api-docs

name	api-docs
description	Official Prem documentation. Use when onboarding to Prem, understanding the platform architecture, managing API keys, handling errors/rate limits, billing, or looking up raw API reference endpoints. Triggers on Confidential API docs, quickstart, API keys, billing, rate limits, attestation, security model, enclave architecture, Prem platform, prem.io, prem labs, prem, prem ai.

Prem Documentation

Official docs for the Prem confidential AI platform. Load specific links below based on the task at hand.

Install official skill: npx skills add docs.prem.io --yes.

---

Basics

Topic	URL	Load when
Quickstart	https://docs.prem.io/basics/build/quickstart.md	Starting a new Prem integration from scratch — covers API key setup, env vars (PREM_API_KEY, PROXY_URL, ENCLAVE_URL), SDK install, local proxy, and first streaming chat completion
Overview	https://docs.prem.io/basics/build/overview.md	Need a navigation map of all Prem docs — chat, audio, auth, encryption, errors, recipes, architecture
Use LLMs / AI agents	https://docs.prem.io/basics/troubleshooting/use-llms.md	Loading Prem docs into an AI coding agent or editor (Claude Code, Cursor, Windsurf) via npx skills add https://docs.prem.io/skill.md, or discovering all doc pages via llms.txt
Contact us	https://docs.prem.io/basics/troubleshooting/contact-us.md	Need support — live chat via dashboard (minutes) or email [e-mail] (1 business day)

---

Developer Resources

Topic	URL	Load when
Encryption	https://docs.prem.io/developer-resources/get-started/encryption.md	Understanding zero-knowledge architecture, KEK/DEK/RAG-DEK key hierarchy, XWing post-quantum KEM, XChaCha20-Poly1305 encryption, AES-KWP key wrapping, two-server model (gateway + enclave), or file encryption flow
API Keys	https://docs.prem.io/developer-resources/get-started/api-keys.md	Creating or rotating API keys (Dashboard → Developers → API Keys), setting IP restrictions (IPv4/IPv6/CIDR), applying least-privilege scopes, or understanding org-level key ownership
Errors	https://docs.prem.io/developer-resources/get-started/errors.md	Debugging error responses — JSON body has status, error, support_id fields; 400 bad request, 401 auth, 403 missing scope, 404 not found, 429 rate limited, 500/503 server errors; save support_id for support tickets (kept 1 week)
Rate Limits	https://docs.prem.io/developer-resources/get-started/rate-limits.md	Understanding rate limit types (RPS, TPM, tokens daily quota, concurrent requests, audio APM/daily quota), tier capacities (BASE to TIER_3), token bucket algorithm, handling 429 with exponential backoff + Retry-After header

---

Billing

Topic	URL	Load when
Models & Pricing	https://docs.prem.io/developer-resources/billing/models-and-pricing.md	Comparing model costs (pay-as-you-go): chat input/output per 1M tokens, audio per minute; or choosing between Explorer (free, shared), Developer (dedicated), Enterprise (air-gapped) tiers
Balance	https://docs.prem.io/developer-resources/billing/balance.md	Understanding prepaid balance (requests blocked at zero), topping up ($1 min via Stripe), or configuring auto top-up (threshold + target amount)
Limits	https://docs.prem.io/developer-resources/billing/limits.md	Monthly spending caps per tier (Base $100, Tier 1 $500, Tier 2 $5k, Tier 3 $20k), resets on 1st of month, adjustable via Dashboard
Payment	https://docs.prem.io/developer-resources/billing/payment.md	Adding/managing payment cards (stored via Stripe), setting default payment method for manual top-ups and auto-replenishment
Usage	https://docs.prem.io/developer-resources/billing/usage.md	Inspecting last 30 days of token usage (prompt/completion/reasoning tokens + cost per day), per-model breakdown, file storage size and count

---

API Reference

API Keys

Endpoint	URL	Load when
Get API key	https://docs.prem.io/developer-resources/reference/api-keys/get-api-key.md	GET /developers/api_keys/{id} — fetch single key with scopes, ip_allow_list, expires_at, usage metrics
Get API key analytics	https://docs.prem.io/developer-resources/reference/api-keys/get-api-key-analytics.md	GET /analytics/api_keys — success rates, latency percentiles (p50/p90/p95/p99), hourly breakdown, 30-day per-key metrics, top keys by usage
List API key scopes	https://docs.prem.io/developer-resources/reference/api-keys/list-api-key-scopes.md	GET /developers/api_keys/scopes — full list of available scopes: api_key.create/read/update/delete, users.read/update, chats.completion, tools.execute, files.encrypted.create/read, audio.transcription/translation
List API keys	https://docs.prem.io/developer-resources/reference/api-keys/list-api-keys.md	GET /developers/api_keys — all org keys with metadata (scopes, IPs, expiry, request count, last used)
Get API key request	https://docs.prem.io/developer-resources/reference/api-keys/get-api-key-request.md	GET /developers/api_keys/requests/{id} — single request log entry with full request/response bodies, headers, duration (ms), memory peak
List API key requests	https://docs.prem.io/developer-resources/reference/api-keys/list-api-key-requests.md	POST /developers/api_keys/requests — paginated request history with filters: api_key_id, date range, status (success/failed), HTTP method, search string; default limit 50

