@mupt-ai/dari-gateway 0.3.4 → 0.3.5

package/README.md

@@ -36,7 +36,9 @@ The gateway is the same runtime that serves Dari's hosted routing service. Choos
 
 ## Run
 
-Requires [Bun](https://bun.sh) 1.3 or later. Write a `dari-gateway.yaml` (start from [`dari-gateway.example.yaml`](dari-gateway.example.yaml)), then:
+The easiest way to run it on your own machine is the [Dari CLI](https://github.com/mupt-ai/dari-cli): choose **Fully local** in `dari gateway setup` (or on your first `dari claude`), and it installs this gateway's standalone binary, runs it in the background, and connects your agents.
+
+To run it yourself, use [Bun](https://bun.sh) 1.3 or later, or the standalone binary for your platform from the [releases](https://github.com/mupt-ai/dari-gateway/releases). A config file is optional: without `DARI_GATEWAY_CONFIG` or a `dari-gateway.yaml` in the working directory, the gateway serves a default router of current Anthropic and OpenAI models, each once its provider has a subscription or API key. To choose models, write a `dari-gateway.yaml` (start from [`dari-gateway.example.yaml`](dari-gateway.example.yaml)). Then:
 
 ```bash
 export DARI_GATEWAY_KEYS="$(openssl rand -hex 24)" # keys your clients send

package/dist/models/canonical.d.ts

@@ -0,0 +1,2 @@
+/** The canonical model ids Dari serves, from the monorepo's router catalog. */
+export declare const CANONICAL_MODEL_IDS: readonly string[];

package/dist/models/canonical.js

@@ -0,0 +1,3 @@
+import canonicalIds from "./canonical_model_ids.json" with { type: "json" };
+/** The canonical model ids Dari serves, from the monorepo's router catalog. */
+export const CANONICAL_MODEL_IDS = canonicalIds;

package/dist/models/canonical_model_ids.json

@@ -0,0 +1,33 @@
+[
+    "openai/gpt-6-astra",
+    "openai/gpt-6.1-terra",
+    "openai/gpt-6.1-sol",
+    "openai/gpt-6-sol",
+    "openai/gpt-6-luna",
+    "openai/gpt-5.6-sol",
+    "openai/gpt-5.6-terra",
+    "openai/gpt-5.6-luna",
+    "anthropic/claude-opus-5-5",
+    "anthropic/claude-opus-5",
+    "anthropic/claude-sonnet-5-5",
+    "anthropic/claude-sonnet-5",
+    "anthropic/claude-fable-5-2",
+    "anthropic/claude-fable-5-1",
+    "anthropic/claude-fable-5",
+    "anthropic/claude-haiku-5-5",
+    "anthropic/claude-haiku-4-5",
+    "meta/muse-wonder-1.0",
+    "meta/muse-spark-1.3",
+    "meta/muse-ember",
+    "meta/muse-spark-1.2",
+    "moonshotai/Kimi-K3",
+    "zai-org/GLM-5.3",
+    "zai-org/GLM-5.3-Flash",
+    "deepseek-ai/DeepSeek-V4-Pro-0813",
+    "zai-org/GLM-5.2",
+    "deepseek-ai/DeepSeek-V4-Flash-0731",
+    "deepseek-ai/DeepSeek-V4.1-Flash-1017",
+    "deepseek-ai/DeepSeek-V4.1-Flash",
+    "xai/grok-4.6",
+    "xai/grok-4.7"
+]

package/dist/models/runtime_overrides.js

@@ -26,6 +26,10 @@ const MODEL_COMPAT_OVERRIDES = {
         forceAdaptiveThinking: true,
         supportsTemperature: false,
     },
+    "anthropic/claude-fable-5-2": {
+        forceAdaptiveThinking: true,
+        supportsTemperature: false,
+    },
     "anthropic/claude-opus-5-5": {
         forceAdaptiveThinking: true,
         supportsTemperature: false,
@@ -65,6 +69,14 @@ const MODEL_COMPAT_OVERRIDES = {
         supportsExplicitPromptCacheMode: true,
         supportsReasoningConfigurationUpdate: true,
     },
+    "openai/gpt-6.1-terra": {
+        supportsStrictMode: true,
+        supportsOpenAIGrammarTools: true,
+        supportsAdditionalTools: true,
+        supportsToolSearch: true,
+        supportsExplicitPromptCacheMode: true,
