Current availability
Hosted paid agent execution is not available on this website today. There is no public upload API, paid-job endpoint or official reqmote MCP server to connect to. Contact hello@reqmote.com to discuss an integration or scoped project.
What an agent can read now
The public keyword-comparison JSON contains model identities, artifact sizes, evaluation definitions, source attribution and retained replay predictions. It requires no sign-in and contains published evidence, not private customer data.
Use the keyword evidence page as the canonical human-readable citation. The board walkthrough describes a separate vision experiment. Reading either resource does not start a new computation.
Preserve each result’s evidence scope: command-clip accuracy is not event recall; false-command windows are not activations per hour; offline evaluation does not establish target-board performance.
Planned paid workflow
- Connect with delegated authority. A paid customer grants the agent specific data, action and budget permissions.
- Upload and verify data. A resumable transfer produces an authorized input reference after validation and integrity checks.
- Preflight the action. Receive the supported target/workflow, expected evidence, estimate, constraints and quote expiry.
- Accept within the approved budget. Submit the exact quoted request using explicit customer or delegated authority.
- Receive a durable job ID. The service continues independently while the agent disconnects.
- Return for the result. An authorized agent or customer retrieves status, results and retained artifacts using that job ID.
This describes intended behavior, not live endpoints or a service-level commitment. Pricing, supported paid workloads, connection instructions and retention windows will be published with the service release.
When the answer takes a day
The job should continue without an open browser tab, HTTP request or tool call. The agent’s host must retain the job ID and arrange a later status check. A webhook can notify a customer-owned orchestration endpoint; it cannot reopen an arbitrary LLM conversation.
The planned HTTP interface and MCP adapter will refer to the same job and evidence. Hosts that support asynchronous task handling can use it; others can receive a prompt job receipt and poll later. A 24-hour example is not a promised turnaround.
What the eventual connection guide must cover
- Actual capability availability, authentication, supported clients and versioned schemas.
- Upload sizes, data handling, delegated permissions and spending limits.
- Idempotent submission, status, missing-input requests, cancellation and recovery after disconnection.
- Result retention, artifact integrity, usage reconciliation and the difference between completion and meeting the requested quality target.
Until those controls and a real execution path are qualified, this page remains an availability guide. The current capabilities and maintainer information explain how to engage today.
