openclaw-amem 2.1.0 → 2.1.2

package/README.md

@@ -10,9 +10,9 @@
 
 **Memory for [OpenClaw](https://github.com/openclaw/openclaw) agents that catches its own contradictions.**
 
-A nightly pass re-reads what changed and flags memories that no longer agree with each other — the failure mode every long-lived memory store eventually has. The rest: facts extracted instead of transcripts stored, notes rewritten as new ones arrive, linked into a Zettelkasten-style graph, retrieved with hybrid BM25 + dense search over 2-hop expansion. Memories are scoped per agent (`owner`/`readers`/`writers`) and per subject, both enforced inside the Qdrant query rather than filtered after it.
+A nightly pass re-reads what changed and flags memories that no longer agree with each other. This is the failure mode every long-lived memory store eventually has. The plugin also extracts facts instead of storing transcripts. It rewrites notes as new ones arrive, links them into a Zettelkasten-style graph, and retrieves them with hybrid BM25 + dense search over 2-hop expansion. The plugin scopes memories per agent (`owner`/`readers`/`writers`) and per subject, and enforces both inside the Qdrant query rather than filtering after it.
 
-Runs on a small model — `gpt-4o-mini`, `haiku`, a local Ollama. No Python.
+The plugin runs on a small model — `gpt-4o-mini`, `haiku`, a local Ollama. It does not require Python.
 
 中文走 [jieba](https://github.com/messense/node-jieba) 分词，提示词有完整中文版（`AMEM_PROMPT_LOCALE=zh`），embedding 模型本身多语言。
 
@@ -26,9 +26,9 @@ Requires a local [Qdrant](https://qdrant.tech). Implements [A-MEM](https://arxiv
 
 - 🔄 **Memories evolve** — new facts update/link related memories (EVOLVE / CONFLICT / EXPAND / NEW), not silent overwrite.
 - 🔍 **Hybrid retrieval** — BM25 (Jieba for Chinese) + dense vectors (RRF) + 2-hop graph expansion with a relevance gate.
-- 🧠 **Knowledge vs episodic** — durable knowledge notes skip consolidation & time-decay; topic tags for precise recall.
+- 🧠 **Knowledge vs episodic** — durable knowledge notes skip consolidation & time-decay. Topic tags for precise recall.
 - 🧹 **Self-consolidating** — daily 02:30 in-process merge of semantic duplicates with link cascading.
-- 🔐 **Per-agent isolation** — private by default; explicit `owner`/`readers`/`writers`; Mode A (shared collection) or Mode B (dedicated collection).
+- 🔐 **Per-agent isolation** — private by default. Explicit `owner`/`readers`/`writers`. Mode A (shared collection) or Mode B (dedicated collection).
 - 🀄 **Chinese-optimized** & local embeddings (Transformers.js) — no Python, no external embedding API.
 
 → Full feature list & internals: **[@amemhq/core README](../amem-core)** · **[docs](https://amem.owo.lc)**.
@@ -36,7 +36,7 @@ Requires a local [Qdrant](https://qdrant.tech). Implements [A-MEM](https://arxiv
 ## Requirements
 
 - OpenClaw v2026.4+
-- Node.js 24 (18+ works; 24/26 supported)
+- Node.js 24 (18+ works, 24/26 supported)
 - Qdrant running on `:6333`
 - An LLM: `ANTHROPIC_API_KEY` by default, or any OpenAI-compatible provider — see [LLM provider](#llm-provider)
 - ~1.1 GB of disk for the embedding model, downloaded once on first run
@@ -57,6 +57,16 @@ pnpm --filter openclaw-amem build
 openclaw plugins install --link ./packages/openclaw-amem
 ```
 
+Updating:
+
+```bash
+openclaw plugins update openclaw-amem
+```
+
+The gateway does not load the new build until it restarts. Run `openclaw gateway restart` after the update.
+
+An update rebuilds `node_modules`. The model cache lives there by default, so the plugin re-downloads 2.27 GB unless `AMEM_MODEL_CACHE` points somewhere outside the plugin directory. Set that once and updates cost nothing.
+
 ### 2. Configure `~/.openclaw/openclaw.json`
 
 Add `openclaw-amem` to your allowed plugins and hook it into the `memory` slot:
@@ -81,7 +91,7 @@ Add `openclaw-amem` to your allowed plugins and hook it into the `memory` slot:
 }
 ```
 