+        supportsReasoningConfigurationUpdate: true,
+    },
     "openai/gpt-6-sol": {
         supportsStrictMode: true,
         supportsOpenAIGrammarTools: true,
@@ -138,7 +150,9 @@ const NATIVE_REASONING_COMPAT_KEYS = new Set([
 const MODEL_INPUT_OVERRIDES = {
     "anthropic/claude-haiku-5-5": ["text", "image"],
     "anthropic/claude-sonnet-5-5": ["text", "image"],
+    "meta/muse-ember": ["text", "image"],
     "openai/gpt-6.1-sol": ["text", "image"],
+    "openai/gpt-6.1-terra": ["text", "image"],
     "openai/gpt-6-sol": ["text", "image"],
     "openai/gpt-6-luna": ["text", "image"],
     "openai/gpt-5.6-luna": ["text", "image"],

package/dist/self_host/config.d.ts

@@ -41,5 +41,11 @@ export type RouterModelConfig = {
     id: string;
     thinking?: ReasoningEffort[];
 };
-export declare function loadGatewayConfig(path: string): Promise<GatewayConfig>;
+export declare const DEFAULT_GATEWAY_CONFIG: GatewayConfig;
+/**
+ * Load the config at `path`, which must exist. Without a path, load
+ * `dari-gateway.yaml` from the working directory when present, and otherwise
+ * serve DEFAULT_GATEWAY_CONFIG.
+ */
+export declare function loadGatewayConfig(path: string | undefined): Promise<GatewayConfig>;
 export declare function parseGatewayConfig(raw: unknown): GatewayConfig;

package/dist/self_host/config.js

@@ -2,11 +2,40 @@ import { isReasoningEffort, isRecord, } from "@mupt-ai/dari-router/policy-engine
 const ROUTER_ID_PATTERN = /^[A-Za-z0-9_-]+$/;
 const PROVIDER_ID_PATTERN = /^[a-z0-9][a-z0-9.-]*$/;
 const MODEL_ID_PATTERN = /^[a-z0-9][a-z0-9.-]*\/\S+$/;
+// Served when no config file is set or found. A model is served once its
+// provider has a credential (a connected subscription or the provider's API
+// key); dari/routing without Auto Routing serves the first available one.
+export const DEFAULT_GATEWAY_CONFIG = parseGatewayConfig({
+    routers: {
+        default: {
+            models: [
+                "anthropic/claude-sonnet-5-5",
+                "anthropic/claude-opus-5-5",
+                "anthropic/claude-fable-5-2",
+                "anthropic/claude-haiku-5-5",
+                "openai/gpt-6.1-sol",
+                "openai/gpt-6.1-terra",
+            ],
+        },
+    },
+});
+const DEFAULT_CONFIG_PATH = "dari-gateway.yaml";
+/**
+ * Load the config at `path`, which must exist. Without a path, load
+ * `dari-gateway.yaml` from the working directory when present, and otherwise
+ * serve DEFAULT_GATEWAY_CONFIG.
+ */
 export async function loadGatewayConfig(path) {
+    if (path === undefined) {
+        if (!(await Bun.file(DEFAULT_CONFIG_PATH).exists()))
+            return DEFAULT_GATEWAY_CONFIG;
+        path = DEFAULT_CONFIG_PATH;
+    }
     const file = Bun.file(path);
     if (!(await file.exists())) {
         throw new Error(`Gateway config not found at ${path}. Write one from dari-gateway.example.yaml, `
-            + "or set DARI_GATEWAY_CONFIG (in the container image, mount it at /config/dari-gateway.yaml).");
+            + "or unset DARI_GATEWAY_CONFIG to serve the default router "
+            + "(in the container image, mount it at /config/dari-gateway.yaml).");
     }
     const text = await file.text();
     const raw = path.endsWith(".json") ? JSON.parse(text) : Bun.YAML.parse(text);

package/dist/self_host/main.d.ts

@@ -3,7 +3,8 @@
  * Start a self-hosted Dari Gateway from environment variables:
  *
  * - `DARI_GATEWAY_KEYS` (required): comma-separated bearer keys clients use.
