@columenlabs/lumen-ai 0.1.278 → 0.1.279

package/dist/agent/models.js

@@ -63,7 +63,16 @@ export function planFetch(base = fetch, name = 'Assistant') {
             ];
         const streamed = body.stream === true;
         body.stream = true;
-        const res = await base(input, { ...init, body: JSON.stringify(body) });
+        // ChatGPT keeps a plan's prompt cache by the request's `session-id` header, not the body's
+        // `prompt_cache_key` (OpenAI's own Codex client: "ChatGPT derives cache affinity from the
+        // Responses session-id header", codex-rs/core/src/client.rs). Without it each call of a plan
+        // went to a fresh cache and read 0 (live 2026-11-14), while the same calls on a key read 92%.
+        const headers = new Headers(init?.headers);
+        if (typeof body.prompt_cache_key === 'string' && !headers.has('session-id')) {
+            headers.set('session-id', body.prompt_cache_key);
+            headers.set('thread-id', body.prompt_cache_key);
+        }
+        const res = await base(input, { ...init, headers, body: JSON.stringify(body) });
         // A refusal is logged with OpenAI's own reason; the person sees a plain sentence.
         if (!res.ok)
             console.error(JSON.stringify({

package/dist/assistant/engine.js

@@ -1211,7 +1211,11 @@ export class Assistant {
                 ...(plan ? {} : { max_output_tokens: max }),
                 ...(effort &&
                     effort !== 'default' && { reasoning: { effort: effort === 'max' ? 'high' : effort } }),
-            }, signal ? { signal } : {});
+            }, {
+                ...(signal ? { signal } : {}),
+                // A ChatGPT plan keeps its prompt cache by this header, not the body's key (planFetch).
+                ...(cacheKey ? { headers: { 'session-id': cacheKey, 'thread-id': cacheKey } } : {}),
+            });
             for await (const e of stream) {
                 if (e.type === 'response.output_text.delta' && typeof e.delta === 'string')
                     text += e.delta;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@columenlabs/lumen-ai",
-  "version": "0.1.278",
+  "version": "0.1.279",
   "description": "Lumen's assistant and agent parts: a chat assistant's server (Claude tiers, tools over MCP, the web, hand off, spend caps), serving an MCP server from a Worker, read-only tool sets and metered tool calls.",
   "license": "MIT",
   "type": "module",
@@ -57,8 +57,8 @@
     "@anthropic-ai/sdk": "0.133.0",
     "@cloudflare/computer": "0.8.0",
     "@cloudflare/think": "0.23.0",
-    "@columenlabs/lumen-ops": "0.1.278",
-    "@columenlabs/lumen-ui-types": "0.1.278",
+    "@columenlabs/lumen-ops": "0.1.279",
+    "@columenlabs/lumen-ui-types": "0.1.279",
     "agents": "0.32.0",
     "ai": "7.0.162",
     "just-bash": "3.6.0",