Raw: https://promptreports.ai/docs/products/promptcompute.md

# PromptCompute

Approvals: the exact action goes to a person before it runs, and each decision is signed. Also holds agent identity and PromptReports ID.

- [PromptCompute docs →](https://promptcompute.ai/docs/decide-requests)
- [promptcompute.ai](https://promptcompute.ai/)

## How an agent reaches it

|   |   |
|---|---|
| MCP | `https://mcp.promptcompute.ai/mcp` Built |
| Auth today | OAuth 2.1 for MCP clients (Client ID Metadata Documents preferred), connect keys per product, signed agent requests (RFC 9421). |
| ID agent token | Planned (LDG-2264) |

## MCP tools

`request_approval` `get_approval` `cancel_approval` `answer_approver` `report_execution` `report_outcome` `list_action_types` `list_resource_lessons`

## Actions a person may need to approve

`grant.create` `access.grant` `access.revoke` `engine.switch` `agent.output.release`

From the PromptCompute action catalog.

## Workflows that use it

- [Launch with proof](https://promptreports.ai/use-cases/launch-with-proof/) Partly built
- [Customer escalation to recovery](https://promptreports.ai/use-cases/escalation-to-recovery/) Partly built
- [Hire an agent](https://promptreports.ai/use-cases/hire-an-agent/) Planned
- [A trust packet a buyer can check](https://promptreports.ai/use-cases/trust-packet/) Partly built
- [Cost per outcome](https://promptreports.ai/use-cases/cost-per-outcome/) Partly built
- [Discount and promise guard](https://promptreports.ai/use-cases/discount-and-promise-guard/) Partly built
- [Incident to policy](https://promptreports.ai/use-cases/incident-to-policy/) Partly built
- [Graded research](https://promptreports.ai/use-cases/graded-research/) Partly built
- [Investor update with proof](https://promptreports.ai/use-cases/investor-update/) Partly built
- [Evaluate an agent before you buy it](https://promptreports.ai/use-cases/evaluate-an-agent/) Planned
- [Ask across the family](https://promptreports.ai/use-cases/ask-the-family/) Planned
- [Contain an agent that went wrong](https://promptreports.ai/use-cases/contain-an-agent/) Proposed
- [Switch models with proof](https://promptreports.ai/use-cases/switch-models-with-proof/) Proposed
- [Answer a security questionnaire from the record](https://promptreports.ai/use-cases/security-questionnaire/) Proposed
- [SOC 2 and EU AI Act evidence from the record](https://promptreports.ai/use-cases/audit-evidence/) Planned
- [An agent buys context it can trust](https://promptreports.ai/use-cases/buy-context-you-can-trust/) Proposed
- [What your agents did this week](https://promptreports.ai/use-cases/weekly-agent-digest/) Proposed