- * - `DARI_GATEWAY_CONFIG`: router config file (default `dari-gateway.yaml`).
+ * - `DARI_GATEWAY_CONFIG`: router config file. Unset, `dari-gateway.yaml`
+ *   serves when present, and otherwise the default router.
  * - `DARI_GATEWAY_DB`: SQLite database for subscriptions and prefix-cache
  *   state (default `dari-gateway.db`). It holds subscription tokens.
  * - `DARI_AUTO_ROUTING_API_KEY`: choose among a router's models with Dari
@@ -15,3 +16,5 @@ export declare function startSelfHostedGateway(env?: Record<string, string>): Pr
     server: import("../http/http2_server.js").Http2RoutingServer | Bun.Server<undefined>;
     close: () => void;
 }>;
+/** Run the gateway as a process: the `dari-gateway` bin and compiled binary. */
+export declare function runSelfHostedGateway(): Promise<void>;

package/dist/self_host/main.js

@@ -11,7 +11,8 @@ import { createPrioritySelectorBackend } from "../strategies/priority_selector.j
 import { createSelfHostedPlatform } from "./platform.js";
 import { providerForModel } from "@mupt-ai/dari-router/policy-engine";
 import { checkProviderKey } from "../providers/key_check.js";
-import { loadGatewayConfig } from "./config.js";
+import { CANONICAL_MODEL_IDS } from "../models/canonical.js";
+import { DEFAULT_GATEWAY_CONFIG, loadGatewayConfig } from "./config.js";
 import { configuredProviderApiKey, gatewayKeyIndex, providersWithoutCredentials } from "./resolve.js";
 import { PrefixStore } from "./prefix_store.js";
 import { SessionPinStore } from "./session_pin_store.js";
@@ -21,7 +22,8 @@ const SUBSCRIPTION_ROUTES = "/v1/organizations/current/credentials";
  * Start a self-hosted Dari Gateway from environment variables:
  *
  * - `DARI_GATEWAY_KEYS` (required): comma-separated bearer keys clients use.
- * - `DARI_GATEWAY_CONFIG`: router config file (default `dari-gateway.yaml`).
+ * - `DARI_GATEWAY_CONFIG`: router config file. Unset, `dari-gateway.yaml`
+ *   serves when present, and otherwise the default router.
  * - `DARI_GATEWAY_DB`: SQLite database for subscriptions and prefix-cache
  *   state (default `dari-gateway.db`). It holds subscription tokens.
  * - `DARI_AUTO_ROUTING_API_KEY`: choose among a router's models with Dari
@@ -37,7 +39,7 @@ export async function startSelfHostedGateway(env = process.env) {
     if (gatewayKeys.length === 0) {
         throw new Error("Set DARI_GATEWAY_KEYS to the comma-separated keys clients authenticate with.");
     }
-    const config = await loadGatewayConfig(env.DARI_GATEWAY_CONFIG ?? "dari-gateway.yaml");
+    const config = await loadGatewayConfig(env.DARI_GATEWAY_CONFIG?.trim() || undefined);
     const databasePath = env.DARI_GATEWAY_DB ?? "dari-gateway.db";
     const created = databasePath !== ":memory:" && !existsSync(databasePath);
     const db = new Database(databasePath, { create: true });
@@ -134,7 +136,30 @@ async function checkConfiguredProviderKeys(config, providerApiKey) {
         console.warn("dari-gateway could not check these provider API keys", JSON.stringify({ providers: unchecked }));
     }
 }
-if (import.meta.main) {
+/** Run the gateway as a process: the `dari-gateway` bin and compiled binary. */
+export async function runSelfHostedGateway() {
+    // `--print-config` prints the router config the gateway would serve, as
+    // JSON, and exits; `--print-default-config` prints the built-in default.
