@mupt-ai/dari-gateway 0.3.2 → 0.3.3

package/README.md

@@ -36,27 +36,26 @@ The gateway is the same runtime that serves Dari's hosted routing service. Choos
 
 ## Run
 
-Requires [Bun](https://bun.sh) 1.3 or later.
+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:
 
 ```bash
-cp dari-gateway.example.yaml dari-gateway.yaml   # edit routers and models
 export DARI_GATEWAY_KEYS="$(openssl rand -hex 24)" # keys your clients send
 export ANTHROPIC_API_KEY=... OPENAI_API_KEY=...    # optional with subscriptions
-bun install
-bun run start
+bunx @mupt-ai/dari-gateway
 ```
 
-Or with Docker, from this directory (in the monorepo, run `docker build -t dari-gateway dari-gateway` from the repository root):
+Or with Docker:
 
 ```bash
-docker build -t dari-gateway .
 docker run -p 8080:8080 \
   -e DARI_GATEWAY_KEYS -e ANTHROPIC_API_KEY \
   -v "$PWD/dari-gateway.yaml:/config/dari-gateway.yaml" \
   -v dari-gateway-data:/data \
-  dari-gateway
+  ghcr.io/mupt-ai/dari-gateway
 ```
 
+From a clone, `bun install && bun run start` or `docker build -t dari-gateway .` run the same thing.
+
 | Variable | Meaning |
 | --- | --- |
 | `DARI_GATEWAY_KEYS` | Required. Comma-separated bearer keys clients authenticate with. |

package/dist/context/interaction_fingerprint.js

@@ -59,7 +59,7 @@ export function interactionLooseChain(routerId, request) {
 // transcript step. Growing history and tool changes preserve identity, while
 // unrelated work reusing the same harness session gets a new group.
 export function interactionConversationPrefix(request) {
-    const firstStep = conversationFingerprintSteps(looseFingerprintSteps(request.sourceProtocol, request.items)).at(0);
+    const firstStep = conversationFingerprintSteps(looseFingerprintSteps(request.sourceProtocol, request.items.filter((item) => !isOpenClawRuntimeContext(item)))).at(0);
     return sha256(stableStringify({
         version: CONVERSATION_PREFIX_VERSION,
         firstStep: firstStep === undefined
@@ -67,6 +67,15 @@ export function interactionConversationPrefix(request) {
             : canonicalInteractionFingerprintStep(firstStep),
     }));
 }
+// OpenClaw appends a user message of runtime context (active exec sessions,
+// subagents) to each request and drops it from later history. On a first
+// turn it merges into the first transcript step, so leaving it in would give
+// the first turn a different conversation than the turns after it.
+const OPENCLAW_RUNTIME_CONTEXT = "<<<BEGIN_OPENCLAW_INTERNAL_CONTEXT>>>";
+function isOpenClawRuntimeContext(item) {
+    return item.type === "message" && item.role === "user" && item.content.length > 0 &&
+        item.content.every((part) => part.type === "text" && part.text.trimStart().startsWith(OPENCLAW_RUNTIME_CONTEXT));
+}
 export function interactionPrefixContext(routerId, request, targets) {
     const chainByModel = new Map();
     const itemIndexByHashByModel = new Map();

package/dist/ingress/codex.js

@@ -8,7 +8,7 @@ import { openAIResponsesToInteraction } from "../protocol/input/openai_responses
 import { parseJsonRequest } from "./json.js";
 import { museCodeClientId } from "./muse_code.js";
 import { handleRoutedCompletion } from "../runtime/routed_completion.js";
-import { codexConversationIdentity, codexRequestPurpose, codexRequestContext, } from "../runtime/harness_identity/index.js";
+import { codexConversationIdentity, codexRequestPurpose, codexRequestContext, museCodeConversationIdentity, } from "../runtime/harness_identity/index.js";
 export async function handleResponses(request, routerId, selectorBackend) {
     try {
         const apiKey = bearerToken(request.headers);
@@ -23,14 +23,16 @@ export async function handleResponses(request, routerId, selectorBackend) {
             ...(agentContext === undefined ? {} : { agentContext }),
             ...(museClientId === undefined ? {} : { museCodeClientId: museClientId }),
         };
-        const harnessConversationIdentity = codexConversationIdentity(request.headers, rawPayload);
+        const codexIdentity = codexConversationIdentity(request.headers, rawPayload);
         // Codex applies its global model_reasoning_effort setting to custom model
         // providers. That preference describes a direct model, not a heterogeneous
         // router, so let Dari select the reasoning effort for verified Codex turns.
         // Parsing above still validates malformed Responses reasoning controls.
-        if (harnessConversationIdentity !== undefined && explicitModelId(payload.model) === null) {
+        if (codexIdentity !== undefined && explicitModelId(payload.model) === null) {
             payload.reasoning = undefined;
         }
+        const harnessConversationIdentity = codexIdentity
+            ?? (museClientId === undefined ? undefined : museCodeConversationIdentity(request.headers));
         return await handleRoutedCompletion({
             request,
             routerId,

