pi-opencode-go-provider 1.0.16 → 1.0.18

package/.pi/fabric/mesh/state.json

@@ -0,0 +1,31 @@
+{
+  "format": 1,
+  "entries": {
+    "topology/peer-seq": {
+      "key": "topology/peer-seq",
+      "value": {
+        "format": 1,
+        "next": 1
+      },
+      "version": 1,
+      "updatedAt": 1788421061231,
+      "updatedBy": {
+        "id": "session:01a06632-9590-73a7-8e4c-bf4f4d6d8277",
+        "name": "main",
+        "kind": "main",
+        "sessionId": "01a06632-9590-73a7-8e4c-bf4f4d6d8277"
+      }
+    }
+  },
+  "versions": {
+    "topology/peer-seq": 1,
+    "topology/hosts/ae5878888fad007023c5ff775ac6356930ec9eaacf6e5b09ceebb18d9d4e54c8": 419,
+    "sessions/01a06632-9590-73a7-8e4c-bf4f4d6d8277": 418,
+    "topology/participants/ae5878888fad007023c5ff775ac6356930ec9eaacf6e5b09ceebb18d9d4e54c8": 419
+  },
+  "tombstoneOrder": [
+    "sessions/01a06632-9590-73a7-8e4c-bf4f4d6d8277",
+    "topology/participants/ae5878888fad007023c5ff775ac6356930ec9eaacf6e5b09ceebb18d9d4e54c8",
+    "topology/hosts/ae5878888fad007023c5ff775ac6356930ec9eaacf6e5b09ceebb18d9d4e54c8"
+  ]
+}

package/README.md

@@ -20,6 +20,7 @@ _Go-optimized endpoints for lower latency — 14+ models for [pi](https://github
 - **Fast & Efficient** - Go-optimized endpoints for lower latency
 - **Cost Tracking** with per-model pricing for budget management
 - **Reasoning Models** with thinking level maps for proper effort control
+- **Prompt-cache session affinity** — sends `x-opencode-session` and `x-opencode-client` so OpenCode Go can pin a session to the same cache node
 
 ## Installation
 

package/index.ts

@@ -339,7 +339,40 @@ async function resolveApiKey(modelRegistry: ModelRegistry): Promise<void> {
   cachedApiKey = await modelRegistry.getApiKeyForProvider("opencode-go") ?? undefined;
 }
 
-// ─── Extension Entry Point ────────────────────────────────────────────────────
+// OpenCode Go prompt-cache routing headers.
+// Pi core already injects these in mergeProviderAttributionHeaders for
+// provider "opencode-go" / host opencode.ai (see #4847). This hook is a
+// backstop: it fills them only when missing, so older pi builds and any
+// request path that skipped transformHeaders still pin the session.
+// Per-request via before_provider_headers — not registerProvider.headers —
+// because provider-wide headers leak to raw helper streams
+// (getApiKeyAndHeaders) that must not inherit the main session's cache
+// lineage. No custom streamSimple is needed; headers are applied before
+// provider dispatch.
+const OPENCODE_SESSION_HEADER = "x-opencode-session";
+const OPENCODE_CLIENT_HEADER = "x-opencode-client";
+const OPENCODE_CLIENT = "pi";
+
+function headerLookup(headers: Record<string, string | null | undefined>, name: string): string | undefined {
+  const lower = name.toLowerCase();
+  for (const [key, value] of Object.entries(headers)) {
+    if (key.toLowerCase() === lower) return value ?? undefined;
+  }
+  return undefined;
+}
+
+export function applyOpenCodeSessionHeaders(
+  headers: Record<string, string | null>,
+  sessionId: string | undefined,
+): Record<string, string | null> {
+  if (sessionId && !headerLookup(headers, OPENCODE_SESSION_HEADER)) {
+    headers[OPENCODE_SESSION_HEADER] = sessionId;
+  }
+  if (!headerLookup(headers, OPENCODE_CLIENT_HEADER)) {
+    headers[OPENCODE_CLIENT_HEADER] = OPENCODE_CLIENT;
+  }
+  return headers;
+}
 
 export default function (pi: ExtensionAPI) {
   const embeddedModels = modelsData as JsonModel[];
@@ -369,6 +402,11 @@ export default function (pi: ExtensionAPI) {
     })),
   });
 
+  pi.on("before_provider_headers", (event, ctx) => {
+    if (ctx.model?.provider !== PROVIDER_ID) return;
+    applyOpenCodeSessionHeaders(event.headers, ctx.sessionManager.getSessionId());
+  });
+
   pi.on("session_start", async (_event, ctx) => {
     revalidateAbort?.abort();
     revalidateAbort = new AbortController();