+    // The Dari CLI edits a local gateway's models from them.
+    if (process.argv.includes("--print-canonical-models")) {
+        console.log(JSON.stringify(CANONICAL_MODEL_IDS));
+        return;
+    }
+    if (process.argv.includes("--print-default-config")) {
+        console.log(JSON.stringify(DEFAULT_GATEWAY_CONFIG));
+        return;
+    }
+    if (process.argv.includes("--print-config")) {
+        try {
+            const config = await loadGatewayConfig(process.env.DARI_GATEWAY_CONFIG?.trim() || undefined);
+            console.log(JSON.stringify(config));
+            return;
+        }
+        catch (error) {
+            console.error(`dari-gateway: ${error instanceof Error ? error.message : String(error)}`);
+            process.exit(1);
+        }
+    }
     // Startup errors are configuration mistakes: print the message, not a stack.
     const started = await startSelfHostedGateway().catch((error) => {
         console.error(`dari-gateway: ${error instanceof Error ? error.message : String(error)}`);
@@ -142,3 +167,5 @@ if (import.meta.main) {
     });
     installShutdownHandlers(started.server, async () => started.close());
 }
+if (import.meta.main)
+    await runSelfHostedGateway();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mupt-ai/dari-gateway",
-  "version": "0.3.4",
+  "version": "0.3.5",
   "description": "Self-hostable LLM gateway for coding agents and APIs: one endpoint for Anthropic Messages, OpenAI Chat Completions, and OpenAI Responses, with subscription pooling, failover, and prompt-cache-aware routing.",
   "license": "Apache-2.0",
   "repository": {
@@ -42,6 +42,7 @@
   ],
   "scripts": {
     "start": "bun run src/self_host/main.ts",
+    "compile": "bun scripts/compile.ts",
     "build": "rm -rf dist && tsc -p tsconfig.build.json",
     "typecheck": "tsc --noEmit",
     "test": "bun test",

package/src/models/canonical.ts

@@ -0,0 +1,4 @@
+import canonicalIds from "./canonical_model_ids.json" with { type: "json" };
+
+/** The canonical model ids Dari serves, from the monorepo's router catalog. */
+export const CANONICAL_MODEL_IDS: readonly string[] = canonicalIds;

package/src/models/canonical_model_ids.json

@@ -0,0 +1,33 @@
+[
+  "openai/gpt-6-astra",
+  "openai/gpt-6.1-terra",
+  "openai/gpt-6.1-sol",
+  "openai/gpt-6-sol",
+  "openai/gpt-6-luna",
+  "openai/gpt-5.6-sol",
+  "openai/gpt-5.6-terra",
+  "openai/gpt-5.6-luna",
+  "anthropic/claude-opus-5-5",
+  "anthropic/claude-opus-5",
+  "anthropic/claude-sonnet-5-5",
+  "anthropic/claude-sonnet-5",
+  "anthropic/claude-fable-5-2",
+  "anthropic/claude-fable-5-1",
+  "anthropic/claude-fable-5",
+  "anthropic/claude-haiku-5-5",
+  "anthropic/claude-haiku-4-5",
+  "meta/muse-wonder-1.0",
+  "meta/muse-spark-1.3",
+  "meta/muse-ember",
+  "meta/muse-spark-1.2",
+  "moonshotai/Kimi-K3",
+  "zai-org/GLM-5.3",