-> **⚠️ Memory slot conflict:** If your `openclaw.json` already assigns the `memory` slot to another plugin (e.g. `memory-core`), you **must** change it to `openclaw-amem`. The gateway only loads one plugin per slot — any additional `memory`-kind plugins are **silently skipped**. Set `"memory-core": { "enabled": false }` in `entries` to disable the old plugin.
+> **⚠️ Memory slot conflict:** If your `openclaw.json` already assigns the `memory` slot to another plugin (for example, `memory-core`), you **must** change it to `openclaw-amem`. On OpenClaw 2026.8.1 the slot alone is not enough. The other plugin still loads and takes the `memory_search` tool name, because the gateway loads built-in plugins before installed ones. Memory keeps working through the slot, but the tool an agent calls is then the other plugin's. Set `"memory-core": { "enabled": false }` in `entries` as well. Note that `memory-core` also writes its dream diary, and disabling it stops that too.
 
 ### 3. Restart OpenClaw
 
@@ -92,11 +102,11 @@ openclaw gateway restart
 First run downloads the embedding model (`bge-m3`, 2.27 GB) and caches it. Later
 restarts are instant. If that is too much, the only smaller model worth setting is
 `AMEM_EMBED_MODEL=Xenova/bge-small-zh-v1.5` — 25 MB, Chinese only, and it caps at
-512 tokens. Set it **before** you have memories, since changing it afterwards means
+512 tokens. Set it **before** you have memories, because changing it afterwards means
 a [migration](https://amem.owo.lc/reference/embedding-models#changing-the-model-on-a-store-you-already-have).
 
 Upgrading from 1.x downloads nothing. Your existing memories keep the model that
-built them, and the plugin says so at startup along with the one command that
+built them, and the plugin shows this at startup along with the one command that
 moves them.
 
 ## LLM provider
@@ -104,7 +114,7 @@ moves them.
 The plugin calls an LLM for note construction, linking, and evolution. Pick the backend with `AMEM_LLM_PROVIDER`:
 
 - **`anthropic`** (default) — the Anthropic Messages API. Set `ANTHROPIC_API_KEY`.
-- **`openai`** — the OpenAI Chat Completions API, which every OpenAI-compatible endpoint speaks. Set `AMEM_LLM_PROVIDER=openai`, point `AMEM_LLM_BASE_URL` at the endpoint, and set `AMEM_LLM_API_KEY` (or the standard `OPENAI_API_KEY`). Covers **OpenAI, DeepSeek, OpenRouter, Groq, Together**, and local servers (**Ollama, vLLM, LM Studio** — no key needed). Reasoning models (`o1`, `o3`, `gpt-5`) are handled automatically.
+- **`openai`** — the OpenAI Chat Completions API, which every OpenAI-compatible endpoint speaks. Set `AMEM_LLM_PROVIDER=openai`, point `AMEM_LLM_BASE_URL` at the endpoint, and set `AMEM_LLM_API_KEY` (or the standard `OPENAI_API_KEY`). Covers **OpenAI, DeepSeek, OpenRouter, Groq, Together**, and local servers (**Ollama, vLLM, LM Studio** — no key needed). The plugin handles reasoning models (`o1`, `o3`, `gpt-5`) automatically.
 
 Choose the model with `AMEM_LLM_MODEL`. Full env-var reference and examples: **[amem.owo.lc/reference/configuration](https://amem.owo.lc/reference/configuration)**.
 
