@nowledge/openclaw-nowledge-mem 0.6.8 → 0.6.9

package/CHANGELOG.md

@@ -2,6 +2,16 @@
 
 All notable changes to the Nowledge Mem OpenClaw plugin will be documented in this file.
 
+## [0.6.9] - 2026-03-05
+
+### Fixed
+
+- **Auto-recall searched with full prompt instead of user message**: `event.prompt` contains the entire conversation history (system prompt + all prior turns), making recalled memories irrelevant as conversation grows. Now uses a tiered query strategy: substantial messages (>=40 chars) search alone; short messages (3-39 chars, likely follow-ups like "explain that more") include the last 3 messages as context for topic grounding; tiny messages (<3 chars) skip recall. Falls back to truncated `event.prompt` (500 chars) only when `event.messages` is unavailable.
+
+### Added
+
+- **`recallMinScore` config option** (0-100, default 0): Minimum relevance score threshold for auto-recalled memories. Set to e.g. 30 to filter out low-confidence results. Configurable via OpenClaw Config UI, config file, or `NMEM_RECALL_MIN_SCORE` env var.
+
 ## [0.6.8] - 2026-02-27
 
 ### Changed

package/CLAUDE.md

@@ -91,6 +91,7 @@ Optional config file at `~/.nowledge-mem/openclaw.json`. Falls through to OpenCl
 | `sessionDigest` | boolean | `true` | `NMEM_SESSION_DIGEST` | Thread capture + LLM distillation at session end |
 | `digestMinInterval` | integer 0-86400 | `300` | `NMEM_DIGEST_MIN_INTERVAL` | Minimum seconds between session digests |
 | `maxContextResults` | integer 1-20 | `5` | `NMEM_MAX_CONTEXT_RESULTS` | How many memories to inject at prompt time |
+| `recallMinScore` | integer 0-100 | `0` | `NMEM_RECALL_MIN_SCORE` | Min relevance score (%) to include in auto-recall |
 | `apiUrl` | string | `""` | `NMEM_API_URL` | Remote server URL. Empty = local (127.0.0.1:14242) |
 | `apiKey` | string | `""` | `NMEM_API_KEY` | API key. Never logged. |
 

package/README.md

@@ -51,6 +51,11 @@ Or use a config file at `~/.nowledge-mem/openclaw.json`:
 
 The `apiKey` is injected as `NMEM_API_KEY` into the nmem CLI process. Never passed as a CLI argument, never logged.
 
+### Configure via WebUI of OpenClaw
+
+<img width="1192" height="2007" alt="openclaw-config" src="https://github.com/user-attachments/assets/a4abd855-db56-4a1f-b0ac-0114909d0c20" />
+
+
 ## How It Works
 
 ### Per-Turn Flow
@@ -332,6 +337,7 @@ To change settings, use the OpenClaw plugin settings UI. Changes take effect on
 | `sessionDigest` | boolean | `true` | Thread capture + LLM distillation at session end |
 | `digestMinInterval` | integer | `300` | Minimum seconds between session digests for the same thread (0-86400) |
 | `maxContextResults` | integer | `5` | Max memories to inject at prompt time (1-20, only used when sessionContext is enabled) |
+| `recallMinScore` | integer | `0` | Min relevance score (0-100%) to include in auto-recall. 0 = include all |
 | `apiUrl` | string | `""` | Remote server URL. Empty = local (`http://127.0.0.1:14242`) |
 | `apiKey` | string | `""` | API key for remote access. Injected as `NMEM_API_KEY` env var, never logged |
 
