The unified inference layer
One API. 500+ models. Credits that last a year.
RootInference is one OpenAI-compatible API for 500+ models from 60+ providers. Smart routing picks the best model per request, automatic failover keeps you up when providers go down — and every credit you buy stays valid a full 12 months.
base_url="https://api.rootinference.com/v1"
| Model | Provider | Latency | Credits | Status |
|---|---|---|---|---|
| anthropic/claude-sonnet-4-5 | provider-virginia | 412 ms | 0.8 cr | 200 OK |
| openai/gpt-5-mini | provider-oregon | 238 ms | 0.2 cr | 200 OK |
| meta/llama-4-70b | provider-frankfurt | 301 ms | 0.3 cr | 200 OK |
| mistral/large-2 | provider-berlin | 887 ms | 0.5 cr | 200 OK |
| deepseek/v4 | provider-singapore | 356 ms | 0.2 cr | 200 OK |
| google/gemini-3-pro | provider-iowa | 428 ms | 0.7 cr | 200 OK |
| qwen/qwen-3-coder | provider-tokyo | 264 ms | 0.3 cr | 200 OK |
Routing production inference for teams shipping AI products
Platform
Every model, every provider, one inference layer
Integrating a model provider directly means adopting its SDK, its rate limits, its outages and its bill. RootInference sits between your app and 60+ providers, so routing, failover and spend control become platform features — not code you maintain.
Connect
One OpenAI-compatible API for 500+ models from 60+ providers. Change the base URL, keep your SDK — streaming, tool calling, structured outputs and vision included.
Route
Routing policies pick the best model and provider per request by price, latency and quality — with per-request overrides when you need control.
Failover
A continuously health-checked provider pool reroutes traffic automatically the moment a provider degrades. Better uptime than any single vendor.
Optimize
Unified analytics attribute every credit to a key, team and feature — and credits never rush you: every one stays valid a full 12 months.
Products
One platform. Four capabilities your inference stack is missing.
Unified API
Unified API
Change the base URL, keep your SDK. RootInference exposes a single OpenAI-compatible API that speaks to 500+ models from 60+ providers — streaming, tool calling, structured outputs and vision included. No provider-specific clients, no per-vendor billing, no rewrites.
One API for every model
Smart Routing
Smart Routing
Define routing policies once and RootInference picks the best model and provider per request. Optimise for cost, latency or quality, set fallback chains, pin rules per API key, and A/B new models against production traffic — all without touching application code.
Route every request optimally
Reliability
Reliability Engine
Every provider has bad days — rate limits, degraded latency, outages. The Reliability Engine health-checks a pool of providers continuously and reroutes traffic automatically the moment one degrades, so your app stays up when any single vendor goes down.
Failover before your users notice
Observability
Usage Observability
Unified request logs, latency and cost dashboards, and spend attribution by key, team and feature — across every provider you use. Set budget alerts, export the ledger, and answer 'what did we spend on inference and why' in seconds instead of spreadsheet archaeology.
See exactly where inference spend goes
Live gateway
Watch a request get routed, and a provider get benched
Every request is routed by live price, latency and health telemetry. Every token, millisecond and credit is logged and attributed — across all providers, in one place.
curl https://api.rootinference.com/v1/chat/completionsRequest received for anthropic/claude-sonnet-4-5, streaming on
Resolved via unified schema — no client changes required
Tool call emitted: get_weather, city Berlin
First token in 212 ms at 96 tok/s
Usage metered: 1,412 in / 388 out, priced at 0.9 credits
Same request shape works on 500+ models
rootinference status --watchProvider pool checked: 9 healthy, 1 degraded
provider-3 degraded: error rate 7.2%, p95 latency 4.8 s
provider-3 removed from rotation
Traffic rerouted to provider-1 / provider-5 in 340 ms
Client impact: 0 failed requests
provider-3 re-admitted after 12 consecutive healthy checks
Integrations
The models and tools you already build with
Frontier and open models from OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek and more — plus first-class support for the OpenAI SDK, Vercel AI SDK, LangChain and your observability stack.
Security
Your prompts are your business — literally
RootInference never trains on your data, gives you per-provider data-policy controls, and is built GDPR-aligned with SOC 2 on the compliance roadmap.
Credit validity
Credits that don't quietly expire
Prepaid AI balances are the industry's favourite breakage machine. RootInference credits — 1 credit ≈ $0.01 of inference, redeemable against OpenAI, Anthropic, Google, Meta and every other model on the platform — stay valid a full 12 months.
12-month validity
RootInference
Every purchased credit valid 12 months. Subscription credits roll over 12 months. FIFO by expiry, tracked in your dashboard.
30–90 day expiry
Typical gateways
Balances expire in 30–90 days — or the expiry clause is buried in the terms and you find out from a zeroed balance.
Pricing
Simple credit pricing across every model
1 credit ≈ $0.01 of inference. Route economy models for pennies, escalate to frontier models when the request deserves it — one balance covers all of it, valid for 12 months.
| Inference workload | Credits |
|---|---|
| 1M tokens, economy models (Llama / Mistral Small class) | 5–15 |
| 1M tokens, mid-tier models (mini / Haiku class) | 15–60 |
| 1M tokens, frontier models (GPT-5 / Opus class) | 150–900 |
| Typical 10-turn chat conversation | <1 |
| RAG query with 8K context | 1–3 |
| Multi-step agent run | 5–50 |
| Batch-summarize 1,000 documents | 50–200 |
| Dedicated throughput | Custom |
Starter
2,000 credits / month
$19/mo
Pro
12,000 credits / month
$99/mo
Scale
70,000 credits / month
$499/mo
Use cases
Built for teams running LLMs in production
Ship on frontier models without lock-in — switch providers with a string change.
One gateway, one bill and per-team budgets across every provider.
Fallback chains keep long-running agent workflows alive through provider errors.
Run million-token context workloads on economical models — 5–15 credits per 1M tokens.
Latency-first routing keeps time-to-first-token low, every message.
Compare 500+ models on your own traffic through one interface.
Change the base URL. Keep your SDK.
Point your existing OpenAI client at RootInference and get 500+ models, smart routing and automatic failover — in the time it takes to edit one line.