Go · 48+ providers · OpenAI-compatible

One endpoint.
All models.

High-performance AI gateway unifying 48+ LLM providers behind a single OpenAI-compatible API. Format translation, model combos, token compression, and a built-in dashboard.

View Source Download
OpenAI
Anthropic
Gemini
Groq
DeepSeek
Mistral
xAI
Together
Fireworks
OpenRouter
Ollama
+37
/v1/chat/completions
Model Combos

Three routing strategies.

Define how requests flow through providers. Automatic failover, load distribution, or parallel synthesis.

→
→
fallback

Priority Chain

Try primary. On failure, exponential backoff, then next provider. Zero downtime.

↔
↔
round_robin

Sticky Rotation

Distribute across providers. Sticky sessions per conversation. Multi-account support.

↓
fusion

Parallel Synthesis

Fan-out to multiple providers. Judge model synthesizes the best response.

Capabilities

What's inside.

Format Translation

Auto-detect OpenAI, Claude, Gemini, Vertex, Ollama formats. Pivot architecture through OpenAI canonical.

RTK Compression

20-40% token savings. Auto-detects git-diff, grep, ls, tree, log output. Fail-open on errors.

Multi-Account

Round-robin across API keys. Exponential backoff (5s → 10s → 20s → cap 5min). Per-account limits.

OAuth Refresh

Proactive refresh 5min before expiry. Reactive on 401/403. Per-account locking prevents races.

Web Dashboard

React SPA: provider/account management, model browser, combo editor, request logs, usage charts.

SSE Streaming

Full streaming with format translation. Client disconnect detection. OpenAI-compatible event stream.

API

OpenAI-compatible endpoints.

Drop-in replacement. Use existing OpenAI SDKs and tools against any provider.

POST
/v1/chat/completions
Chat completions with streaming support
GET
/v1/models
List all available models
POST
/v1/embeddings
Generate text embeddings
POST
/v1/images/generations
Image generation (DALL-E, SDXL)
POST
/v1/audio/speech
Text-to-speech synthesis
POST
/v1/audio/transcriptions
Speech-to-text (Whisper)
Use with any OpenAI SDK
python
from openai import OpenAI client = OpenAI( base_url="http://localhost:8080/v1", api_key="your-api-key-or-jwt-token", ) response = client.chat.completions.create( model="gpt-4o-mini", # or any model from 48+ providers messages=[{"role": "user", "content": "Hello!"}], ) print(response.choices[0].message.content)
RTK

Request Token Killer.

Auto-compress tool results by 20-40%. Detects content patterns and applies the best filter.

git-diff 30-50%
git-log 25-40%
git-status 20-35%
grep 30-50%
find 20-30%
ls 15-25%
tree 20-30%
dedup-log 40-60%
smart-truncate fallback
Providers

191+ models pre-configured.

OpenAI12
Anthropic6
Gemini8
Groq10
DeepSeek4
Mistral8
xAI4
Cohere6
Together15
Fireworks12
OpenRouter20+
Perplexity5
Replicate10+
Ollamalocal
NVIDIA8
HuggingFace10+
GitHub4
Cerebras3
Architecture

Hexagonal design.

Ports and adapters. Inbound HTTP, domain logic, outbound providers. Clean separation.

architecture
inbound HTTP/REST, SSE Stream, React SPA | ports ProxyService, AuthService | use cases ProxyUsecase, AdminUsecase, AuthUsecase | domain Model, Provider, Account, Combo, Error | outbound ProviderRegistry, SQLite, Translator
Stack

Built with.

Go 1.21+
chi v5
SQLite
React 18
Tailwind
Zustand
JWT
Caddy
Get Started

Three install methods.

Docker Compose

git clone https://github.com/ardiannurcahya/svclabs-router.git cd svclabs-router cp .env.example .env make compose-up # includes Caddy reverse proxy with auto-HTTPS # dashboard at http://localhost:8080

From Source

git clone https://github.com/ardiannurcahya/svclabs-router.git cd svclabs-router cp config.yaml.example config.yaml make full-build # run ./bin/router-svclabs -config config.yaml

Enable Providers

# set API keys as env vars export OPENAI_API_KEY=sk-... export ANTHROPIC_API_KEY=sk-ant-... export GOOGLE_API_KEY=AI... # or add via admin API curl -X POST /admin/accounts \ -d '{"provider":"openai",...}'