Guides
Guides
Direct answerFind reviewed guidance by task, from first requests and model selection to production troubleshooting.
Updated
AI API production reliability: timeouts, retries, limits, and SLOsDesign end-to-end deadlines, bounded retries, concurrency protection, circuit breaking, degradation, observability, and load acceptance for AI APIs.AI API quickstart: from API key to first requestSet the API origin to https://api.openfox.vip and choose a protocol below. OpenAI SDKs normally use a base URL ending in /v1; direct HTTP examples combine the origin with the full endpoint. Avoid duplicated /v1 segments.AI API security, privacy, and data governanceCreate verifiable boundaries for API keys, input data, file URLs, prompt injection, tool authorization, logs, and upstream retention.AI gateway admin runbook: channel rollout, retries, and supplier healthAdministrators maintain channels, models, groups and billing through the existing dashboard. Check public visibility and group permissions, then validate changes in an isolated environment. Tuzi channel diagnostics, supplier consoles and administrative APIs are not added by this migration.AI model API platform beginner guide: from signup to calls and billing checksSign in, select an exact model ID and group in the marketplace, check your balance and create a model key on the Keys page. Run one minimal request and inspect usage logs. Account management uses a personal access token, not a model key.API endpoint and capability matrixChoose endpoints by operation, not by guessing from model names. Text may use Chat, Responses, Messages or Gemini; images use generation or editing; asynchronous results use task queries. The list below points to reviewed examples.Billing, usage, and cost reconciliationUnderstand model prices, groups, reservations, and settlement; distinguish online, Fast, batch, audio cache, and storage metering to reconcile request cost.Choosing and calling coding modelsFilter coding models by repository size, tool use, reasoning depth, and latency, then validate on real tasks.Common AI API error codes and troubleshootingFor 401/403 check credential type, scopes, model permissions and IP rules. For 429 check quotas and rate limits. Preserve a non-secret request ID for 5xx errors. HTTP 200 may still carry a business failure. A timeout does not prove creation failed; inspect records before retrying.Doubao billing: online, Fast, batch, and cacheOpenFox billing follows configured model prices, group ratios and billing expressions. Tuzi Doubao allowances, discounts and Live tiers must not be imported as OpenFox pricing. Existing prices stay unchanged; unknown prices are not guessed.File and media input, output, and lifecycleImage edits upload local files; validate size, format and access rights on the server. Download external artifacts with a separate request without platform credentials or cookies, and do not follow untrusted redirects. Tuzi media-library endpoints do not apply here.General text, chat, and reasoning model guideChoose general text models by reasoning depth, context, structured output, tools, latency, and cost.Image generation and vision model guideUse multimodal chat for image understanding and dedicated Images endpoints for generation and editing.Media and object-storage admin runbook: profiles, access, and lifecycleMedia retention, download availability and storage permissions depend on task plugins and providers. Do not promise unverified retention periods. Record task identifiers, expiration behavior, cleanup policies and recovery procedures.Model evaluation, canarying, and version migrationBuild a representative evaluation set and pinned baseline, then compare offline, canary, roll back, and monitor before a model upgrade or retirement.Model originators, hosted clouds, aggregators, and local runtimesSeparate model knowledge from the route used to call it across originators, hosted clouds, aggregators, workflows, local runtimes, and account bridges.RAG, embedding, and reranking model guideEvaluate chunking, embedding, retrieval, reranking, and evidence-grounded generation separately.SSE streaming and WebSocket realtime connectionsHandle SSE events, in-stream errors, cancellation, proxy buffering, backpressure, and realtime WebSocket sessions from first byte to confirmed completion.Speech, transcription, and realtime audio guideSpeech synthesis, transcription, and realtime conversation use distinct endpoints and transports.Tool use and AI agent model guideAgents should prioritize tool schemas, long-horizon state, failure recovery, and least privilege.Unified async tasks: submit, poll, and retain resultsOpenFox uses task plugins and unified task resources. Save the real task_id from creation, query /v1/tasks/{task_id}, and retrieve artifacts after completion. Tuzi generic /async/* and /get-async wrappers are not available here.Video generation and asynchronous API guideVideo generation is usually asynchronous: persist task IDs, limit polling, and validate result media.