@@ -345,6 +351,7 @@ For persistent or scripted config, create `~/.nowledge-mem/openclaw.json`:
   "sessionDigest": true,
   "digestMinInterval": 300,
   "maxContextResults": 5,
+  "recallMinScore": 0,
   "apiUrl": "",
   "apiKey": ""
 }
@@ -361,6 +368,7 @@ NMEM_SESSION_CONTEXT=true
 NMEM_SESSION_DIGEST=true
 NMEM_DIGEST_MIN_INTERVAL=300
 NMEM_MAX_CONTEXT_RESULTS=5
+NMEM_RECALL_MIN_SCORE=0
 NMEM_API_URL=https://...
 NMEM_API_KEY=your-key
 ```

package/openclaw.plugin.json

@@ -1,15 +1,15 @@
 {
 	"id": "openclaw-nowledge-mem",
-	"version": "0.6.8",
+	"version": "0.6.9",
 	"kind": "memory",
 	"uiHints": {
 		"sessionContext": {
 			"label": "Session context injection",
-			"help": "Inject Working Memory and relevant memories at prompt time. Off by default \u2014 the agent can call memory_search on demand instead."
+			"help": "Inject Working Memory and relevant memories at prompt time. Off by default; the agent can call memory_search on demand instead."
 		},
 		"sessionDigest": {
 			"label": "Session digest at end",
-			"help": "Capture conversation threads and distill key memories via LLM at session end. On by default \u2014 negligible cost, one triage call per session."
+			"help": "Capture conversation threads and distill key memories via LLM at session end. On by default. Negligible cost, one triage call per session."
 		},
 		"digestMinInterval": {
 			"label": "Minimum digest interval (seconds)",
@@ -17,7 +17,11 @@
 		},
 		"maxContextResults": {
 			"label": "Max context results",
-			"help": "How many memories to inject for each context cycle (1\u201320)"
+			"help": "How many memories to inject for each context cycle (1-20)"
+		},
+		"recallMinScore": {
+			"label": "Min recall score (%)",
+			"help": "Only inject memories with relevance score above this threshold (0-100). Set to 0 to include all results."
 		},
 		"apiUrl": {
 			"label": "Server URL (remote mode)",
@@ -25,7 +29,7 @@
 		},
 		"apiKey": {
 			"label": "API key (remote mode)",
-			"help": "Required when connecting to a remote server. Never logged or passed as a CLI argument \u2014 injected as NMEM_API_KEY env var only.",
+			"help": "Required when connecting to a remote server. Never logged or passed as a CLI argument. Injected as NMEM_API_KEY env var only.",
 			"secret": true
 		}
 	},
@@ -35,7 +39,7 @@
 			"sessionContext": {
 				"type": "boolean",
 				"default": false,
-				"description": "Inject Working Memory and recalled memories at prompt time. Off by default \u2014 the agent queries memory on demand via tools."
+				"description": "Inject Working Memory and recalled memories at prompt time. Off by default; the agent queries memory on demand via tools."
 			},
 			"sessionDigest": {
 				"type": "boolean",
@@ -56,6 +60,13 @@
 				"maximum": 20,
 				"description": "Maximum memories to inject at prompt time (only used when sessionContext is enabled)"
 			},
+			"recallMinScore": {
+				"type": "integer",
+				"default": 0,
+				"minimum": 0,
+				"maximum": 100,
+				"description": "Minimum relevance score (0-100%) to include a memory in auto-recall. Set to 0 to include all results."
+			},
 			"apiUrl": {
 				"type": "string",
 				"default": "",

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@nowledge/openclaw-nowledge-mem",
-	"version": "0.6.8",
+	"version": "0.6.9",
 	"type": "module",
 	"description": "Nowledge Mem memory plugin for OpenClaw, local-first personal knowledge base",
 	"author": {

package/src/config.js

@@ -19,6 +19,7 @@ const ALLOWED_KEYS = new Set([
 	"sessionDigest",
 	"digestMinInterval",
 	"maxContextResults",
