@columenlabs/lumen-ai 0.1.277 → 0.1.279

package/dist/agent/agent.d.ts

@@ -177,6 +177,8 @@ export interface AskInput {
     outcomes?: AssistantScreenOutcome[];
     /** The person kept this message in this chat after a fresh chat was offered (`offerFresh`). */
     stay?: boolean;
+    /** How many images went with the message (set by the agent: the images go in its parts). */
+    image_count?: number;
 }
 /** A quote as sent, checked: its text cut to `ASSISTANT_QUOTE_MAX`, and whose message it came from. */
 export declare function quoteOf(q: unknown): AssistantQuote | undefined;

package/dist/agent/agent.js

@@ -413,7 +413,12 @@ export class LumenAgent extends Think {
         this.#steered = [];
         // The images go in the message's parts; the turn's plan reads the rest.
         const { images: _images, ...asked } = input;
-        this.#input = { ...asked, ...(map ? { screen: map } : {}) };
+        const pictures = imageParts(input.images);
+        this.#input = {
+            ...asked,
+            ...(map ? { screen: map } : {}),
+            ...(pictures.length ? { image_count: pictures.length } : {}),
+        };
         if (!map)
             delete this.#input.screen;
         let failed = false;
@@ -424,7 +429,7 @@ export class LumenAgent extends Think {
                 parts: [
                     ...(quote ? [{ type: 'text', text: quoteLead(quote) }] : []),
                     { type: 'text', text: input.message },
-                    ...imageParts(input.images),
+                    ...pictures,
                 ],
                 metadata: meta,
             }, {

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.d.ts

@@ -278,6 +278,8 @@ export interface AgentTurnInput {
     tier?: string;
     /** The browser's IANA zone, for the bearings. */
     time_zone?: string;
+    /** How many images went with the message, counted in the chat's usage (`UsageEntry.images`). */
+    image_count?: number;
     /** When the chat's last message was, for "since you last asked". */
     last_at?: number;
     /**

package/dist/assistant/engine.js

@@ -549,6 +549,9 @@ export class Assistant {
                     id: `chat:${input.chat}`,
                     kind: 'chat',
                     title: message.split('\n')[0] || 'Chat',
+                    turns: 1,
+                    images: input.image_count ?? 0,
+                    refs: input.refs?.length ?? 0,
                     own_usd: ownKeyUsd(paying, onPayer),
                     own_plan_calls: ownPlanCalls(paying, onPayer),
                     own_plan_usd: ownPlanUsd(paying, onPayer),
@@ -1208,7 +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/dist/assistant/state.d.ts

@@ -316,6 +316,11 @@ export declare class State {
         kind: UsageEntry['kind'];
         task?: string;
         title: string;
+        /** Messages the person sent with this use (1 a chat turn; a run's calls send none). */
+        turns?: number;
+        /** Images and picks (refs) that went with those messages. */
+        images?: number;
+        refs?: number;
         own_usd?: number;
         own_plan_calls?: number;
         own_plan_usd?: number;

package/dist/assistant/state.js

@@ -429,6 +429,9 @@ export class State {
             const own = (prev?.own_usd ?? 0) + (e.own_usd ?? 0);
             const plan = (prev?.own_plan_calls ?? 0) + (e.own_plan_calls ?? 0);
             const planUsd = (prev?.own_plan_usd ?? 0) + (e.own_plan_usd ?? 0);
+            const turns = (prev?.turns ?? 0) + (e.turns ?? 0);
+            const images = (prev?.images ?? 0) + (e.images ?? 0);
+            const refs = (prev?.refs ?? 0) + (e.refs ?? 0);
             entries.unshift({
                 id: e.id,
                 kind: e.kind,
@@ -444,6 +447,9 @@ export class State {
                 ...(own > 0 ? { own_usd: own } : {}),
                 ...(plan > 0 ? { own_plan_calls: plan } : {}),
                 ...(planUsd > 0 ? { own_plan_usd: planUsd } : {}),
+                ...(turns > 0 ? { turns } : {}),
+                ...(images > 0 ? { images } : {}),
+                ...(refs > 0 ? { refs } : {}),
                 parts: merged,
             });
             return { entries: entries.slice(0, USAGE_KEPT) };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@columenlabs/lumen-ai",
-  "version": "0.1.277",
+  "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": "0.1.277",
-    "@columenlabs/lumen-ui-types": "0.1.277",
+    "@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",