@poncho-ai/cli 0.24.0 → 0.24.1

package/.turbo/turbo-build.log

@@ -1,5 +1,5 @@
 
-> @poncho-ai/cli@0.24.0 build /home/runner/work/poncho-ai/poncho-ai/packages/cli
+> @poncho-ai/cli@0.24.1 build /home/runner/work/poncho-ai/poncho-ai/packages/cli
 > tsup src/index.ts src/cli.ts --format esm --dts
 
 CLI Building entry: src/cli.ts, src/index.ts
@@ -8,11 +8,11 @@
 CLI Target: es2022
 ESM Build start
 ESM dist/cli.js                          94.00 B
-ESM dist/run-interactive-ink-GA5V5ATO.js 56.74 KB
 ESM dist/index.js                        857.00 B
-ESM dist/chunk-CX2JHBBS.js               399.73 KB
-ESM ⚡️ Build success in 65ms
+ESM dist/run-interactive-ink-IEB4MZ2C.js 56.74 KB
+ESM dist/chunk-3ETNDULB.js               400.75 KB
+ESM ⚡️ Build success in 57ms
 DTS Build start
-DTS ⚡️ Build success in 3966ms
+DTS ⚡️ Build success in 3876ms
 DTS dist/cli.d.ts   20.00 B
 DTS dist/index.d.ts 3.59 KB

package/CHANGELOG.md

@@ -1,5 +1,16 @@
 # @poncho-ai/cli
 
+## 0.24.1
+
+### Patch Changes
+
+- [`096953d`](https://github.com/cesr/poncho-ai/commit/096953d5a64a785950ea0a7f09e2183e481afd29) Thanks [@cesr](https://github.com/cesr)! - Improve time-to-first-token by lazy-loading the recall corpus
+
+  The recall corpus (past conversation summaries) is now fetched on-demand only when the LLM invokes the `conversation_recall` tool, instead of blocking every message with ~1.3s of upfront I/O. Also adds batch `mget` support to Upstash/Redis/DynamoDB conversation stores, parallelizes memory fetch with skill refresh, debounces skill refresh in dev mode, and caches message conversions across multi-step runs.
+
+- Updated dependencies [[`096953d`](https://github.com/cesr/poncho-ai/commit/096953d5a64a785950ea0a7f09e2183e481afd29)]:
+  - @poncho-ai/harness@0.22.1
+
 ## 0.24.0
 
