cli: answer --version and --help instead of starting the server

README.md

@@ -42,6 +42,7 @@ curl -sN localhost:8080/v1/messages \
 
 # Your own config
 GW_CONFIG=conf/gateway.yaml cargo run -p gw-server
+gw --version                       # the built binary takes no other arguments
 
 # Go live: give an account `endpoint` + `api_key_env` in the config — that's it.
 # GW_TRANSPORT=mock forces zero egress; GW_TRANSPORT=http disables the mock.

crates/server/src/main.rs

@@ -22,12 +22,26 @@ use tracing_subscriber::{EnvFilter, Layer as _};
 const BATCH_STALE_SECS: i64 = 120;
 const BATCH_POLL: Duration = Duration::from_secs(2);
 const CONFIG_FEED_RETRY: Duration = Duration::from_secs(5);
+const USAGE: &str = "usage: gw [--version | --help]\nconfiguration comes from the environment: GW_CONFIG, GW_TRANSPORT, GW_PORT";
 
 #[global_allocator]
 static GLOBAL: tikv_jemallocator::Jemalloc = tikv_jemallocator::Jemalloc;
 
 #[tokio::main]
 async fn main() -> anyhow::Result<()> {
+    if let Some(flag) = env::args().nth(1) {
+        return match flag.as_str() {
+            "--version" | "-V" => {
+                println!("gw {}", env!("CARGO_PKG_VERSION"));
+                Ok(())
+            }
+            "--help" | "-h" => {
+                println!("{USAGE}");
+                Ok(())
+            }
+            _ => Err(anyhow::anyhow!("unknown argument {flag}\n{USAGE}")),
+        };
+    }
     let tracer_provider = init_tracing()?;
 
     // reloads re-read this captured source

crates/server/tests/cli.rs

@@ -0,0 +1,24 @@
+use std::process::Command;
+
+#[test]
+fn version_flag_prints_the_version_and_exits() {
+    let out = Command::new(env!("CARGO_BIN_EXE_gw"))
+        .arg("--version")
+        .output()
+        .expect("run gw");
+    assert!(out.status.success());
+    assert_eq!(
+        String::from_utf8_lossy(&out.stdout).trim(),
+        format!("gw {}", env!("CARGO_PKG_VERSION"))
+    );
+}
+
+#[test]
+fn an_unknown_argument_is_refused_with_usage() {
+    let out = Command::new(env!("CARGO_BIN_EXE_gw"))
+        .arg("--bogus")
+        .output()
+        .expect("run gw");
+    assert!(!out.status.success());
+    assert!(String::from_utf8_lossy(&out.stderr).contains("usage: gw"));
+}

docs/providers.md

@@ -27,7 +27,7 @@ models:
 | `gemini` | `https://generativelanguage.googleapis.com` | gemini, realtime | `x-goog-api-key` (realtime = the Live API socket, live-verified: the bridge admits on `clientContent.turnComplete`, relays the binary frames, and settles `usageMetadata` — audio output tokens at their own weight) |
 | `deepseek` | `https://api.deepseek.com` | openai-chat | `Bearer` |
 | `openrouter` | `https://openrouter.ai/api` | openai-chat | `Bearer` (its `reasoning_details` shape is the one this gateway emits, so signed Anthropic reasoning round-trips through tool loops; verified live on free and paid models) |
-| `moonshot` | `https://api.moonshot.cn` | openai-chat | `Bearer` (Kimi K2 thinking: `reasoning_content` in and out, `thinking: {type: disabled}` passes through; the vendor's `/anthropic` base also works as `kind: anthropic` + `endpoint`) |
+| `moonshot` | `https://api.moonshot.cn` | openai-chat | `Bearer` (Kimi K2 thinking: `reasoning_content` in and out, `thinking: {type: disabled}` passes through; the vendor's `/anthropic` base also works as `kind: anthropic` + `endpoint`; the international site is `endpoint: https://api.moonshot.ai` — keys are site-specific and kimi-k3 is only there) |
 | `xai` | `https://api.x.ai` | openai-chat, responses, image, video, realtime | `Bearer` (Grok: `reasoning_effort` per model — grok-4.6 `low`–`xhigh`, grok-4.3 also `none`, others reject values they don't list; usage carries `prompt_tokens_details.cached_tokens` and `completion_tokens_details.reasoning_tokens`; xAI's own Anthropic-compatible surface is deprecated, so Anthropic clients reach Grok through this gateway's `/v1/messages` cross-protocol path; verified live: grok-4.3/4.6/4.20 chat + stream + effort tiers, `/v1/messages` both ways, grok-4.5 through `protocol: responses` natively and from the chat/messages surfaces incl. tool loops, grok-imagine-image-2.0 per-image units, and grok-voice-latest through `/v1/realtime` (`protocol: realtime`; xAI's `response.done` carries an empty `usage`, so the turn bills the delivered-output estimate — transcript tokens plus audio bytes/4 — as `estimated`, input unmetered: price the model per output unit accordingly); grok-imagine-video-1.5 through `/v1/videos/generations` + `GET /v1/videos/{id}` (`protocol: video`, unit price per generated second, vendor cost from `usage.cost_in_usd_ticks`); files/collections and vendor batches are not wired) |
 | `siliconflow` | `https://api.siliconflow.cn` | openai-chat, embeddings, rerank, tts, stt, image, video | `Bearer` (Qwen3 `enable_thinking`, DeepSeek/GLM/Kimi/MiniMax hosted models, bge/Qwen3 embeddings and rerankers, CosyVoice TTS, SenseVoice STT, Kolors images — all verified live) |