@@ -131,13 +141,13 @@ Once installed, the plugin exposes five tools to the agent:
 
 ## Security & data flow
 
-A memory plugin's job is to read configuration from the environment and send memory data to backends **you** control, so registry static scanners flag its `env` + `network` pattern (e.g. ClawHub's `suspicious.env_credential_access`). This is structurally expected for any configurable memory/LLM plugin — the audit outcome is **advisory (`Review`), not `Malicious`**, and VirusTotal reports the bundle clean.
+A memory plugin's job is to read configuration from the environment and send memory data to backends **you** control, so registry static scanners flag its `env` + `network` pattern (for example, ClawHub's `suspicious.env_credential_access`). This is structurally expected for any configurable memory/LLM plugin. The audit outcome is **advisory (`Review`), not `Malicious`**, and VirusTotal reports the bundle clean.
 
 What it actually does — all of it declared in [`openclaw.plugin.json`](openclaw.plugin.json):
 
 - **Environment variables it reads** (its configuration surface, supplied by you): `AMEM_LLM_PROVIDER`, `AMEM_LLM_API_KEY`, `OPENAI_API_KEY`, `AMEM_LLM_BASE_URL`, `AMEM_LLM_MODEL`, `AMEM_COLLECTION`, `AMEM_DATA_DIR`, `AMEM_EVO_COUNTER_PATH`, `AMEM_REVIEW_DIR`, `AMEM_PROMPT_LOCALE`. No credential is bundled, hardcoded, or logged.
 - **Network destinations**: only your **local Qdrant** (`http://localhost:6333`) and your configured **LLM endpoint** (Anthropic by default, or any OpenAI-compatible endpoint via `AMEM_LLM_PROVIDER=openai` + `AMEM_LLM_BASE_URL`). It sends memory text/embeddings there to store and evolve notes — its stated purpose. It does not phone home.
-- **Conversation content** is processed for memory only when you set `hooks.allowConversationAccess: true`. Keep Qdrant and review-output paths scoped to locations you control.
+- The plugin processes **conversation content** for memory only when you set `hooks.allowConversationAccess: true`. Keep Qdrant and review-output paths scoped to locations you control.
 
 ## Development
 

package/dist/index.js

@@ -38,17 +38,22 @@ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: tru
 // ../amem-core/src/config.ts
 function configure(opts) {
   if (opts.dataDir) _dataDir = opts.dataDir;
+  if (opts.warn) _warn = opts.warn;
 }
 function getDataDir() {
   return _dataDir;
 }
-var os, path, _dataDir;
+function warn(msg) {
+  _warn(msg);
+}
+var os, path, _dataDir, _warn;
 var init_config = __esm({
   "../amem-core/src/config.ts"() {
     "use strict";
     os = __toESM(require("os"), 1);
     path = __toESM(require("path"), 1);
     _dataDir = process.env.AMEM_DATA_DIR || path.join(os.homedir(), ".amem");
+    _warn = (msg) => console.warn(msg);
   }
 });
 