 ### Minor Changes

package/dist/{chunk-CX2JHBBS.js → chunk-3ETNDULB.js}

@@ -9331,7 +9331,8 @@ data: ${JSON.stringify(data)}
       response.writeHead(200, {
         "Content-Type": "text/event-stream",
         "Cache-Control": "no-cache",
-        Connection: "keep-alive"
+        Connection: "keep-alive",
+        "X-Accel-Buffering": "no"
       });
       const historyMessages = [...conversation.messages];
       const preRunMessages = [...conversation.messages];
@@ -9392,7 +9393,9 @@ data: ${JSON.stringify(data)}
       try {
         conversation.messages = [...historyMessages, { role: "user", content: userContent }];
         conversation.updatedAt = Date.now();
-        await conversationStore.update(conversation);
+        conversationStore.update(conversation).catch((err) => {
+          console.error("[poncho] Failed to persist user turn:", err);
+        });
         const persistDraftAssistantTurn = async () => {
           const draftSections = [
             ...sections.map((section) => ({
@@ -9425,18 +9428,32 @@ data: ${JSON.stringify(data)}
           conversation.updatedAt = Date.now();
           await conversationStore.update(conversation);
         };
-        const recallCorpus = (await conversationStore.list(ownerId)).filter((item) => item.conversationId !== conversationId && !item.parentConversationId).slice(0, 20).map((item) => ({
-          conversationId: item.conversationId,
-          title: item.title,
-          updatedAt: item.updatedAt,
-          content: item.messages.slice(-6).map((message) => `${message.role}: ${typeof message.content === "string" ? message.content : getTextContent(message)}`).join("\n").slice(0, 2e3)
-        })).filter((item) => item.content.length > 0);
+        let cachedRecallCorpus;
+        const lazyRecallCorpus = async () => {
+          if (cachedRecallCorpus) return cachedRecallCorpus;
+          const _rc0 = performance.now();
+          let recallConversations;
+          if (typeof conversationStore.listSummaries === "function") {
+            const recallSummaries = (await conversationStore.listSummaries(ownerId)).filter((s) => s.conversationId !== conversationId && !s.parentConversationId).slice(0, 20);
+            recallConversations = (await Promise.all(recallSummaries.map((s) => conversationStore.get(s.conversationId)))).filter((c) => c != null);
+          } else {
+            recallConversations = (await conversationStore.list(ownerId)).filter((item) => item.conversationId !== conversationId && !item.parentConversationId).slice(0, 20);
+          }
+          cachedRecallCorpus = recallConversations.map((item) => ({
+            conversationId: item.conversationId,
+            title: item.title,
+            updatedAt: item.updatedAt,
+            content: item.messages.slice(-6).map((message) => `${message.role}: ${typeof message.content === "string" ? message.content : getTextContent(message)}`).join("\n").slice(0, 2e3)
+          })).filter((item) => item.content.length > 0);
+          console.info(`[poncho] recall corpus fetched lazily (${cachedRecallCorpus.length} items, ${(performance.now() - _rc0).toFixed(1)}ms)`);
+          return cachedRecallCorpus;
+        };
         for await (const event of harness.runWithTelemetry({
           task: messageText,
           conversationId,
           parameters: {
             ...bodyParameters ?? {},
-            __conversationRecallCorpus: recallCorpus,
+            __conversationRecallCorpus: lazyRecallCorpus,
             __activeConversationId: conversationId,
             __ownerId: ownerId
           },
@@ -10077,7 +10094,7 @@ var runInteractive = async (workingDir, params) => {
   await harness.initialize();
   const identity = await ensureAgentIdentity2(workingDir);
   try {
-    const { runInteractiveInk } = await import("./run-interactive-ink-GA5V5ATO.js");
+    const { runInteractiveInk } = await import("./run-interactive-ink-IEB4MZ2C.js");
     await runInteractiveInk({
       harness,
       params,

package/dist/cli.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   main
-} from "./chunk-CX2JHBBS.js";
+} from "./chunk-3ETNDULB.js";
 
 // src/cli.ts
 void main();

package/dist/index.js

@@ -23,7 +23,7 @@ import {
   runTests,
   startDevServer,
   updateAgentGuidance
-} from "./chunk-CX2JHBBS.js";
+} from "./chunk-3ETNDULB.js";
 export {
   addSkill,
   buildCli,

package/dist/{run-interactive-ink-GA5V5ATO.js → run-interactive-ink-IEB4MZ2C.js}

@@ -2,7 +2,7 @@ import {
   consumeFirstRunIntro,
   inferConversationTitle,
   resolveHarnessEnvironment
-} from "./chunk-CX2JHBBS.js";
+} from "./chunk-3ETNDULB.js";
 
 // src/run-interactive-ink.ts
 import * as readline from "readline";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@poncho-ai/cli",
-  "version": "0.24.0",
+  "version": "0.24.1",
   "description": "CLI for building and deploying AI agents",
   "repository": {
     "type": "git",
@@ -27,9 +27,9 @@
     "react": "^19.2.4",
     "react-devtools-core": "^6.1.5",
     "yaml": "^2.8.1",
-    "@poncho-ai/harness": "0.22.0",
     "@poncho-ai/messaging": "0.5.1",
-    "@poncho-ai/sdk": "1.5.0"
+    "@poncho-ai/sdk": "1.5.0",
+    "@poncho-ai/harness": "0.22.1"
   },
   "devDependencies": {
     "@types/busboy": "^1.5.4",