+  "zai-org/GLM-5.3-Flash",
+  "deepseek-ai/DeepSeek-V4-Pro-0813",
+  "zai-org/GLM-5.2",
+  "deepseek-ai/DeepSeek-V4-Flash-0731",
+  "deepseek-ai/DeepSeek-V4.1-Flash-1017",
+  "deepseek-ai/DeepSeek-V4.1-Flash",
+  "xai/grok-4.6",
+  "xai/grok-4.7"
+]

package/src/models/runtime_overrides.ts

@@ -29,6 +29,10 @@ const MODEL_COMPAT_OVERRIDES: Record<string, Record<string, unknown>> = {
     forceAdaptiveThinking: true,
     supportsTemperature: false,
   },
+  "anthropic/claude-fable-5-2": {
+    forceAdaptiveThinking: true,
+    supportsTemperature: false,
+  },
   "anthropic/claude-opus-5-5": {
     forceAdaptiveThinking: true,
     supportsTemperature: false,
@@ -68,6 +72,14 @@ const MODEL_COMPAT_OVERRIDES: Record<string, Record<string, unknown>> = {
     supportsExplicitPromptCacheMode: true,
     supportsReasoningConfigurationUpdate: true,
   },
+  "openai/gpt-6.1-terra": {
+    supportsStrictMode: true,
+    supportsOpenAIGrammarTools: true,
+    supportsAdditionalTools: true,
+    supportsToolSearch: true,
+    supportsExplicitPromptCacheMode: true,
+    supportsReasoningConfigurationUpdate: true,
+  },
   "openai/gpt-6-sol": {
     supportsStrictMode: true,
     supportsOpenAIGrammarTools: true,
@@ -144,7 +156,9 @@ const NATIVE_REASONING_COMPAT_KEYS: ReadonlySet<string> = new Set([
 const MODEL_INPUT_OVERRIDES: Record<string, string[]> = {
   "anthropic/claude-haiku-5-5": ["text", "image"],
   "anthropic/claude-sonnet-5-5": ["text", "image"],
+  "meta/muse-ember": ["text", "image"],
   "openai/gpt-6.1-sol": ["text", "image"],
+  "openai/gpt-6.1-terra": ["text", "image"],
   "openai/gpt-6-sol": ["text", "image"],
   "openai/gpt-6-luna": ["text", "image"],
   "openai/gpt-5.6-luna": ["text", "image"],

package/src/self_host/config.ts

@@ -62,12 +62,42 @@ const ROUTER_ID_PATTERN = /^[A-Za-z0-9_-]+$/;
 const PROVIDER_ID_PATTERN = /^[a-z0-9][a-z0-9.-]*$/;
 const MODEL_ID_PATTERN = /^[a-z0-9][a-z0-9.-]*\/\S+$/;
 
-export async function loadGatewayConfig(path: string): Promise<GatewayConfig> {
+// Served when no config file is set or found. A model is served once its
+// provider has a credential (a connected subscription or the provider's API
+// key); dari/routing without Auto Routing serves the first available one.
+export const DEFAULT_GATEWAY_CONFIG: GatewayConfig = parseGatewayConfig({
+  routers: {
+    default: {
+      models: [
+        "anthropic/claude-sonnet-5-5",
+        "anthropic/claude-opus-5-5",
+        "anthropic/claude-fable-5-2",
+        "anthropic/claude-haiku-5-5",
+        "openai/gpt-6.1-sol",
+        "openai/gpt-6.1-terra",
+      ],
+    },
+  },
+});
+
+const DEFAULT_CONFIG_PATH = "dari-gateway.yaml";
+
+/**
+ * Load the config at `path`, which must exist. Without a path, load
+ * `dari-gateway.yaml` from the working directory when present, and otherwise
+ * serve DEFAULT_GATEWAY_CONFIG.
