FAQ
Answers to the questions we hear most from developers evaluating and running on LLM API. Jump to a section below, or use search (⌘K).
Getting started
What is LLM API?
LLM API is a gateway, not a model. Your app talks to a single OpenAI/Anthropic-compatible endpoint with one key, and LLM API forwards each request to the real upstream provider (OpenAI, Anthropic, AWS Bedrock, Google, and others). One key + one URL = access to every model from every vendor. See How it works.
How do I switch from calling OpenAI or Anthropic directly?
Point your existing client's base URL at https://api.llmapi.ai/v1 and use your LLM API key instead of the provider key. The request format is unchanged — usually a one-line edit. See the Quickstart.
Does my existing OpenAI SDK work?
Yes. LLM API speaks the OpenAI API format, so any OpenAI SDK works unchanged — just set base_url/baseURL and the API key. Tools that speak Anthropic's native protocol (like Claude Code) point at the base URL without /v1. See SDKs and Configure your AI tools.
Do I need to change my prompts or application code?
No. LLM API uses the same Chat Completions request/response shape you already use — only the base URL and key change. Your prompts, streaming, and tool-calling code carry over as-is.
How is LLM API different from OpenRouter?
Compared with OpenRouter, LLM API offers:
- Enhanced analytics with deeper insight into model usage and performance
- No fees when using your own provider keys, maximizing cost efficiency
- Greater flexibility and customization for enterprise deployments
Models & providers
Which models and providers are supported?
Hundreds of models across Anthropic, OpenAI/Azure, Google, xAI, DeepSeek, Meta, Qwen, Mistral, and many more — including audio and video providers. Browse the full catalog on the Models page, or fetch the live list your key can use via GET /v1/models.
Can I bring my own provider keys (BYOK)?
Yes. Attach your own provider credentials and LLM API routes through them — inference is billed directly by your provider, with no LLM API fee. Only optional add-ons like full data retention bill to LLM API. See Custom Providers.
How do I add a custom, self-hosted, or fine-tuned model?
Add an OpenAI-compatible provider (a lowercase name, base URL, and token) in your organization's provider settings, then call it as {customName}/{modelName}. This works for private deployments, internal services, and specialized providers. See Custom Providers.
Can I route to one model and fall back to another?
Yes. Provider Fallback lets you attach a per-request chain of alternate (provider, model) targets — if the primary fails with an availability-class error, LLM API transparently retries the next target (e.g. Anthropic → Bedrock for Claude), with no duplicate billing on success.
How does LLM API pick a provider for a model?
When a model is available from more than one upstream provider, LLM API's smart routing selects the best available provider for that model. You can also pin a specific provider when you need to.
Pricing & billing
How does pricing work?
Two tracks, and you can hold keys on both: AI for Coding (a flat monthly subscription with a refreshing credit budget) and AI for Production (pay-as-you-go with volume discounts that grow from 5% up to 50% at enterprise scale).
Is there a fee for using my own keys?
No. When you route through your own provider keys, inference is billed directly by the provider with no LLM API markup. The only LLM-API-side charge is optional storage if you turn on full data retention. See Data Retention.
What
Most organizations run on prepaid credits — top up a balance and usage draws it down. Pay-as-you-go (PAYG) instead invoices you once per billing cycle for what you used, with no balance to maintain. PAYG is enabled per organization by the LLM API team. See Pay-as-you-go Billing.
Is there a starting bonus for new organizations?
Yes. New orgs get a one-time deposit bonus on their first top-up (up to $1,000, matched at +100%), and coding subscriptions get a one-time 50% off the first month (monthly billing).
How much does storing full request data cost?
Metadata-only retention (timestamps, model, tokens, cost) is the free default. Storing complete request/response payloads is opt-in at $0.01 per 1M tokens. See Data Retention.
Reliability & performance
Does the gateway add latency?