@@ -87,6 +92,12 @@ function applyModelPaths(env) {
     env.allowLocalModels = true;
   }
 }
+function humanBytes(n) {
+  if (n >= 1e9) return `${(n / 1e9).toFixed(2)} GB`;
+  if (n >= 1e6) return `${(n / 1e6).toFixed(1)} MB`;
+  if (n >= 1e3) return `${Math.round(n / 1e3)} kB`;
+  return `${n} B`;
+}
 function makeProgressReporter() {
   const lastPct = /* @__PURE__ */ new Map();
   return (e) => {
@@ -95,7 +106,7 @@ function makeProgressReporter() {
     const pct = Math.floor(e.progress / 10) * 10;
     if (lastPct.get(e.file) === pct) return;
     lastPct.set(e.file, pct);
-    const size = e.total ? ` of ${(e.total / 1e9).toFixed(2)} GB` : "";
+    const size = e.total ? ` of ${humanBytes(e.total)}` : "";
     console.log(`[amem] downloading ${e.file}: ${pct}%${size}`);
   };
 }
@@ -294,7 +305,7 @@ async function ensureCollection(collectionName) {
       if (inUse !== null && inUse !== wanted) throw new MixedEmbeddingModelsError(col, wanted, inUse);
       pinEmbeddingModel(wanted);
       if (wanted === LEGACY_DEFAULT_EMBEDDING_MODEL) {
-        console.warn(
+        warn(
           `[amem] "${col}" is on ${LEGACY_DEFAULT_EMBEDDING_MODEL} (${LEGACY_DEFAULT_DIM}-dim).
 [amem] ${DEFAULT_EMBEDDING_MODEL} reads 8192 tokens where that one stops at 128, so anything longer is being truncated before it reaches the vector.
 [amem] To move: npx --package=@amemhq/core amem-migrate --from-collection ${col}`
@@ -698,7 +709,7 @@ function makeCrud(collectionName, modeBIsolated = false) {
             })
           )
         ]).catch((err) => {
-          console.error(`[amem] retrieval tracking patch failed: ${err.message}`);
+          warn(`[amem] retrieval tracking patch failed: ${err.message}`);
         });
         for (const r of queryResults) {
           r.note.retrieval_count = (r.note.retrieval_count || 0) + 1;
@@ -818,6 +829,7 @@ var init_storage = __esm({
   "../amem-core/src/storage.ts"() {
     "use strict";
     init_auth();
+    init_config();
     init_embedding();
     QDRANT_URL = "http://localhost:6333";
     getCollection = () => process.env.AMEM_COLLECTION || "amem_notes";
@@ -1128,7 +1140,7 @@ function configureLlm(cfg) {
 function warnOnce(key, message) {
   if (_warned.has(key)) return;
   _warned.add(key);
-  console.error(message);
+  warn(message);
 }
 function resolveProvider(role = "fast") {
   const raw = role === "strong" ? process.env.AMEM_LLM_STRONG_PROVIDER || _override.strong?.provider || void 0 : void 0;
@@ -1197,9 +1209,11 @@ async function llmCall(prompt, maxTokens = 500, role = "fast") {
   const isThinking = model.includes("gemini") || model.includes("pro-agent");
   const effectiveMaxTokens = isThinking ? Math.max(maxTokens * 8, 4e3) : maxTokens;
   try {
-    return provider === "openai" ? await openaiCall(prompt, model, effectiveMaxTokens, baseURL) : await anthropicCall(prompt, model, effectiveMaxTokens, baseURL);
+    const text = provider === "openai" ? await openaiCall(prompt, model, effectiveMaxTokens, baseURL) : await anthropicCall(prompt, model, effectiveMaxTokens, baseURL);
+    if (text === null) warn(`[amem] ${provider} answered with no text (model ${model})`);
+    return text;
   } catch (e) {
-    console.error(`[amem] LLM call failed: ${e.message}`);
+    warn(`[amem] LLM call failed: ${e.message}`);
     return null;
   }
 }
@@ -1288,7 +1302,8 @@ confidence guide (Story 27):
 
