From 628691d2dfc1d7f1b847bfa277644f29df9c2120 Mon Sep 17 00:00:00 2001
From: jules <cloudtech@juleslive.net>
Date: Sun, 3 May 2026 22:18:18 +1000
Subject: [PATCH] ai: shake-out fixes from first end-to-end search_kb run
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Three problems surfaced when driving search_kb / read_chunk through a
real DeepSeek chat for the first time:

1. vite.config.ts: minisearch wasn't in `aliasedDeps`, so when
   @crema/lexical-rag-ui imports it, Vite couldn't resolve the bare
   specifier from the sibling lib's location. Added it next to the
   other shared deps. tsconfig paths alone is not enough — the Vite
   alias is what propagates resolution to sibling-lib code.

2. ai.tsx reindexKB: was using `toast.show?.()` which doesn't exist on
   useToast()'s API. Optional chaining silently no-op'd, so the button
   click ran the fetch but produced zero UI feedback (success or
   failure). Switched to the actual API: toast.info / toast.success /
   toast.error. Added a console.error in the catch arm so the
   underlying exception is visible in DevTools when something does go
   wrong.

3. ai.tsx MAX_TOOL_ITERATIONS: cap was 3, which is too tight for
   agentic search→read→search loops on real questions. Bumped to 6.
   More importantly, when the cap IS reached, the runner now
   synthesises tool-error messages for each pending tool_call and
   continues the chat — instead of silently dropping them, which left
   the conversation with an assistant.tool_calls turn but no matching
   tool messages. DeepSeek (and the OpenAI spec) reject that
   conversation with 400 ("insufficient tool messages following
   tool_calls"), poisoning the thread. Now the model gets a clean
   "max iterations reached" signal and produces a final answer.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 app/routes/ai.tsx | 43 +++++++++++++++++++++++++++++++------------
 vite.config.ts    |  1 +
 2 files changed, 32 insertions(+), 12 deletions(-)

diff --git a/app/routes/ai.tsx b/app/routes/ai.tsx
index 707ceeb..10e5b28 100644
--- a/app/routes/ai.tsx
+++ b/app/routes/ai.tsx
@@ -127,9 +127,10 @@ async function reindexKB(
       window.sessionStorage.getItem("arcadia_access_token")) ||
     "dev"
   const url = `${baseUrl}/index/${encodeURIComponent(corpus)}/build`
-  toast.show?.({
-    title: "Reindexing…",
-    description: `Rebuilding corpus '${corpus}'.`,
+  // Use useToast's actual API (see lib-notification-ui): tone helpers
+  // success/error/info/warning, plus generic toast({ title, description, tone }).
+  toast.info(`Reindexing '${corpus}'…`, {
+    description: `POST ${url}`,
   })
   try {
     const res = await fetch(url, {
@@ -140,16 +141,14 @@ async function reindexKB(
       throw new Error(`HTTP ${res.status}: ${await res.text()}`)
     }
     const out = (await res.json()) as { chunk_count: number; built_at: string }
-    toast.show?.({
-      title: "Reindex complete",
+    toast.success("Reindex complete", {
       description: `${out.chunk_count} chunks indexed for '${corpus}'.`,
     })
   } catch (err) {
-    toast.show?.({
-      title: "Reindex failed",
-      description: err instanceof Error ? err.message : String(err),
-      tone: "error",
-    })
+    const msg = err instanceof Error ? err.message : String(err)
+    // eslint-disable-next-line no-console
+    console.error("[reindexKB]", err)
+    toast.error("Reindex failed", { description: msg })
   }
 }
 
@@ -818,7 +817,7 @@ function ChatSurface({
       }
     }
   }, [activeAgent, messages, isStreaming, agentHistory])
-  const MAX_TOOL_ITERATIONS = 3
+  const MAX_TOOL_ITERATIONS = 6
   const [pendingConfirm, setPendingConfirm] = useState<{
     /** Message index that emitted the write calls. */
     afterIndex: number
@@ -856,7 +855,27 @@ function ChatSurface({
       toolIterationsRef.current = 0
       return
     }
-    if (toolIterationsRef.current >= MAX_TOOL_ITERATIONS) return
+    if (toolIterationsRef.current >= MAX_TOOL_ITERATIONS) {
+      // Cap reached. Synthesize a tool-error response for each pending
+      // call so the conversation stays valid (every assistant tool_calls
+      // turn must be followed by tool messages — otherwise the next
+      // request to DeepSeek 400s on "insufficient tool messages"). Then
+      // continue the chat so the model writes a final answer with what
+      // it has, or apologizes for not finishing.
+      const cappedMessages = calls.map((c) => ({
+        role: "tool" as const,
+        content: JSON.stringify({
+          error: `MAX_TOOL_ITERATIONS (${MAX_TOOL_ITERATIONS}) reached — please answer with what you have so far.`,
+        }),
+        toolCallId: c.id,
+        name: c.name,
+      }))
+      void continueChat(cappedMessages, {
+        system: systemPrompt,
+        tools: getOpenAITools(),
+      })
+      return
+    }
     toolIterationsRef.current += 1
     void (async () => {
       const { reads, writes } = classifyCalls(calls)
diff --git a/vite.config.ts b/vite.config.ts
index b1fa382..6cd8ec8 100644
--- a/vite.config.ts
+++ b/vite.config.ts
@@ -104,6 +104,7 @@ const aliasedDeps = [
   "@tiptap/extension-link",
   "@tiptap/extension-placeholder",
   "@tiptap/extension-image",
+  "minisearch",
 ]
 const sharedDepAliases = Object.fromEntries(
   aliasedDeps.map((name) => [name, resolvePath(nodeModules, name)]),