+	"recallMinScore",
 	"apiUrl",
 	"apiKey",
 	// Legacy aliases — accepted but not advertised
@@ -135,7 +136,7 @@ function firstDefined(...options) {
  * will configure via OpenClaw's plugin settings UI (pluginConfig).
  *
  * Canonical keys: sessionContext, sessionDigest, digestMinInterval,
- *                 maxContextResults, apiUrl, apiKey
+ *                 maxContextResults, recallMinScore, apiUrl, apiKey
  *
  * Legacy aliases (accepted from all sources; never shown in docs):
  *   autoRecall → sessionContext
@@ -148,6 +149,7 @@ function firstDefined(...options) {
  *   NMEM_SESSION_DIGEST        — true/1/yes to enable (alias: NMEM_AUTO_CAPTURE)
  *   NMEM_DIGEST_MIN_INTERVAL   — seconds (0-86400)
  *   NMEM_MAX_CONTEXT_RESULTS   — integer (1-20)
+ *   NMEM_RECALL_MIN_SCORE      — integer (0-100)
  *   NMEM_API_URL               — remote server URL
  *   NMEM_API_KEY               — API key (never logged)
  */
@@ -234,6 +236,20 @@ export function parseConfig(raw, logger) {
 	const maxContextResults = Math.min(20, Math.max(1, Math.trunc(mcr.value)));
 	_sources.maxContextResults = mcr.source;
 
+	// --- recallMinScore: file > pluginConfig > env > default ---
+	const rmsEnv = envInt("NMEM_RECALL_MIN_SCORE");
+	const rms = firstDefined(
+		{ value: pickNum(resolvedFile, "recallMinScore"), source: "file" },
+		{
+			value: pickNum(resolvedPlugin, "recallMinScore"),
+			source: "pluginConfig",
+		},
+		{ value: rmsEnv, source: "env" },
+		{ value: 0, source: "default" },
+	);
+	const recallMinScore = Math.min(100, Math.max(0, Math.trunc(rms.value)));
+	_sources.recallMinScore = rms.source;
+
 	// --- apiUrl: file > pluginConfig > env > "" ---
 	const fileUrl =
 		typeof resolvedFile.apiUrl === "string" && resolvedFile.apiUrl.trim();
@@ -267,6 +283,7 @@ export function parseConfig(raw, logger) {
 		sessionDigest,
 		digestMinInterval,
 		maxContextResults,
+		recallMinScore,
 		apiUrl,
 		apiKey,
 		_sources,

package/src/hooks/capture.js

@@ -250,7 +250,10 @@ function buildConversationText(normalized) {
 	const parts = [];
 	let total = 0;
 	for (const m of normalized) {
-		const text = truncate(m.fullContent || m.content, MAX_DISTILL_MESSAGE_CHARS);
+		const text = truncate(
+			m.fullContent || m.content,
+			MAX_DISTILL_MESSAGE_CHARS,
+		);
 		const line = `${m.role}: ${text}`;
 		if (total + line.length > MAX_CONVERSATION_CHARS) break;
 		parts.push(line);
@@ -297,7 +300,9 @@ export function buildAgentEndCaptureHandler(client, cfg, logger) {
 
 		//    Skip when no new messages were added (e.g. heartbeat re-sync).
 		if (!result || result.messagesAdded === 0) {
-			logger.debug?.("capture: no new messages since last sync, skipping triage");
+			logger.debug?.(
+				"capture: no new messages since last sync, skipping triage",
+			);
 			return;
 		}
 

package/src/hooks/recall.js

@@ -13,20 +13,148 @@ function escapeForPrompt(text) {
 	);
 }
 
+/** Max query length sent to search — longer messages get truncated. */
+const MAX_QUERY_LENGTH = 500;
+
+/**
+ * Messages shorter than this get augmented with recent conversational
+ * context. Below this threshold, a message is likely referential
+ * ("explain that more", "what about the second point?") and needs
+ * surrounding context to produce useful search results.
+ *
+ * Messages at or above this threshold are substantial enough to
+ * search on their own ("openviking 不好用", "how do I deploy to k8s?").
+ */
+const SHORT_QUERY_THRESHOLD = 40;
+
+/** How many recent messages to include for short-query context. */
+const CONTEXT_MESSAGES = 3;
+