package/dist/runtime/harness_identity/conversation.d.ts

@@ -13,6 +13,10 @@ export type HarnessConversationIdentity = {
     kind: "openai_chat_session";
     value: string;
     qualifiers?: string[];
+} | {
+    kind: "muse_code_thread";
+    value: string;
+    qualifiers?: string[];
 };
 export type HarnessConversationScope = {
     organizationId: string;

package/dist/runtime/harness_identity/extract.d.ts

@@ -6,4 +6,5 @@ export declare function openCodeRequestContext(headers: Headers): AgentRequestCo
 export declare function codexConversationIdentity(headers: Headers, payload: Record<string, unknown>): HarnessConversationIdentity | undefined;
 export declare function codexRequestPurpose(headers: Headers, payload: Record<string, unknown>): InteractionRequestPurpose | undefined;
 export declare function codexRequestContext(headers: Headers, payload: Record<string, unknown>): AgentRequestContext | undefined;
+export declare function museCodeConversationIdentity(headers: Headers): HarnessConversationIdentity | undefined;
 export declare function claudeCodeConversationIdentity(headers: Headers, metadataSessionId?: string): HarnessConversationIdentity | undefined;

package/dist/runtime/harness_identity/extract.js

@@ -129,6 +129,13 @@ export function codexRequestContext(headers, payload) {
         agentType: boundedIdentity(agentKind, "x-codex-turn-metadata.subagent_kind"),
     };
 }
+// Muse Code sends its thread on every Responses request. The main agent's
+// thread is its session, and each subagent has its own thread, so each keeps
+// its own conversation and lease.
+export function museCodeConversationIdentity(headers) {
+    const threadId = boundedIdentity(headers.get("x-tbh-thread-id"), "x-tbh-thread-id");
+    return threadId === undefined ? undefined : { kind: "muse_code_thread", value: threadId };
+}
 export function claudeCodeConversationIdentity(headers, metadataSessionId) {
     const candidates = [
         uuidIdentity(headers.get("x-claude-code-session-id"), "x-claude-code-session-id"),

package/dist/runtime/harness_identity/index.d.ts

@@ -1,2 +1,2 @@
 export { scopedHarnessAgentId, scopedHarnessConversationId, scopedHarnessSessionId, type HarnessConversationIdentity, type HarnessConversationScope, } from "./conversation.js";
-export { chatCompletionsConversationIdentity, claudeCodeConversationIdentity, codexConversationIdentity, codexRequestPurpose, codexRequestContext, openCodeRequestContext, piRequestContext, } from "./extract.js";
+export { chatCompletionsConversationIdentity, claudeCodeConversationIdentity, codexConversationIdentity, codexRequestPurpose, codexRequestContext, museCodeConversationIdentity, openCodeRequestContext, piRequestContext, } from "./extract.js";

package/dist/runtime/harness_identity/index.js

@@ -1,2 +1,2 @@
 export { scopedHarnessAgentId, scopedHarnessConversationId, scopedHarnessSessionId, } from "./conversation.js";
-export { chatCompletionsConversationIdentity, claudeCodeConversationIdentity, codexConversationIdentity, codexRequestPurpose, codexRequestContext, openCodeRequestContext, piRequestContext, } from "./extract.js";
+export { chatCompletionsConversationIdentity, claudeCodeConversationIdentity, codexConversationIdentity, codexRequestPurpose, codexRequestContext, museCodeConversationIdentity, openCodeRequestContext, piRequestContext, } from "./extract.js";

package/dist/self_host/config.js

@@ -5,7 +5,8 @@ const MODEL_ID_PATTERN = /^[a-z0-9][a-z0-9.-]*\/\S+$/;
 export async function loadGatewayConfig(path) {
     const file = Bun.file(path);
     if (!(await file.exists())) {
-        throw new Error(`Gateway config not found at ${path}.`);
+        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).");
     }
     const text = await file.text();
     const raw = path.endsWith(".json") ? JSON.parse(text) : Bun.YAML.parse(text);

package/dist/self_host/main.js

@@ -106,6 +106,10 @@ export async function startSelfHostedGateway(env = process.env) {
     return { server, close: () => db.close() };
 }
 if (import.meta.main) {
-    const { server, close } = await startSelfHostedGateway();
-    installShutdownHandlers(server, async () => close());
+    // 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)}`);
+        process.exit(1);
+    });
+    installShutdownHandlers(started.server, async () => started.close());
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mupt-ai/dari-gateway",
-  "version": "0.3.2",
+  "version": "0.3.3",
   "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": {

package/src/context/interaction_fingerprint.ts

@@ -105,7 +105,10 @@ export function interactionLooseChain(routerId: string, request: InteractionRequ
 // unrelated work reusing the same harness session gets a new group.
 export function interactionConversationPrefix(request: InteractionRequest): string {
   const firstStep = conversationFingerprintSteps(
-    looseFingerprintSteps(request.sourceProtocol, request.items),
+    looseFingerprintSteps(
+      request.sourceProtocol,
+      request.items.filter((item) => !isOpenClawRuntimeContext(item)),
+    ),
   ).at(0);
   return sha256(stableStringify({
     version: CONVERSATION_PREFIX_VERSION,
@@ -115,6 +118,17 @@ export function interactionConversationPrefix(request: InteractionRequest): stri
   }));
 }
 
