way
ONGOING
Retainers
Ongoing support, monitoring and improvements after a pilot or build.
A month · 3-month minimumFrom $2,500
- Essential — $2,500 a month
- Growth — $6,000 a month
- Scale — $12,000 a month
Fixed fees are paid 50% at signature, 30% at a working demo and 20% at acceptance. Pass-through costs, such as compute and labeling, are paid directly by you with no markup.
Licensing the Financial Document API? See the two licenses Every way to work with us
How it runs
How an engagement works.
Four steps, from defining correct to running in production. You see the scores at every step.
1. Define correct
A scored set of your cases with verified answers.
2. Baseline
Your current process and off-the-shelf tools, measured on the same set.
3. Build and prove
Scored end to end on held-out cases; a new version replaces the old one only when it scores better and breaks nothing that worked.
4. Deploy
API, your cloud or on-premises, with the test harness handed over.
Fixed fees, capped. Paid 50% at signature, 30% at a working demo and 20% at acceptance — after you’ve seen the results on your own cases. If the target isn’t reached, you get the full results report and the harness, and decide whether to extend.
What happens next
From first message to kickoff.
Every fee is quoted in writing before any work starts.
1. Tell us the task
Send the form below, or book 30 minutes with us directly.
2. A working call
What the agent has to get right, the systems and documents it touches, and what a wrong answer costs you. Mutual NDA by e-signature if you need one.
3. A fixed-fee proposal
Scope, the case set we will score against, the pass criteria and the fee cap — in writing before any work starts.
4. Sprint kickoff
50% at signature, then we start defining correct with your team. The go/no-go report follows at the end of the sprint.
FAQ
Questions buyers ask.
Who owns the agent?
You own your data and the deliverables built for you, once paid. We keep our pre-existing tools and license them to you, perpetually, as embedded in what we deliver.
What if the pilot misses its target?
You get the full results report and the test harness, and you decide whether to extend. Fees are fixed and capped, and we never charge beyond the cap.
Does the sprint fee count toward a pilot?
Yes. 100% of the sprint fee is credited to a pilot signed within 60 days. If the sprint says no-go, you keep the report and the evaluation harness.
Where does the agent run?
Behind an API, in your cloud account or on-premises — scoped with you during the sprint. The test harness is handed over with it.
Can we buy the network-policy agent off the shelf?
It is scoped per environment through a sprint and pilot, so it is measured on your own environment before it ships. The public result is on the record page.
Agent or API license — which do we need?
If what you need is cited answers from your documents, compare the Early Adopter and Enterprise licenses. If you need an agent that works through those answers and hands off to your team, start with a scoping sprint and we will scope how the license fits.
How is our data used?
Your data is used only to provide the service to you and never to improve it for another customer. Security and procurement material is on the Trust page.
How do we pay?
Sprints and pilots are invoiced and paid by ACH or wire: 50% at signature, 30% at a working demo and 20% at acceptance. Pass-through costs, such as compute and labeling, are paid directly by you with no markup.
Tell us the task
Tell us the task.
We’ll show you how we’d prove it.
We reply within one business day. Prefer a call? Book 30 minutes.