Every frontier model,through one endpoint.
One OpenAI- and Anthropic-compatible endpoint for 50+ models. Streaming, tools, BYOK, and semantic cache — zero markup.
Every provider behind
a single integration.
33 models, 8 providers, one OpenAI- and Anthropic-compatible endpoint. Switch providers with a string change.
claude-opus-5
Anthropic
Frontier · Tools · 1M context
gpt-6-astra
OpenAI
Frontier · Tools · 1M context
grok-4.6
xAI
Reasoning · Tools · 500k
kimi-k3
Moonshot
Open weights · Tools · 1M
glm-5.3
Zhipu
Open weights · Tools · 1M
deepseek-v4-pro
DeepSeek
Reasoning · Tools · 1M
claude-opus-5
Anthropic
Frontier · Tools · 1M context
gpt-6-astra
OpenAI
Frontier · Tools · 1M context
grok-4.6
xAI
Reasoning · Tools · 500k
kimi-k3
Moonshot
Open weights · Tools · 1M
glm-5.3
Zhipu
Open weights · Tools · 1M
deepseek-v4-pro
DeepSeek
Reasoning · Tools · 1M
claude-sonnet-5
Anthropic
Balanced · Tools · 1M context
gpt-5.6-terra
OpenAI
Balanced · Tools · 1M context
grok-4.5
xAI
Reasoning · Tools · 500k
kimi-k2.7-code
Moonshot
Code · Open weights · 256k
minimax-m3
MiniMax
Open weights · Tools · 1M
glm-5.3-flash-derisked
Zhipu
Derisked · Hosted · 1M
claude-sonnet-5
Anthropic
Balanced · Tools · 1M context
gpt-5.6-terra
OpenAI
Balanced · Tools · 1M context
grok-4.5
xAI
Reasoning · Tools · 500k
kimi-k2.7-code
Moonshot
Code · Open weights · 256k
minimax-m3
MiniMax
Open weights · Tools · 1M
glm-5.3-flash-derisked
Zhipu
Derisked · Hosted · 1M
claude-haiku-4.5
Anthropic
Fast · Tools · 200k
gpt-5.4-mini
OpenAI
Fast · Tools · 400k
gpt-5.3-codex
OpenAI
Code · Tools · 400k
deepseek-v4-flash
DeepSeek
Fast · Open weights · 1M
doubao-seed-2.1-turbo
ByteDance
Fast · Tools · 256k
claude-fable-5.1
Anthropic
Frontier · Tools · 1M context
claude-haiku-4.5
Anthropic
Fast · Tools · 200k
gpt-5.4-mini
OpenAI
Fast · Tools · 400k
gpt-5.3-codex
OpenAI
Code · Tools · 400k
deepseek-v4-flash
DeepSeek
Fast · Open weights · 1M
doubao-seed-2.1-turbo
ByteDance
Fast · Tools · 256k
claude-fable-5.1
Anthropic
Frontier · Tools · 1M context
Dedicated GPUs,
private by default.
Single-tenant pod
One GPU per customer — never shared.
Bring your model
HuggingFace id or Docker image, your token encrypted at rest.
Auto-stop when idle
Spins down after your idle window; restarts automatically on the next request.
Enterprise controls,
built in.
Spend caps, BYOK, ZDR, semantic cache, and audit logs — working out of the box.
OpenAI & Anthropic SDK compatible
Change one line. Streaming, tools, JSON mode, and multi-modal all keep working.
▍
Semantic cache
Serves cached responses for near-duplicate prompts above your similarity threshold.
threshold
0.95
cosine
Per-key opt-in.
ZDR keys never read or write.
Bring your own key
Bill to your own provider account. Encrypted at rest, decrypted at the edge per request.
$ payload (AES-256-GCM)
3a18f6d4b2907e5c3a18f6d4b2907e5c3a18f6d4b2907e5c3a18f6d4b2907e5c
$ decrypted at edge
sk-proj-xxxx••••••••
Per-key & org-level hard caps
SpendTightest cap wins. Alerts at 80 / 90 / 100% via in-app, email, or signed webhook.
Zero Data Retention
ZDRPer-key toggle. Prompts & responses never logged. Caches skip. Billing metadata only.
Usage & audit
LogsDaily spend, p50/p95/p99, per-key cache hit rate, CSV export, append-only audit log.
Routing presets & signed webhooks
Named fallback chains via one header. HMAC-SHA256 webhooks on every inference event.
Header
X-Ahura-Preset: prod-cheap-first
Webhook signature
sha256=3a18f6…
Frequently asked questions
Everything you need to know about our platform and services.
50+ models across 12 providers — OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, Qwen, Cohere, and more. The catalog refreshes as new models ship; your code doesn't change.
Yes. Point the OpenAI SDK at our base URL and it works unchanged. An Anthropic-compatible /v1/messages endpoint is also available for existing Anthropic SDK code.
Pass-through pricing at each model's published rate — zero markup. Metered per request against your credit balance. With BYOK, the upstream call bills to your account instead. Off-peak discounts apply automatically.
Per-key rate limits (1–10,000 RPM) and per-key plus org-level monthly hard caps — the tightest wins, returning 402 when reached. Alerts at 80 / 90 / 100%. All configurable in the dashboard.
Supported on every model that supports them upstream. The catalog exposes per-model capability flags (streaming, tools, JSON mode, vision, context window) so you can select programmatically.
Enable Zero Data Retention per key and prompts and completions are never logged — only billing metadata (tokens, latency, model, status). ZDR keys bypass the semantic cache.
Two cache layers. Exact-match returns free cached responses (5-min TTL). Semantic cache (opt-in per key) matches near-duplicates within a tunable cosine threshold. Hit rate is visible per key and per org.
An edge gateway across 200+ regions, TLS 1.3 terminated at the edge. With BYOK, the provider key is held encrypted (AES-256-GCM) and decrypted per request at the edge — plaintext never crosses the control-plane boundary.