package/src/index.ts

@@ -3526,6 +3526,7 @@ export const createRequestHandler = async (options?: {
         "Content-Type": "text/event-stream",
         "Cache-Control": "no-cache",
         Connection: "keep-alive",
+        "X-Accel-Buffering": "no",
       });
       const historyMessages = [...conversation.messages];
       const preRunMessages = [...conversation.messages];
@@ -3584,10 +3585,14 @@ export const createRequestHandler = async (options?: {
       });
 
       try {
-        // Persist the user turn immediately so refreshing mid-run keeps chat context.
+        // Persist the user turn so refreshing mid-run keeps chat context.
+        // Fire-and-forget: the write chain in the store serializes file ops,
+        // and persistDraftAssistantTurn won't run until LLM events arrive.
         conversation.messages = [...historyMessages, { role: "user", content: userContent }];
         conversation.updatedAt = Date.now();
-        await conversationStore.update(conversation);
+        conversationStore.update(conversation).catch((err) => {
+          console.error("[poncho] Failed to persist user turn:", err);
+        });
 
         const persistDraftAssistantTurn = async (): Promise<void> => {
           const draftSections: Array<{ type: "text" | "tools"; content: string | string[] }> = [
@@ -3626,27 +3631,45 @@ export const createRequestHandler = async (options?: {
           await conversationStore.update(conversation);
         };
 
-        const recallCorpus = (await conversationStore.list(ownerId))
-          .filter((item) => item.conversationId !== conversationId && !item.parentConversationId)
-          .slice(0, 20)
-          .map((item) => ({
-            conversationId: item.conversationId,
-            title: item.title,
-            updatedAt: item.updatedAt,
-            content: item.messages
-              .slice(-6)
-              .map((message) => `${message.role}: ${typeof message.content === "string" ? message.content : getTextContent(message)}`)
-              .join("\n")
-              .slice(0, 2000),
-          }))
-          .filter((item) => item.content.length > 0);
+        let cachedRecallCorpus: unknown[] | undefined;
+        const lazyRecallCorpus = async () => {
+          if (cachedRecallCorpus) return cachedRecallCorpus;
+          const _rc0 = performance.now();
+          let recallConversations: Conversation[];
+          if (typeof conversationStore.listSummaries === "function") {
+            const recallSummaries = (await conversationStore.listSummaries(ownerId))
+              .filter((s) => s.conversationId !== conversationId && !s.parentConversationId)
+              .slice(0, 20);
+            recallConversations = (
+              await Promise.all(recallSummaries.map((s) => conversationStore.get(s.conversationId)))
+            ).filter((c): c is NonNullable<typeof c> => c != null);
+          } else {
+            recallConversations = (await conversationStore.list(ownerId))
+              .filter((item) => item.conversationId !== conversationId && !item.parentConversationId)
+              .slice(0, 20);
+          }
+          cachedRecallCorpus = recallConversations
+            .map((item) => ({
+              conversationId: item.conversationId,
+              title: item.title,
+              updatedAt: item.updatedAt,
+              content: item.messages
+                .slice(-6)
+                .map((message) => `${message.role}: ${typeof message.content === "string" ? message.content : getTextContent(message)}`)
+                .join("\n")
+                .slice(0, 2000),
+            }))
+            .filter((item) => item.content.length > 0);
+          console.info(`[poncho] recall corpus fetched lazily (${cachedRecallCorpus.length} items, ${(performance.now() - _rc0).toFixed(1)}ms)`);
+          return cachedRecallCorpus;
+        };
 
         for await (const event of harness.runWithTelemetry({
           task: messageText,
           conversationId,
           parameters: {
             ...(bodyParameters ?? {}),
-            __conversationRecallCorpus: recallCorpus,
+            __conversationRecallCorpus: lazyRecallCorpus,
             __activeConversationId: conversationId,
             __ownerId: ownerId,
           },
@@ -3675,7 +3698,6 @@ export const createRequestHandler = async (options?: {
             runCancelled = true;
           }
           if (event.type === "model:chunk") {
-            // If we have tools accumulated and text starts again, push tools as a section
             if (currentTools.length > 0) {
               sections.push({ type: "tools", content: currentTools });
               currentTools = [];