 Text: ${content}`;
   const raw = await llmCall(prompt, 400);
-  if (!raw)
+  if (!raw) {
+    warn("[amem] note construction got nothing back; storing with no keywords, tags or context");
     return {
       keywords: [],
       tags: [],
@@ -1298,6 +1313,7 @@ Text: ${content}`;
       topics: [],
       confidence: "medium"
     };
+  }
   try {
     const data = parseJsonLoose(raw);
     const rawCategory = typeof data.category === "string" ? data.category : "General";
@@ -1316,7 +1332,7 @@ Text: ${content}`;
       confidence
     };
   } catch (e) {
-    console.error(`[amem] Note construction parse failed: ${e.message}`);
+    warn(`[amem] Note construction parse failed: ${e.message}`);
     return {
       keywords: [],
       tags: [],
@@ -1345,9 +1361,15 @@ async function llmCrudDecision(userText, assistantText, existingMemories) {
     const raw = await llmCall(prompt, 400, resolveCrudRole());
     if (!raw) return [];
     const match = stripReasoning(raw).match(/\[.*\]/s);
-    if (!match) return [];
+    if (!match) {
+      warn("[amem] llmCrudDecision found no array in the response; nothing from this turn is stored");
+      return [];
+    }
     const parsed = JSON.parse(match[0]);
-    if (!Array.isArray(parsed)) return [];
+    if (!Array.isArray(parsed)) {
+      warn("[amem] llmCrudDecision parsed a non-array; nothing from this turn is stored");
+      return [];
+    }
     const ops = [];
     for (const item of parsed) {
       if (!item || typeof item !== "object") continue;
@@ -1366,7 +1388,7 @@ async function llmCrudDecision(userText, assistantText, existingMemories) {
     }
     return ops.slice(0, 3);
   } catch (e) {
-    console.error(`[amem] llmCrudDecision failed: ${e.message}`);
+    warn(`[amem] llmCrudDecision failed: ${e.message}`);
     return [];
   }
 }
@@ -1376,20 +1398,26 @@ async function llmShouldMerge(contentA, contentB) {
   if (!raw) return { shouldMerge: false };
   try {
     const data = parseJsonLoose(raw);
-    if (typeof data.shouldMerge !== "boolean") return { shouldMerge: false };
+    if (typeof data.shouldMerge !== "boolean") {
+      warn("[amem] llmShouldMerge got no boolean verdict; treating the pair as distinct");
+      return { shouldMerge: false };
+    }
     if (data.shouldMerge && typeof data.merged === "string") {
       return { shouldMerge: true, merged: data.merged };
     }
     return { shouldMerge: false };
   } catch (e) {
-    console.error(`[amem] llmShouldMerge parse failed: ${e.message}`);
+    warn(`[amem] llmShouldMerge parse failed: ${e.message}`);
     return { shouldMerge: false };
   }
 }
 async function llmEvolutionJudge(oldContent, newContent) {
   const prompt = t.evolutionJudge(oldContent, newContent);
   const raw = await llmCall(prompt, 300, "strong");
-  if (!raw) return { type: "NEW" };
+  if (!raw) {
+    warn("[amem] evolution judge got nothing back; defaulting the pair to NEW");
+    return { type: "NEW" };
+  }
   try {
     const data = parseJsonLoose(raw);
     const type = VALID_EVOLUTION_TYPES.has(data.type) ? data.type : "NEW";
@@ -1398,7 +1426,7 @@ async function llmEvolutionJudge(oldContent, newContent) {
       mergedContent: typeof data.mergedContent === "string" ? data.mergedContent : void 0
     };
   } catch (e) {
-    console.error(`[amem] llmEvolutionJudge parse failed: ${e.message}`);
+    warn(`[amem] llmEvolutionJudge parse failed: ${e.message}`);
     return { type: "NEW" };
   }
 }
@@ -1437,7 +1465,7 @@ ${linkedStr}`;
       tagsToUpdate: Array.isArray(data.tags_to_update) ? data.tags_to_update.map(String) : []
     };
   } catch (e) {
-    console.error(`[amem] Evolution parse failed: ${e.message}`);
+    warn(`[amem] Evolution parse failed: ${e.message}`);
     return { tags: null, context: null, shouldStrengthen: false, suggestedConnections: [], tagsToUpdate: [] };
   }
 }
@@ -1449,9 +1477,15 @@ async function llmConflictScan(contents) {
     if (!raw) return [];
     const cleaned = stripReasoning(raw);
     const match = cleaned.match(/\[[\s\S]*\]/);
-    if (!match) return [];
+    if (!match) {
+      warn("[amem] llmConflictScan found no array in the response; this sweep reports no pairs");
+      return [];
+    }
     const parsed = JSON.parse(match[0]);
-    if (!Array.isArray(parsed)) return [];
+    if (!Array.isArray(parsed)) {
+      warn("[amem] llmConflictScan parsed a non-array; this sweep reports no pairs");
+      return [];
+    }
     const pairs = [];
     const seen = /* @__PURE__ */ new Set();
     for (const item of parsed) {
@@ -1475,7 +1509,7 @@ async function llmConflictScan(contents) {
     }
     return pairs;
   } catch (e) {
-    console.error(`[amem] llmConflictScan failed: ${e.message}`);
+    warn(`[amem] llmConflictScan failed: ${e.message}`);
     return [];
   }
 }
@@ -1486,6 +1520,7 @@ var init_llm = __esm({
     import_sdk = __toESM(require("@anthropic-ai/sdk"), 1);
     import_openai = __toESM(require("openai"), 1);
     init_prompts();
+    init_config();
     _override = {};
     _warned = /* @__PURE__ */ new Set();
     DEFAULT_TIMEOUT_MS = 3e4;
@@ -1817,7 +1852,7 @@ async function addMemory(content, agentId = "main", opts) {
       }
     }
   } catch (e) {
-    console.error(`[warn] Link/Evolution phase failed: ${e.message}`);
+    warn(`[amem] link/evolution phase failed: ${e.message}`);
   }
   console.log(`[done] Note added: ${note.id}`);
   return note.id;
@@ -1942,7 +1977,12 @@ async function searchMemory(query, topK = 5, agentId = "main", opts) {
       keywords: note.keywords,
       links: note.links,
       timestamp: note.timestamp,
-      similarity: embSimMap.get(id) ?? bfsSimMap.get(id) ?? 0,
+      // Neither map covers a note that got here on BM25 alone: it was never in
+      // the dense results, and it was not expanded into. That is a real cosine
+      // nobody had measured, not a zero — and reporting 0 made a lexical match
+      // look like the least relevant row in the list. Both vectors are already
+      // in hand, so measuring it is one dot product.
+      similarity: embSimMap.get(id) ?? bfsSimMap.get(id) ?? cosineSimilarity(queryEmbedding, note.embedding),
       rrf: rrfMap.get(id) ?? 0,
       via,
       topics: note.topics ?? [],
@@ -2078,8 +2118,8 @@ async function consolidateMemories(agentId, logger, storageCtx) {
   const ctx = storageCtx ?? defaultCtx();
   const log = {
     info: (msg) => logger ? logger.info(msg) : console.log(msg),
-    warn: (msg) => logger ? logger.warn(msg) : console.warn(msg),
-    error: (msg) => logger ? logger.error(msg) : console.error(msg)
+    warn: (msg) => logger ? logger.warn(msg) : warn(msg),
+    error: (msg) => logger ? logger.error(msg) : warn(msg)
   };
   log.info(`[Consolidation] Starting consolidation for agentId: ${agentId}`);
   const rawNotes = await ctx.listNotes(agentId);
@@ -2483,7 +2523,7 @@ async function migrateCollection(opts) {
   const refreshFields = opts.refreshFields !== false;
   const dryRun = opts.dryRun !== false;
   const log = opts.logger?.info ?? ((m) => console.log(m));
-  const warn = opts.logger?.warn ?? ((m) => console.warn(m));
+  const warn2 = opts.logger?.warn ?? warn;
   if (from === to) throw new Error(`migrate: source and target are the same collection ("${from}")`);
   const model = getEmbeddingModel();
   const targetDim = await getEmbeddingDim();
@@ -2551,7 +2591,7 @@ async function migrateCollection(opts) {
         if (!note.context) note.context = built.context;
         refreshed++;
       } catch (e) {
-        warn(`[migrate] re-extract failed for ${note.id.slice(0, 8)} \u2014 keeping as-is: ${e.message}`);
+        warn2(`[migrate] re-extract failed for ${note.id.slice(0, 8)} \u2014 keeping as-is: ${e.message}`);
       }
     }
     const point = noteToPoint({ ...note, embedding: await encode(buildEmbedText(note)) });
@@ -2564,7 +2604,7 @@ async function migrateCollection(opts) {
   await flush();
   const finalCount = await countPointsRaw(to);
   if (finalCount !== notes.length) {
-    warn(`[migrate] target holds ${finalCount} point(s) but the source had ${notes.length} \u2014 check before switching`);
+    warn2(`[migrate] target holds ${finalCount} point(s) but the source had ${notes.length} \u2014 check before switching`);
   }
   log(`[migrate] done: ${migrated} written, ${alreadyDone.size} already present, ${refreshed} re-extracted.`);
   return {
@@ -2619,6 +2659,7 @@ var init_migrate = __esm({
     init_embedding();
     init_llm();
     init_memory();
+    init_config();
   }
 });
 