Models

Endpoint	URL	Load when
List models	https://docs.prem.io/developer-resources/reference/models/list-models.md	GET /models?type=CHAT|AUDIO_TRANSCRIPTION|AUDIO_SPEECH|IMAGE — model id, type, name, description, input_modalities, enabled, price_config

Chats & Audio

Endpoint	URL	Load when
Chat completions	https://docs.prem.io/developer-resources/reference/rvenc/chat-completions.md	POST /rvenc/chat/completions — raw encrypted REST schema; payload fields: encryptedInference, cipherText, nonce; SSE streaming with encrypted chunks (XChaCha20-Poly1305); OpenAI-compatible decrypted format
Audio transcriptions	https://docs.prem.io/developer-resources/reference/rvenc/audio-transcriptions.md	POST /rvenc/audio/transcriptions — encrypted multipart: encryptedInference, cipherText, nonce, encryptedFile, fileNonce, encryptedFileName, fileNameNonce; max 25MB; supports language, response_format (json/text/verbose_json), timestamp_granularities
Audio translations	https://docs.prem.io/developer-resources/reference/rvenc/audio-translations.md	POST /rvenc/audio/translations — same encrypted structure as transcriptions; Whisper-only, always translates to English; max 25MB; non-streaming

---

Recipes

Topic	URL	Load when
Authentication	https://docs.prem.io/recipes/authentication.md	Working auth example: createRvencClient with apiKey + clientKEK, pre-generating encryption keys once via generateEncryptionKeys() for reuse across requests, custom requestTimeoutMs / maxBufferSize
Chat completions	https://docs.prem.io/recipes/rvenc-chat-completion.md	Code examples for: text chat, image analysis, document/receipt JSON extraction (response_format: json_object), product catalog from images, image Q&A, multi-image comparison, streaming vision
Audio transcription	https://docs.prem.io/recipes/audio-transcription.md	Code examples for: basic transcription (whisper-large-v3), language hint, verbose_json with word/segment timestamps, custom domain prompt for accuracy
Audio translation	https://docs.prem.io/recipes/audio-translation.md	Code examples for: basic translation to English, contextual prompt, verbose_json with detected language + duration

---

Learn More

Topic	URL	Load when
What is Prem API	https://docs.prem.io/basics/learn-more/overview.md	Explaining the platform: TEE-based confidential computing, zero-knowledge guarantee, quantum-resistant encryption, OpenAI drop-in replacement, target use cases (healthcare, finance, legal, sensitive IP)
Developer Experience	https://docs.prem.io/basics/learn-more/developer-experience.md	Understanding the DX: encryption is invisible in app code, TypeScript SDK vs local proxy for other languages, standard OpenAI-style errors and rate limits
Platform Status & Roadmap	https://docs.prem.io/basics/learn-more/platform-status.md	Current production features (streaming, tool calling, Whisper/Deepgram, AMD SEV-SNP + Intel TDX + NVIDIA GPU attestation); gap: image reproducibility not yet available; roadmap: deterministic builds, Python/Go/Rust SDKs, enclave-to-enclave chaining, verification dashboard

---

Architecture & Security

Topic	URL	Load when
How it works	https://docs.prem.io/basics/learn-more/how-it-works.md	Full request flow: SDK encrypts on device → proxy gateway (blind, sees only metadata) → sealed enclave decrypts + runs inference + re-encrypts → response decrypts on device; all supporting services (model router, LLM, vector DB, STT) run inside CVMs
Security model	https://docs.prem.io/basics/learn-more/security-model.md	What Prem cannot see (request content, keys, CVM data) vs can see (timing, payload sizes, API keys); protected against: network eavesdropping, server/cloud operator access, rogue employees, code tampering; not protected: hardware side-channels, compromised manufacturer keys, metadata
Attestation	https://docs.prem.io/basics/learn-more/attestation.md	4-step flow (challenge → TEE-signed report → browser verification → decision); hardware: AMD SEV-SNP, Intel TDX, NVIDIA Hopper/Blackwell; Rust+WASM stack (@premai/reticle); GPU session binding; what to verify: code hashes, signature chains, debug disabled, nonce present