Tool-use task data
Deepen AI's trained teams write realistic tasks for agents that call tools and APIs, with the right sequence of calls and a verified answer for each one.
Rate in your plan before you pay A plan within 2 working days No sales call
What a tool-use task looks like
Each task comes with the calls an agent should make and an answer checked by running them.
- Task
- What's the total of Acme's unpaid invoices this quarter, in euros?
- Expected calls
- Find the customer, list unpaid invoices from 1 July, convert the total from dollars to euros.
- Checked as
- VerifiedAnswer: €12,480.
- Reason
- A second person ran the calls against your test data and got the same answer.
- Task
- Move my 3 pm meeting with Dana to tomorrow, same time.
- Expected calls
Find the meeting, then move it to 3 pm tomorrow.- Checked as
- FixedAn availability check was added before the move.
- Reason
- Dana is busy at that time in your test data, and your guidelines require a check before moving meetings.
- Task
- Book a meeting room for the team.
- Expected calls
- Ask the user which day, what time and how many people.
- Checked as
- VerifiedThe right first step is a question, not a booking.
- Reason
- Your guidelines include tasks where the right move is to ask, not to guess.
- Task
- Refund the customer's last order.
- Expected calls
- Get the last order, then issue the refund for its full amount.
- Checked as
- Sent to youYour refund tool takes an optional reason code. Should every refund task include one?
- Reason
- Your answer is applied to every refund task in the set.
How we start on your work
No sales call. A plan in your inbox, then a short paid pilot on your real tools.
What we need from you
- Your tool and API definitions: names, parameters and what they return
- A test environment or sample data the calls can run against
- 10 to 20 tasks you've already written, or the kinds you need
- 5 minutesTell us the task
Which tools your agent uses, how many tasks you need, and the format.
- Within 2 working daysGet a plan and a price
People, hours, the rate, start date and the total price.
- 2 weeksRun the pilot
We train the team on your tools, write and verify real tasks and send two weekly reports. Then you decide.
Who writes tasks, and how answers are verified
Trained people, working in English. Each one learns your tools and guidelines before writing tasks, and every answer is verified by a second person running the calls.
Every reviewer's accuracy is measured
Items you've already answered are hidden in the real work, so we know how accurate each person is.
Nobody guesses on unclear cases
The team lead decides, or asks you. Your answer joins the guidelines.
You see the results every week
A weekly report shows work done, QA results and what's improving.
Standard task writing covers business tools such as CRM, calendar, email, search, orders and billing. Tasks that need specialist knowledge or code are quoted in your plan.
How much does tool-use task data cost?
Tool-use task data is quoted in your plan, because the rate depends on your tools and how complex the tasks are. You pay for the hours people work, with a team lead, QA checks and a weekly report included.
Set before you pay. Your plan shows the rate, team size, hours and total price.
Starts with a two-week pilot on your real tools, paid upfront, with two weekly reports.
Your data stays confidential. We work under NDA with the access you set. Deepen AI holds SOC 2 Type II, ISO 27001, ISO 9001 and TISAX.
Questions
How are answers verified?
A second person runs the expected calls against your test environment or sample data, and must get the same answer before a task is accepted.
Can tasks need several turns or tools?
Yes. Tasks can chain calls across tools, take several turns with the user, or need the agent to recover when a call fails.
Do you write tasks where the agent should ask or decline?
Yes. Some tasks are written so the right move is a question or a refusal, so your agent learns when not to act.
Can tasks cover our real edge cases?
Yes. Send real requests or failure logs, and tasks are written to cover the cases your agent gets wrong.
What format do you deliver in?
Your schema, such as JSON lines with the task, the expected calls and their arguments, and the verified answer.
Does our data stay confidential?
Yes. Your tools, test data and tasks are used only for your work, under NDA, with access you control.