> For the complete documentation index, see [llms.txt](https://vordai.gitbook.io/vordai-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://vordai.gitbook.io/vordai-docs/getting-started.md). # Getting Started ### Quickstart #### 1. Read the live catalogs Start every integration by reading: * GET for model IDs, capability flags, privacy metadata, context windows, limits, and prices * GET for the current exact built-in tool schemas The model catalog is a signed taxonomy-v2 bundle. Clients may verify model\_taxonomy\_version, bundle\_digest, source\_bundle\_digest, publisher\_key\_id, and publisher\_signature. Interpret privacy\_tier, privacy\_capabilities, model\_sets, and traits as separate fields. The uncensored model set applies only to models explicitly listed in model\_sets. The most\_uncensored trait is only a recommendation signal; it does not establish model-set membership or guarantee unrestricted model behavior. The current live catalog publishes empty model\_sets arrays. GET is the authoritative built-in tool directory. Use the returned exact function schemas, tool\_catalog\_version, and tool\_catalog\_digest instead of a hard-coded list. The current snapshot contains calculator, fetch\_url, current\_time, and web\_search. #### 2. Select a privacy tier Vord publishes two privacy tiers: anonymized and private. TEE attestation and E2EE are separate signed privacy capabilities, not additional tiers. The current catalog contains 28 anonymized routes and two private routes: tee-glm-5.2 and tee-kimi-k2.6. Both private routes currently declare e2ee and tee\_attestation. Verify the live route metadata and the documented evidence boundary before making a security claim. #### 3. Open a session Inference requests require an active Vord Session credential. Send it as: Authorization: Bearer sk-... Keep account cookies and account tokens out of inference requests. Use a stable Idempotency-Key when a request may be retried. #### 4. Send a chat completion POST follows the OpenAI Chat Completions format and supports streaming. Choose a model by exact live ID. Only send tool declarations when the model's capabilities array includes tools, and only enable web search when the model declares websearch. #### 5. Parse a file File parsing is available at POST . The current public contract supports PDF, EPUB, DOCX, PPTX, XLSX, and XLS documents, requires an active Vord Session, and limits each file to 25 MiB. Text and source files are read locally where possible. Original files and extracted text are not persisted in ordinary database, object-storage, log, or backup paths. Parsed content is sent to a model only when the user submits the request. #### 6. Verify the node when needed Node attestation is available at GET . The response provides Intel TDX node evidence bound to the current TLS public key. A successful verification supports only the bounded conclusion TDX NODE VERIFIED. It does not verify the workload image, source code, runtime arguments, or operator exclusion. ### Recovery behavior Direct-mode recovery can look up request and billing state, but it does not replay or reconstruct response text. A separate opt-in Response Spool path is Beta / rollout pending. When enabled, it resumes the same accepted request from short-lived encrypted frames and never creates a second provider execution. The feature gate is not generally open. ### Credits Credits purchase and the device wallet are Beta / rollout pending. Credits are blind-signed, device-held bearer credentials protected by P2PK keys. Anyone with the corresponding proof and private-key material can spend them; exporting that material transfers practical spending control. ### Coming Soon Optional device-local App Screen Lock with a 4-digit or 6-digit passcode is Coming Soon. It is a screen-privacy control and does not replace the recovery key or add a new Credits-encryption layer. Video generation APIs and media-capable routes are Coming Soon. They remain under integration and are not generally available. Other planned surfaces include a Python SDK, additional providers, and more tools. ### Availability labels Available means the live service or catalog exposes the capability. Beta / rollout pending means implementation or a public contract exists but general availability has not been confirmed. Coming Soon means an approved product or implementation plan exists, but the capability is not yet available.