+ */
+export async function loadGatewayConfig(path: string | undefined): Promise<GatewayConfig> {
+  if (path === undefined) {
+    if (!(await Bun.file(DEFAULT_CONFIG_PATH).exists())) return DEFAULT_GATEWAY_CONFIG;
+    path = DEFAULT_CONFIG_PATH;
+  }
   const file = Bun.file(path);
   if (!(await file.exists())) {
     throw new Error(
       `Gateway config not found at ${path}. Write one from dari-gateway.example.yaml, `
-        + "or set DARI_GATEWAY_CONFIG (in the container image, mount it at /config/dari-gateway.yaml).",
+        + "or unset DARI_GATEWAY_CONFIG to serve the default router "
+        + "(in the container image, mount it at /config/dari-gateway.yaml).",
     );
   }
   const text = await file.text();

package/src/self_host/main.ts

@@ -13,7 +13,8 @@ import type { SelectorBackend } from "../strategies/routing.js";
 import { createSelfHostedPlatform } from "./platform.js";
 import { providerForModel } from "@mupt-ai/dari-router/policy-engine";
 import { checkProviderKey } from "../providers/key_check.js";
-import { loadGatewayConfig, type GatewayConfig } from "./config.js";
+import { CANONICAL_MODEL_IDS } from "../models/canonical.js";
+import { DEFAULT_GATEWAY_CONFIG, loadGatewayConfig, type GatewayConfig } from "./config.js";
 import { configuredProviderApiKey, gatewayKeyIndex, providersWithoutCredentials } from "./resolve.js";
 import { PrefixStore } from "./prefix_store.js";
 import { SessionPinStore } from "./session_pin_store.js";
@@ -25,7 +26,8 @@ const SUBSCRIPTION_ROUTES = "/v1/organizations/current/credentials";
  * Start a self-hosted Dari Gateway from environment variables:
  *
  * - `DARI_GATEWAY_KEYS` (required): comma-separated bearer keys clients use.
- * - `DARI_GATEWAY_CONFIG`: router config file (default `dari-gateway.yaml`).
+ * - `DARI_GATEWAY_CONFIG`: router config file. Unset, `dari-gateway.yaml`
+ *   serves when present, and otherwise the default router.
  * - `DARI_GATEWAY_DB`: SQLite database for subscriptions and prefix-cache
  *   state (default `dari-gateway.db`). It holds subscription tokens.
  * - `DARI_AUTO_ROUTING_API_KEY`: choose among a router's models with Dari
@@ -41,7 +43,7 @@ export async function startSelfHostedGateway(env: Record<string, string> = proce
   if (gatewayKeys.length === 0) {
     throw new Error("Set DARI_GATEWAY_KEYS to the comma-separated keys clients authenticate with.");
   }
-  const config = await loadGatewayConfig(env.DARI_GATEWAY_CONFIG ?? "dari-gateway.yaml");
+  const config = await loadGatewayConfig(env.DARI_GATEWAY_CONFIG?.trim() || undefined);
   const databasePath = env.DARI_GATEWAY_DB ?? "dari-gateway.db";
   const created = databasePath !== ":memory:" && !existsSync(databasePath);
   const db = new Database(databasePath, { create: true });
@@ -153,7 +155,29 @@ async function checkConfiguredProviderKeys(
   }
 }
 
-if (import.meta.main) {
+/** Run the gateway as a process: the `dari-gateway` bin and compiled binary. */
+export async function runSelfHostedGateway(): Promise<void> {
+  // `--print-config` prints the router config the gateway would serve, as
+  // JSON, and exits; `--print-default-config` prints the built-in default.
+  // The Dari CLI edits a local gateway's models from them.
+  if (process.argv.includes("--print-canonical-models")) {
+    console.log(JSON.stringify(CANONICAL_MODEL_IDS));
+    return;
+  }
+  if (process.argv.includes("--print-default-config")) {
+    console.log(JSON.stringify(DEFAULT_GATEWAY_CONFIG));
+    return;
+  }
+  if (process.argv.includes("--print-config")) {
+    try {
+      const config = await loadGatewayConfig(process.env.DARI_GATEWAY_CONFIG?.trim() || undefined);
+      console.log(JSON.stringify(config));
+      return;
+    } catch (error) {
+      console.error(`dari-gateway: ${error instanceof Error ? error.message : String(error)}`);
+      process.exit(1);
+    }
+  }
   // Startup errors are configuration mistakes: print the message, not a stack.
   const started = await startSelfHostedGateway().catch((error: unknown) => {
     console.error(`dari-gateway: ${error instanceof Error ? error.message : String(error)}`);
@@ -161,3 +185,5 @@ if (import.meta.main) {
   });
   installShutdownHandlers(started.server, async () => started.close());
 }
+
+if (import.meta.main) await runSelfHostedGateway();