Your keys.
Your models.
One endpoint.
GenAff is an OpenAI-compatible gateway over the provider accounts you already pay for. Connect OpenAI, Anthropic, or DeepSeek once, then point your entire stack at one unified shape with zero markup.
API shape
OpenAI-compatible
Credentials
Yours, encrypted
Markup
0% · Direct cost
Zero markup routing: Requests stream directly to your upstream provider accounts.
Universal Drop-in
One line to switch. Zero SDK lock-in.
Use standard OpenAI client packages in Python, Node.js, Go, or cURL. Simply point the baseURL to GenAff.
curl https://api.genaff.cloud/v1/chat/completions \
-H "Authorization: Bearer $GENAFF_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "anthropic:claude-3-5-sonnet",
"messages": [{"role": "user", "content": "Explain vector indexing"}],
"stream": false
}'Two planes, deliberately separated.
Your dashboard session and your gateway key are different credentials. The web app never handles a gk_live_ key, and your servers never receive your dashboard token.
Dashboard Plane: Uses short-lived, memory-stored JWTs with Web Locks preventing token replay across tabs.
Gateway Plane: Uses secret-isolated gk_live_ bearer tokens with fast, stateless authentication.
Dashboard plane
Short-lived access token, provider connections, GenAff keys and usage metrics.
Authentication
Bearer <session JWT>
Gateway plane
Model discovery and chat completions from your own backend application or SDK.
Authentication
Bearer gk_live_…
No silent fallback, no shared keys.
A failed model does not route to another provider. Provider credentials are encrypted at rest with AES-256 and are never returned by the API.
The sub-millisecond request lifecycle
How POST /v1/chat/completions authenticates, resolves models, vaults keys, and relays to upstream AI providers without touching disk.
Key Authentication
Fast SHA-256 lookup verifies gk_live_ scopes and owner identity against PostgreSQL.
Sub-millisecond auth
Model Resolution
Parses provider:model, verifies the caller holds an active, healthy provider connection.
Deterministic routing
In-Memory Decryption
Decrypts upstream provider key exclusively in memory. Never logged, never written to disk, never echoed in errors.
Zero credential leak
Upstream Dispatch
Transforms payload to native provider format and calls hardcoded, verified upstream endpoints over secure HTTPS.
Native provider adapters
Metadata Recording
Normalizes response to OpenAI format. Records latency, status, and tokens. Zero prompts or completions are stored.
Zero prompt storage
Modular Monolith Architecture
Engineered in Python with FastAPI & PostgreSQL. Eliminates microservice network hops, giving predictable sub-millisecond gateway latency.
SSRF Hardening
GenAff communicates solely with strictly code-configured, verified provider endpoints. Arbitrary caller-supplied base URLs are strictly forbidden.
Deterministic Error Mapping
Upstream errors are mapped into standard HTTP statuses (402 Quota Exceeded, 429 Rate Limit, 502 Upstream Error) with request ID tracing.
Ecosystem
Connected providers
Read live from this deployment. Only providers with an adapter and passing contract tests appear here.
Anthropic
anthropic
Capabilities
- model discovery
- chat
DeepSeek
deepseek
Capabilities
- model discovery
- chat
- reasoning effort
- extra body thinking
OpenAI
openai
Capabilities
- model discovery
- chat
Coming soon to GenAff
Native adapters and contract tests are actively in development for the next wave of frontier and ultra-low latency inference providers.
Google Gemini
gemini
Multimodal 2.0 Flash and Pro models with 1M+ token context windows.
- chat
- multimodal
- 1m context
- function calling
Mistral AI
mistral
Mistral Large 2, Codestral, and fast Mixtral Mixture-of-Experts reasoning.
- chat
- code generation
- moe
- json mode
Groq
groq
Ultra-low latency LPU inference engines powering Llama 3.3 at 500+ tok/s.
- chat
- sub 100ms
- lpu speed
- open weights
Together AI
together
High-throughput open source clusters, DeepSeek R1 distillation, and fine-tunes.
- chat
- open weights
- turbo endpoints
- embeddings
Designed for AI Builders
The gateway without the middleman tax
Most gateways charge 10% to 30% inference markup and proxy your tokens through proprietary servers. GenAff operates on a pure BYOK model.
Zero Markup, Always
Inference is billed directly to your OpenAI, Anthropic, or DeepSeek credit cards. GenAff never marks up tokens, charges per-call fees, or introduces credit traps.
Unified Model Catalog
Call /v1/models to discover all active upstream models across all connected providers formatted in the official OpenAI schema.
End-to-End Key Security
Provider API keys are encrypted at rest with AES-256 and never returned. Gateway keys are displayed once upon creation and stored as salted hashes.
Sub-Millisecond Overhead
Ultra-lightweight proxy pipeline normalizes request bodies and headers with less than 1ms of routing latency before streaming directly upstream.
Metadata-Only Logging
We never store, log, or inspect prompts or completions. Only request metadata (HTTP status, latency, token count, model ID) is recorded for analytics.
Deterministic Errors
Upstream rate-limits and quotas (like 402 Upstream Quota Exceeded) are surfaced with precise error codes and copyable request IDs for rapid debugging.
What Phase 1 does not do.
We would rather name the boundary upfront than ship a stub that pretends to work.
- No managed inference. Your provider bills you directly.
- No credits, subscriptions or mock purchases.
- No streaming yet. Chat is request/response with
stream: false. - No password reset. Contact support for account recovery.
- One connection per provider. Replace the credential instead of adding another.
export GENAFF_API_KEY="gk_live_YOUR_KEY"
curl https://api.genaff.cloud/v1/models \
-H "Authorization: Bearer $GENAFF_API_KEY"Replace the placeholder with a key you create in the dashboard. The web app never stores or proxies this key.
Start routing through GenAff in seconds
Create an account, plug in your provider API keys, and test your first unified request. Completely free for bring-your-own-keys.