+/** Per-message cap when building context window. */
+const CONTEXT_MSG_MAX_CHARS = 150;
+
+/**
+ * Extract text from message content (string or structured blocks).
+ * Mirrors the pattern from capture.js.
+ */
+function extractText(content) {
+	if (typeof content === "string") return content.trim();
+	if (!Array.isArray(content)) return "";
+	const parts = [];
+	for (const block of content) {
+		if (!block || typeof block !== "object") continue;
+		if (block.type === "text" && typeof block.text === "string") {
+			const text = block.text.trim();
+			if (text) parts.push(text);
+		}
+	}
+	return parts.join("\n").trim();
+}
+
+/**
+ * Normalize a raw message entry to { role, text }.
+ * Handles both flat { role, content } and nested { message: { role, content } }.
+ * Returns null for non-user/assistant messages, empty content, or slash commands.
+ */
+function normalizeMessage(raw) {
+	if (!raw || typeof raw !== "object") return null;
+	const msg =
+		raw.message && typeof raw.message === "object" ? raw.message : raw;
+	const role = typeof msg.role === "string" ? msg.role : "";
+	if (role !== "user" && role !== "assistant") return null;
+	const text = extractText(msg.content);
+	if (!text) return null;
+	// Skip slash commands — they're plugin directives, not conversational
+	if (role === "user" && text.startsWith("/")) return null;
+	return { role, text };
+}
+
+/**
+ * Build the search query from the hook event.
+ *
+ * Three tiers:
+ *   >= 40 chars : latest user message alone (self-contained)
+ *   3-39 chars  : latest user message + recent context (likely referential)
+ *   < 3 chars   : skip recall (too short — "ok", emoji, etc.)
+ *
+ * The event object provides:
+ * - event.messages: structured array of {role, content} messages (preferred)
+ * - event.prompt: the full formatted prompt (fallback, truncated)
+ */
+function buildSearchQuery(event) {
+	const messages = event?.messages;
+
+	if (Array.isArray(messages) && messages.length > 0) {
+		// Find the latest user message (walk backwards)
+		let latestUserIdx = -1;
+		let latestUserText = "";
+		for (let i = messages.length - 1; i >= 0; i--) {
+			const norm = normalizeMessage(messages[i]);
+			if (norm && norm.role === "user") {
+				latestUserIdx = i;
+				latestUserText = norm.text;
+				break;
+			}
+		}
+
+		if (!latestUserText || latestUserText.length < 3) return "";
+
+		// Substantial message — use it alone
+		if (latestUserText.length >= SHORT_QUERY_THRESHOLD) {
+			return latestUserText.slice(0, MAX_QUERY_LENGTH);
+		}
+
+		// Short message — likely a follow-up. Add recent context so
+		// the search engine sees what topic the conversation is about.
+		// Collect up to CONTEXT_MESSAGES messages preceding the latest user
+		// message (both user and assistant, for topic grounding).
+		const contextParts = [];
+		const scanFrom = Math.max(0, latestUserIdx - CONTEXT_MESSAGES);
+		for (let i = scanFrom; i < latestUserIdx; i++) {
+			const norm = normalizeMessage(messages[i]);
+			if (!norm) continue;
+			const truncated =
+				norm.text.length > CONTEXT_MSG_MAX_CHARS
+					? `${norm.text.slice(0, CONTEXT_MSG_MAX_CHARS)}…`
+					: norm.text;
+			contextParts.push(truncated);
+		}
+
+		if (contextParts.length > 0) {
+			// Latest message first (primary intent), then context
+			const combined = `${latestUserText}\n\n${contextParts.join("\n")}`;
+			return combined.slice(0, MAX_QUERY_LENGTH);
+		}
+
+		// No prior context available (first message) — use as-is
+		return latestUserText;
+	}
+
+	// Fallback: use event.prompt but truncate aggressively.
