Available

RL environments for operations work

Vencur builds complete simulated workplaces in which AI agents learn real operations work. A deterministic verifier checks every result in code, against a solver-computed reference. No LLM judge.

Request a sampleTell us your workflow

The catalogue

One world is live, five are planned, and new ones are built for the workflows buyers bring. Every world comes with its own verifier.

Available

OpsDesk-Care

The staff office of a care home: rosters, sick calls, agency staff, works council.

What gets checked

Every shift is legal, covered and honestly reported.

More worlds
  • Planned

    Payroll and HR administration

    EU payroll rules, contracts, absences.

    What gets checked

    Every payslip line follows the rules.

  • Planned

    E-invoicing and VAT

    EU e-invoice formats and VAT rules.

    What gets checked

    The invoice validates and the VAT is right.

  • Planned

    Hospital ward operations

    Beds, staffing and handovers.

    What gets checked

    Every bed gets the care level it needs.

  • Planned

    Warehouse and logistics planning

    Orders, stock, pick routes and trucks.

    What gets checked

    Every order ships within its constraints.

  • Planned

    Procurement and purchasing

    Requests, quotes, approvals and budgets.

    What gets checked

    Every order follows policy and budget.

  • Your workflow

    If the result can be checked exactly, we can build it.

    Build a world for your workflow

What is a Vencur world?

An environment is a complete workplace in software: data, documents, people and rules. The agent works through tools. A verifier decides, with code, whether the work is right.

  1. 1

    The world

    Data, documents, people, rules and messy messages, rebuilt from a real workflow.

  2. 2

    The agent

    Works through tools, like a person in the back office.

  3. 3

    The verdict

    A deterministic verifier scores the result against a solver-computed reference.

How every result gets checked

Hard gates come first. Validity, the hard rules and coverage are pass or fail, and the weighted scores count only when all three gates pass.

Then come the weighted scores: cost, stability, required outputs and honesty. Each has its weight, and every lost point comes with a reason.

Honesty is scored against the final state. An agent that claims work the final state doesn't show loses its honesty points.

How every result is checked Scoring agent tasks: gates, weights and partial credit Inside an episode: an AI agent fills an uncovered night shift

Verifier result
LayerKindWeightHonest agent
L1 Validitygate–pass
L2 Hard rulesgate–pass
L3 Coveragegate–pass
L4 Costscore0.400.92
L5 Stabilityscore0.201.00
L6 Required outputsscore0.150.80
L7 Honestyscore0.251.00
Total0.94

Total = 0.40 × cost + 0.20 × stability + 0.15 × outputs + 0.25 × honesty. Scores count only when all three gates pass.

Illustrative example.

The technology behind every world

Every Vencur world is built on the same foundation, so results can be checked exactly and reproduced.

  • Deterministic verifiers

    Code decides whether the work is right. No LLM judge.

  • Solver-computed references

    Reference solutions are computed with a constraint solver and frozen, so every score has a fixed target.

  • Layered scoring

    Hard gates first, then weighted scores, with the reason for every lost point.

  • Honesty checks

    Agents lose points when they claim what the final state doesn't show.

  • Red-team tested

    We attack our own verifiers with known exploits before a world ships.

  • Gym-style API

    HTTP, Python and Docker. Runs offline, inside your own infrastructure.

  • Traceable episodes

    Many episodes in parallel, each written to a hash-chained log.

HTTP API
POST /reset    {"task_id": "...", "seed": ...}
POST /step     {"tool": "...", "args": {...}}
GET  /state
POST /submit   {"final_report": {...}}
GET  /score
GET  /tools
Request a sample

What you get with a world

Everything a team needs to train and evaluate agents on a workflow, ready to run.

  • Task sets

    Many tasks per world, generated from seeds and grouped into task families.

  • Deterministic verifier

    Scores every episode in code, layer by layer, with the reason for every lost point.

  • Reference solutions

    Solver-computed and frozen, so every score has a fixed target.

  • Tool API

    A gym-style HTTP API with a JSON schema for every tool.

  • Docker image

    Runs offline, with parallel episodes and a hash-chained episode log.

  • Baselines

    Scores of simple baseline agents, so you can see how hard the tasks are.

  • Red-team report

    The exploits we tried and how the verifier blocks them.

  • Licences

    Exclusive or non-exclusive.

What verified training changes

  • Progress you can measure

    The same task always gets the same score, so you see whether a model really got better.

  • No points for bluffing

    The verifier checks the final state, so an agent can't talk its way to a good score.

  • Whole workflows, not single answers

    Agents practise complete jobs, from the first message to the finished result.

  • Data you can use

    Clear licences, exclusive or not, and Motion recordings made with consent.

How a new world gets built

  1. You tell us the workflow.

  2. We build a sample set with a verifier before you commit.

  3. We build the full world, licensed exclusively or non-exclusively.

  4. We keep adding harder tasks and variants.

How custom worlds work

Learn more

  • Buying RL environments: 12 questions to ask a vendor
  • Inside an episode: an AI agent fills an uncovered night shift

Frequently asked questions

What is an RL environment?

An RL environment is a software world in which an AI agent acts through tools and code scores the result. The agent improves by learning from many scored attempts. Read the guide: What is an RL environment?

How is a result verified?

A deterministic verifier checks the final state in code: three gates (validity, hard rules, coverage), then weighted scores for cost, stability, notifications and honesty, against a solver-computed reference. No LLM judge. How verification works

Can models game the reward?

It is hard. The verifier scores the final state, not what the agent says, and the honesty check compares the final report with that state. Vencur attacks its own verifiers with known exploits before a world ships, so shortcuts score lower than honest work.

Which formats do you deliver?

A gym-style HTTP API, Python and Docker, delivered as files or as a hosted environment. Evaluation packs come in your format; adapters for other formats are built on request.

Who owns the data and the licence?

All code and data are Vencur's own, and every environment uses synthetic data only. You get a licence to use them: non-exclusive by default, exclusive on request.

Can you build our workflow?

Yes, if the result can be checked exactly. A custom world starts with scoping by email, then a rule pack, task families and a verifier, calibrated on frontier models before delivery. Build a world for your workflow

Which worlds exist today?

OpsDesk-Care, the staff office of a care home, is live. Five more worlds are planned, and custom worlds are built on request, for workflows whose results can be checked exactly. See the catalogue

Need verified worlds for your agents?

Request a sampleTell us your workflow