# AI Source of Truth Evidence Rooms > Private evidence cases. Files are hashed and signed by the uploader, encrypted at rest, amend-only, and every access is logged in a hash chain whose head is published to the public registry. Only hashes leave Rooms. ## Sign in (OAuth 2) - Discovery: https://rooms.aisourceoftruth.com/.well-known/oauth-authorization-server - Protected resource metadata: https://rooms.aisourceoftruth.com/.well-known/oauth-protected-resource - Token endpoint: https://rooms.aisourceoftruth.com/oauth/token - Device sign-in: at PromptReports ID, https://accounts.promptreports.ai/oauth/device_authorization - Authorization endpoint: https://rooms.aisourceoftruth.com/oauth/authorize - JWKS: https://rooms.aisourceoftruth.com/.well-known/jwks.json An agent is an account a person creates at https://rooms.aisourceoftruth.com/app/agents (client id and secret). Four grants, all at the token endpoint: 1. client_credentials: the agent acts on its own case grants. 2. urn:ietf:params:oauth:grant-type:device_code (RFC 8628): headless, now at PromptReports ID. Start at https://accounts.promptreports.ai/oauth/device_authorization; a person approves the user code at https://accounts.promptreports.ai/device; the ID agent token it gives is sent here with DPoP (below). Rooms starts no new device codes. 3. authorization_code with PKCE S256 (RFC 7636): in a browser; a person approves at https://rooms.aisourceoftruth.com/oauth/authorize. 4. urn:ietf:params:oauth:grant-type:token-exchange (RFC 8693): a person's token plus the agent's token gives one token where the agent acts for the person, optionally limited to one case (case_id). Send the access token as "Authorization: Bearer ". The access log records the agent and the person it acts for. Scopes: cases:read, cases:write, export, billing:read, billing:write PromptReports ID agent tokens are accepted too: a token the ID mints for https://rooms.aisourceoftruth.com with scope aisot-rooms:read or aisot-rooms:write, sent as "Authorization: DPoP " with a DPoP proof (RFC 9449) on each call. The token's org must be linked (an owner or admin signs in to Rooms with PromptReports ID once) and the person must be a member there. A refusal names a shared code and the next step. Family contract: https://rooms.aisourceoftruth.com/.well-known/promptreports-connector.json and https://rooms.aisourceoftruth.com/.well-known/promptreports-inventory.json ## Main routes - GET /api/me who the token is - GET /api/cases cases you can see - POST /api/orgs/{org}/cases {title} a new case - GET /api/cases/{id} folders, items, every version - POST /api/cases/{id}/folders {name, parent_id?} - POST /api/signing-keys {public_jwk, proof} an EC P-256 or OKP Ed25519 upload key; proof is that key's signature over "rooms-key-v1\n" - POST /api/cases/{id}/uploads file bytes; headers x-rooms-file-name, x-rooms-sha256, x-rooms-key-id, x-rooms-signature, optional x-rooms-amends and x-rooms-reason - POST /api/cases/{id}/versions/{ver}/renderings/{page} a PNG page image, x-rooms-sha256 - GET /api/cases/{id}/versions/{ver}/file the original (logged) - POST /api/cases/{id}/invites {email, role, days?} - POST /api/cases/{id}/grants {principal, role, days?} - GET /api/cases/{id}/log the access log - GET /api/cases/{id}/log/verify re-check the whole chain - GET /api/cases/{id}/log/{seq}/proof an inclusion proof - POST /api/cases/{id}/exports a court bundle or tip pack (tar) The upload statement an uploader signs is these lines joined by a newline: "rooms-upload-v1", case id, sha256 (hex), size in bytes, media type, file name, the version it amends (or empty). ## Consent by API An invited agent (or a person) sees nothing on a case until it accepts the case's current privacy notice and terms. A call that needs it answers 403 with code consent_required. Then: - GET /api/cases/{id}/consent the texts and their ids - POST /api/cases/{id}/consent {accept: [ids]} logged with the agent and the person it acts for ## Paying Over a plan limit, a call answers 402 payment_required with a machine-readable price sheet (plans, prices, routes to pay). Reading, verifying and exporting existing evidence never need payment. ## MCP A Model Context Protocol server runs at https://rooms.aisourceoftruth.com/mcp with the same tokens and scopes. ## Errors Every error is JSON { code, message, remediation, retryable, docs_url }. The codes: https://rooms.aisourceoftruth.com/docs/errors (also as markdown: https://rooms.aisourceoftruth.com/docs/errors?format=md).