@@ -2777,7 +2818,10 @@ function register(api) {
   const pluginId = api.id ?? "openclaw-amem";
   const convBlocked = isConvAccessBlocked(api.config, pluginId);
   if (convBlocked) logger.warn(BLOCKED_WARNING_LOG);
-  configure({ dataDir: path5.join(os2.homedir(), ".openclaw") });
+  configure({
+    dataDir: path5.join(os2.homedir(), ".openclaw"),
+    warn: (msg) => logger.warn(msg)
+  });
   const hasStrong = !!(pluginConfig.llmStrongProvider || pluginConfig.llmStrongModel || pluginConfig.llmStrongBaseURL);
   if (pluginConfig.llmProvider || pluginConfig.llmModel || pluginConfig.llmBaseURL || pluginConfig.llmCrudRole || hasStrong) {
     configureLlm({
@@ -3127,7 +3171,12 @@ ${text}${hookWarning}` }],
           const msgs = event.messages || [];
           const lastUser = [...msgs].reverse().find((m) => m.role === "user");
           const lastAssistant = [...msgs].reverse().find((m) => m.role === "assistant");
-          if (!lastUser || !lastAssistant) return;
+          if (!lastUser || !lastAssistant) {
+            logger.info(
+              `openclaw-amem: agent_end skipped \u2014 no ${!lastUser ? "user" : "assistant"} message in ${msgs.length} messages`
+            );
+            return;
+          }
           const userText = typeof lastUser.content === "string" ? lastUser.content : lastUser.content.filter((b) => b.type === "text").map((b) => b.text).join(" ");
           const assistantText = typeof lastAssistant.content === "string" ? lastAssistant.content : lastAssistant.content.filter((b) => b.type === "text").map((b) => b.text).join(" ");
           if (!userText || !shouldProcessTurn2(userText)) return;