LLM API adds a single lightweight proxy hop — it authenticates the request, picks a route, and streams the response straight through without extra buffering or additional model calls. We don't publish a fixed overhead number because it depends on your network path and the provider's own latency, but every request's real latency is broken out in your dashboard so you can measure it for your own traffic. See Usage & analytics.
What happens if a provider goes down?
Turn on Provider Fallback and LLM API fails a request over to the next (provider, model) in your chain when the primary returns an availability-class error (like a 429 or an outage) — within a deadline you define, and without a client-side retry loop.
Does LLM API cache responses?
Yes. Caching serves identical requests from cache instead of re-calling the provider, cutting both cost and latency for repetitive workloads.
Is there an uptime SLA?
A specific uptime SLA isn't published on this page. If your deployment needs a contractual availability guarantee, contact support about enterprise terms — and use Provider Fallback to add cross-provider resilience to individual requests today.
Security, privacy & data
Do you store my prompts and responses?
Not by default. LLM API stores only request metadata (timestamps, model, tokens, cost) unless you explicitly opt in to full payload retention. See Data Retention.
Can I run with zero data retention?
Yes — metadata-only is the default, so your prompts and completions are not stored unless you turn on "Retain All Data" per organization. See Data Retention.
Do you use my data to train models?
No. LLM API does not use your prompts or completions to train models. Requests are forwarded to the upstream provider you target, so that provider's own data policies apply to the forwarded call.
How is stored data secured?
When retention is enabled, stored data is encrypted at rest, access is restricted to organization members with the right permissions, and data is automatically deleted after your configured retention period. You can also request immediate deletion of specific records through support. See Data Retention.
Teams & access control
Can I issue a separate key per team member or project?
Yes. Create unlimited API keys, and use one-click team sync to invite your whole team and issue a unique key to each member. See API Keys & IAM Rules.
Can I set spend limits per key?
Yes. Each API key can carry its own usage/spend limit, usage is tracked per key, and keys can be disabled and re-enabled without deleting them. See API Keys & IAM Rules.
Can I restrict which models or providers a key can use?
Yes. IAM rules give fine-grained access control over which models, providers, and pricing a key may use. See API Keys & IAM Rules.
Do you support self-hosting?
Yes. Self-hosted deployments give you full control over data retention and store data in your own PostgreSQL database, with no additional storage charges. See Data Retention → Self-Hosted Deployments.
Do you support SSO, SAML, or advanced RBAC?
Access is scoped to organization members with permissions today. For enterprise identity (SSO/SAML/SCIM) and advanced role-based access control, contact support to discuss your requirements.
Observability & analytics
How do I track usage and cost?
The dashboard shows requests, tokens, total spend, and average cost per 1K tokens across 7- or 30-day windows. See Usage & analytics.
Can I break down spend by model and provider?
Yes. Usage and spend break down per provider and per model, so you can quickly spot expensive outliers. See Usage & analytics.
Can I inspect individual requests?
With full data retention enabled, you can view request history and inspect complete request/response payloads, filtered by model and date range. See Data Retention.
What metrics are available?
Request volume, token usage, total and per-model cost, plus error rate, cache-hit rate, and reliability trends — all in the dashboard.
Advanced capabilities
Is streaming supported?
Yes. Pass stream: true on any request and LLM API proxies the provider's event stream (SSE) straight through. See the Quickstart.
Does tool / function calling work?
Yes, on models that support it — using the standard OpenAI tool-calling schema. Filter the Models page by Tools to see which models qualify.
Are multimodal models and embeddings supported?
Yes — vision, audio (TTS & STT), image and video generation, and embeddings are all available. See the Features section and filter the Models page by capability.
Do you support batch processing?
Yes. See Batch Processing for submitting large asynchronous jobs.
How do rate limits work?
Rate limits depend on your account status and the models you use. See Rate limits for the details and how to request higher limits.
Still stuck? Reach out via the Support button in the header, or email support@llmapi.ai.