Go Feather RouteRoute models. Keep the footprint small.

A focused OpenAI-compatible gateway for model APIs, private provider boundaries, and streaming workloads.

Start routing →

View on GitHub

Docker Hub

Benchmark details

⚡

Fast startup

A static Go binary with a narrow dependency surface and no database, queue, or dashboard required.

◌

Small footprint

Bounded bodies, concurrency, timeouts, and memory-conscious container packaging for small hosts.

≋

Streaming first

Forward Server-Sent Events promptly and cancel upstream work when clients disconnect.

⇄

OpenAI compatible

Keep existing clients while routing model aliases to OpenAI-compatible provider endpoints.

▣

Provider isolation

Provider credentials stay server-side behind one private gateway boundary.

◈

Thingd boundary

Planned data-aware capabilities through Thingd MCP without embedding Thingd in the router.

⌁

Safe diagnostics

Optional loopback-only runtime profiles for investigating streams, cancellations, and goroutine leaks.

Reference performance: the benchmark harness reports latency, throughput, CPU, memory, I/O, process count, cgroup peaks, and OOM state where the host exposes those measurements. See the benchmark guide for the measurement environment and interpretation rules.

The first Thingd Cloud integration canary returned 10/10 successful chat requests through both Go Feather Route and LiteLLM, with lower Go Feather Route p95 latency and streaming time-to-first-byte in that sample. Embeddings and comparable resource usage remain qualification work; see the full canary results.

Gateway idle RSS

~4.5 MiB

Reference container measurement

Proxy p50

0.64 ms

16 requests / concurrency 4

Gateway image

8.7 MB

Static arm64 image

Streaming

SSE

Forwarded without full buffering

How a request moves

Choose a starting path

Start routing

Configure one provider and send an OpenAI-compatible request.

Deploy with Docker

Use the non-root multi-architecture image with runtime-injected secrets.

Measure the gateway

Compare routing overhead and resource use with the pinned LiteLLM reference image.

Inspect runtime health

Check readiness, metrics, and opt-in loopback diagnostics for stalled streams.

Plan a Thingd boundary

Keep the router standalone while defining a separate MCP capability boundary.

Long-term direction

The project is designed to remain a small operational boundary as its capabilities grow: more compatible providers, multimodal requests, per-tenant quotas, usage metrics, health-aware routing, graceful degradation, memory-aware deployment profiles, and a planned Thingd MCP data boundary.

Start in one command

docker run --rm -p 4000:4000 \
  -e GOFEATHERROUTE_API_KEY=gateway-key \
  -e OPENAI_API_KEY=your-key \
  sayanmohsin/go-feather-route:0.1.0

The router owns authentication, limits, provider selection, streaming, and operational boundaries. Provider credentials are injected at runtime. Read the security model, API reference, and deployment guide before placing it behind a public endpoint.