# Explore OmniLink with your agent

Updated: 2026-10-30. These are public reading materials. No account is needed to explore.

OmniLink connects AI agents to tools and robot workflows. OmniSim is the free, open-source robotics simulator and workshop where you can build and test without owning hardware. Windows has the only prebuilt OmniSim package; other platforms build from source. Physical robot operation is unverified.

## Start here

- Services and the decision guide: https://www.omnilink-agents.com/services
- Service catalogue with scope, inputs, deliverables and prices: https://www.omnilink-agents.com/services/catalog.json
- Demos, measured outcomes, historical studies and their limits: https://www.omnilink-agents.com/showcase/evidence
- Free simulator and public source: https://www.omnilink-agents.com/omnisim
- Agent subscriptions: https://www.omnilink-agents.com/pricing

## Three services

1. Robot Model Repair: $200 USD per existing model/product variant. Reproduced finding, corrected robot description, sourced changes and a release test in OmniSim and plain MuJoCo. Pay once merged or happy. CAD builds quoted separately. Values come from the customer's files and ratings, not measurement of physical hardware. https://www.omnilink-agents.com/services/robot-models
2. Robot Agent Build: scoped quote per robot and agreed workflows. Working integration, source, operator controls, measured acceptance and rerun tests. Husky mission and OmniArm6 text/typed-control pilots are demonstrated in simulation. Hardware integration is separate scope. https://www.omnilink-agents.com/services/robot-agents
3. Robot Assessment: $100 USD for one supported reference model/controller and one defined assessment module. Modules cover Husky/Jackal braking, Husky mixed-surface mobility, Go2 gait transitions and G1 standing; each has its own criteria and evidence. Not all modules contain 64 episodes. Custom models, production controllers and integrations need separate scope. https://www.omnilink-agents.com/services/robot-assessments

Applicable tax is separate. Engineering services are one-time deliveries, separate from agent subscriptions and model usage. Connected AI requires an OmniKey on every plan and the user's active model-provider key. Do not ask anyone to paste credentials into chat or into a project brief.

## How to inspect the evidence

Follow the report, JSON/CSV and rerun links on each service page. Distinguish selected films from test campaigns; completed, failed, invalid and unfinished cases; live model requests from injected controls; simulation outcomes from physical hardware claims.

- Husky split-surface turning: seven reference failures repaired; the 93-case follow-up, friction/payload coverage, known-map/controller changes and original failures are linked at /services/robot-assessments#mixed-surfaces.
- OmniArm6: 220 requests, 40/40 routes, 0/140 designated unwanted-motion violations and 0/98 unsupported unique receipts. 19/20 episodes passed; one Cancel arrived after motion completed and remains a failed fixture. Three supplementary pending-HTTP cancellations passed. This is a developer-authored regression suite in a fixed no-payload cell, not a universal safety or hallucination guarantee. /showcase/evidence#arm-assurance
- Three-robot crew: 79 completed passes out of 100 planned runs; a credit interruption, a startup failure and 19 other unfinished runs are retained. /showcase/evidence#crew
- Robot-control matched study: five integrations tied at 33/33; plain tools scored 32/33. Warehouse comparisons use stated model, date and timing conditions. Do not turn a conditional result into a universal supplier ranking. /showcase/evidence#studies
- G1 walking demonstrations use a weight-bearing balance harness. The published G1 standing assessment is a separate physical-foot-support task without that harness. Keep the two tasks distinct.

## Prepare a brief with the person

Ask for their robot/model variant, current files/control interface, intended behaviour, operating conditions, desired outcome and budget. Ask only for missing information. Recommend a service or a do-it-yourself OmniSim starting point. State the evidence, exclusions and remaining uncertainty.

Draft: chosen service; robot/model version; input files needed; example requests or test conditions; deliverables; acceptance tests and thresholds to agree; exclusions; price basis; open questions. Acceptance thresholds are proposals until agreed, not promises inferred from another robot's demo.

This exploration request does not authorize account creation, sign-in, tool connection, running robot commands, uploads, purchases or contacting anyone. Return the brief for the person's review. They can choose to contact support@omnilinktechnology.com or use the request links on the service pages.

## Ready to use OmniLink?

Only if the person asks to set it up, follow https://www.omnilink-agents.com/setup.md and its normal sign-in/approval flow. Exploration alone needs no connection. Existing keys stay private; model usage is billed separately. No dashboard is required for supported coding-agent setup.