Start where your data already lives
INDIVIDUAL
Use the public playground with synthetic or authorized TXT, Markdown, CSV, or JSON files. Inspect the protected output before copying it to another system.
PROJECT
Run the gateway beside the application. Bind every operation to a policy and purpose, then keep Vault credentials in application state.
AI CLIENT
Point an existing non-streaming OpenAI client at the local /v1 endpoint and attach SHREDDAR request fields.
AI AGENT
Expose only five agent-safe tools and install the portable privacy workflow. Reconstruction stays outside model context.
The source determines the boundary
| SOURCE | START WITH | WHY |
|---|---|---|
| Synthetic product evaluation | Browser playground | Fast proof with no external inference |
| Internal documents | Self-hosted gateway | Policy, Vault, and receipts remain inside your environment |
| Existing AI application | OpenAI-compatible adapter | Minimal client change with protected messages |
| Autonomous tool use | MCP server + Agent Skill | Safe tool catalog and fail-closed workflow |
| Agent commerce | ACP blueprint — HOLD | Identity, signer, pricing, escrow, and recovery are not verified |
| HTTP token settlement | x402 — HOLD | No posted rate, destination, facilitator, or settlement receipt exists |
Prove the privacy path before adding commerce
- 01
Evaluate
Use synthetic data to inspect detection, transformations, route choice, and receipt evidence.
- 02
Self-host
Place the gateway and Vault inside the environment where authorized source data already resides.
- 03
Integrate
Choose API, MCP, or a runtime-specific Agent Skill path and test failure behavior.
- 04
Measure
Track task utility, missed sensitive classes, context reduction, latency, and output-guard escapes.
- 05
Activate settlement later
Only after a posted price, payment route, recovery plan, and independently verified receipts exist.