+// OpenClaw appends a user message of runtime context (active exec sessions,
+// subagents) to each request and drops it from later history. On a first
+// turn it merges into the first transcript step, so leaving it in would give
+// the first turn a different conversation than the turns after it.
+const OPENCLAW_RUNTIME_CONTEXT = "<<<BEGIN_OPENCLAW_INTERNAL_CONTEXT>>>";
+
+function isOpenClawRuntimeContext(item: InteractionItem): boolean {
+  return item.type === "message" && item.role === "user" && item.content.length > 0 &&
+    item.content.every((part) => part.type === "text" && part.text.trimStart().startsWith(OPENCLAW_RUNTIME_CONTEXT));
+}
+
 export function interactionPrefixContext(
   routerId: string,
   request: InteractionRequest,

package/src/ingress/codex.ts

@@ -13,6 +13,7 @@ import {
   codexConversationIdentity,
   codexRequestPurpose,
   codexRequestContext,
+  museCodeConversationIdentity,
 } from "../runtime/harness_identity/index.js";
 
 export async function handleResponses(
@@ -33,17 +34,16 @@ export async function handleResponses(
       ...(agentContext === undefined ? {} : { agentContext }),
       ...(museClientId === undefined ? {} : { museCodeClientId: museClientId }),
     };
-    const harnessConversationIdentity = codexConversationIdentity(
-      request.headers,
-      rawPayload,
-    );
+    const codexIdentity = codexConversationIdentity(request.headers, rawPayload);
     // Codex applies its global model_reasoning_effort setting to custom model
     // providers. That preference describes a direct model, not a heterogeneous
     // router, so let Dari select the reasoning effort for verified Codex turns.
     // Parsing above still validates malformed Responses reasoning controls.
-    if (harnessConversationIdentity !== undefined && explicitModelId(payload.model) === null) {
+    if (codexIdentity !== undefined && explicitModelId(payload.model) === null) {
       payload.reasoning = undefined;
     }
+    const harnessConversationIdentity = codexIdentity
+      ?? (museClientId === undefined ? undefined : museCodeConversationIdentity(request.headers));
     return await handleRoutedCompletion({
       request,
       routerId,

package/src/runtime/harness_identity/conversation.ts

@@ -19,6 +19,11 @@ export type HarnessConversationIdentity =
       kind: "openai_chat_session";
       value: string;
       qualifiers?: string[];
+    }
+  | {
+      kind: "muse_code_thread";
+      value: string;
+      qualifiers?: string[];
     };
 
 export type HarnessConversationScope = {

package/src/runtime/harness_identity/extract.ts

@@ -183,6 +183,16 @@ export function codexRequestContext(
   };
 }
 
+// Muse Code sends its thread on every Responses request. The main agent's
+// thread is its session, and each subagent has its own thread, so each keeps
+// its own conversation and lease.
+export function museCodeConversationIdentity(
+  headers: Headers,
+): HarnessConversationIdentity | undefined {
+  const threadId = boundedIdentity(headers.get("x-tbh-thread-id"), "x-tbh-thread-id");
+  return threadId === undefined ? undefined : { kind: "muse_code_thread", value: threadId };
+}
+
 export function claudeCodeConversationIdentity(
   headers: Headers,
   metadataSessionId?: string,

package/src/runtime/harness_identity/index.ts

@@ -11,6 +11,7 @@ export {
   codexConversationIdentity,
   codexRequestPurpose,
   codexRequestContext,
+  museCodeConversationIdentity,
   openCodeRequestContext,
   piRequestContext,
 } from "./extract.js";

package/src/self_host/config.ts

@@ -65,7 +65,10 @@ const MODEL_ID_PATTERN = /^[a-z0-9][a-z0-9.-]*\/\S+$/;
 export async function loadGatewayConfig(path: string): Promise<GatewayConfig> {
   const file = Bun.file(path);
   if (!(await file.exists())) {
-    throw new Error(`Gateway config not found at ${path}.`);
+    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).",
+    );
   }
   const text = await file.text();
   const raw: unknown = path.endsWith(".json") ? JSON.parse(text) : Bun.YAML.parse(text);

package/src/self_host/main.ts

@@ -118,6 +118,10 @@ export async function startSelfHostedGateway(env: Record<string, string> = proce
 }
 
 if (import.meta.main) {
-  const { server, close } = await startSelfHostedGateway();
-  installShutdownHandlers(server, async () => close());
+  // 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)}`);
+    process.exit(1);
+  });
+  installShutdownHandlers(started.server, async () => started.close());
 }