+	// This path only fires when OpenClaw doesn't provide event.messages.
+	if (typeof event?.prompt === "string" && event.prompt.length >= 5) {
+		return event.prompt.slice(0, MAX_QUERY_LENGTH);
+	}
+
+	return "";
+}
+
 /**
- * Builds the before_agent_start hook handler.
+ * Builds the before_prompt_build hook handler.
  *
- * Injects two layers of context at session start:
+ * Injects two layers of context at prompt time:
  * 1. Working Memory — today's focus, priorities, unresolved flags
- * 2. Relevant memories — with types, labels, and source provenance
+ * 2. Relevant memories — searched using the user's latest message
  *
  * Tool guidance is minimal — the agent already sees full tool descriptions
  * in its tool list. We only add a brief behavioral note.
  */
 export function buildRecallHandler(client, cfg, logger) {
+	const minScore = (cfg.recallMinScore ?? 0) / 100; // config is 0-100, API is 0-1
+
 	return async (event) => {
-		const prompt = event.prompt;
-		if (!prompt || prompt.length < 5) return;
+		const searchQuery = buildSearchQuery(event);
+		if (!searchQuery) return;
 
 		const sections = [];
 
@@ -44,9 +172,17 @@ export function buildRecallHandler(client, cfg, logger) {
 
 		// 2. Relevant memories — enriched with scoring signals and labels
 		try {
-			const results = await client.searchRich(prompt, cfg.maxContextResults);
-			if (results.length > 0) {
-				const lines = results.map((r) => {
+			const results = await client.searchRich(
+				searchQuery,
+				cfg.maxContextResults,
+			);
+			// Filter by minimum score if configured
+			const filtered =
+				minScore > 0
+					? results.filter((r) => (r.score ?? 0) >= minScore)
+					: results;
+			if (filtered.length > 0) {
+				const lines = filtered.map((r) => {
 					const title = r.title || "(untitled)";
 					const score = `${(r.score * 100).toFixed(0)}%`;
 					const labels =
@@ -82,7 +218,9 @@ export function buildRecallHandler(client, cfg, logger) {
 			"</nowledge-mem>",
 		].join("\n");
 
-		logger.debug?.(`recall: injecting ${context.length} chars`);
+		logger.debug?.(
+			`recall: injecting ${context.length} chars (query: ${searchQuery.slice(0, 80)}…)`,
+		);
 		return { prependContext: context };
 	};
 }

package/src/tools/connections.js

@@ -293,9 +293,7 @@ export function createConnectionsTool(client, logger) {
 						const otherTitle = isOlderNode
 							? edge.newer_title || "(untitled)"
 							: edge.older_title || "(untitled)";
-						const otherId = isOlderNode
-							? edge.newer_id
-							: edge.older_id;
+						const otherId = isOlderNode ? edge.newer_id : edge.older_id;
 						const direction = isOlderNode ? "→" : "←";
 						const idLine = otherId ? `\n    → id: ${otherId}` : "";
 						return `  ${direction} ${otherTitle}${relLabel ? ` — ${relLabel}` : ""}${idLine}`;

package/src/tools/status.js

@@ -106,6 +106,9 @@ export function createStatusTool(client, _logger, cfg) {
 			lines.push(
 				`  maxContextResults: ${cfg.maxContextResults} (${sources.maxContextResults || "?"})`,
 			);
+			lines.push(
+				`  recallMinScore: ${cfg.recallMinScore}% (${sources.recallMinScore || "?"})`,
+			);
 
 			details.config = {
 				sessionContext: {
@@ -124,6 +127,10 @@ export function createStatusTool(client, _logger, cfg) {
 					value: cfg.maxContextResults,
 					source: sources.maxContextResults,
 				},
+				recallMinScore: {
+					value: cfg.recallMinScore,
+					source: sources.recallMinScore,
+				},
 				apiUrl: { value: cfg.apiUrl || "(local)", source: sources.apiUrl },
 				apiKey: {
 					value: cfg.apiKey ? "(set)" : "(not set)",