package/openclaw.plugin.json

@@ -3,7 +3,7 @@
   "name": "amem",
   "description": "Catches memories that contradict each other. Notes rewrite themselves as new ones arrive, link into a graph, and stay separated per agent and per person. Runs on a small model. Requires a local Qdrant. 中文走 jieba 分词，提示词有中文版。",
   "icon": "https://amem.owo.lc/logo.png",
-  "version": "2.1.0",
+  "version": "2.1.2",
   "kind": "memory",
   "contracts": {
     "tools": [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "openclaw-amem",
-  "version": "2.1.0",
+  "version": "2.1.2",
   "description": "Catches memories that contradict each other. Notes rewrite themselves as new ones arrive, link into a graph, and stay separated per agent and per person. Runs on a small model. Requires a local Qdrant. 中文走 jieba 分词，提示词有中文版。",
   "license": "MIT",
   "main": "dist/index.js",
@@ -16,7 +16,7 @@
     }
   },
   "dependencies": {
-    "@anthropic-ai/sdk": "^0.112.5",
+    "@anthropic-ai/sdk": "^0.122.0",
     "@huggingface/transformers": "^4.2.0",
     "@node-rs/jieba": "^2.0.1",
     "@qdrant/js-client-rest": "^1.18.0",
@@ -26,7 +26,6 @@
   "devDependencies": {
     "@eslint/js": "^10.0.1",
     "@types/node": "^26.1.1",
-    "@types/uuid": "^11.0.0",
     "eslint": "^10.7.0",
     "prettier": "^3.9.6",
     "tsup": "^8.4.0",