@xdarkicex/openclaw-memory-libravdb 1.8.6 → 1.8.8

package/README.md

@@ -318,12 +318,22 @@ libravdbd migrate --from ~/.libravdbd/data.libravdb --tenant <tenantId>
 
 ## Vector Service Configuration (YAML) & Kubernetes
 
-`libravdbd` is heavily configurable via environment variables or a YAML configuration file. The vector service looks for `config.yaml` in this order:
-1. `LIBRAVDB_CONFIG=/path/to/config.yaml`
+`libravdbd` is configured via environment variables or a YAML configuration file. The vector service looks for a config file in this order:
+
+1. `LIBRAVDB_CONFIG=/path/to/config.yaml` (env var — set this for custom paths)
 2. `/etc/libravdbd/config.yaml`
 3. `~/.libravdbd/config.yaml`
 
-Example `config.yaml` for a Kubernetes StatefulSet deployment in multi-tenant mode:
+Env vars override YAML values. All fields are optional — the daemon ships with sensible defaults.
+
+**Reference YAML files:**
+
+| Backend | File | Description |
+|---------|------|-------------|
+| GGUF (recommended) | [`docs/yaml/default-gguf.yaml`](docs/yaml/default-gguf.yaml) | llama.cpp backend, 3-5x faster on CPU |
+| ONNX (fallback) | [`docs/yaml/default-onnx.yaml`](docs/yaml/default-onnx.yaml) | ONNX Runtime backend, wider platform support |
+
+**Kubernetes example** (multi-tenant mode with GGUF):
 
 ```yaml
 # /etc/libravdbd/config.yaml
@@ -333,6 +343,7 @@ tenant_max_open: 128
 grpc_endpoint: "tcp:0.0.0.0:9090"
 embedding_backend: "gguf"
 embedding_profile: "nomic-embed-text-v1.5"
+llama_lib_path: "/var/lib/libravdbd/models/llama/llama-linux-amd64/lib/libllama.so"
 drain_timeout: "25s" # Must be less than k8s terminationGracePeriodSeconds
 ```
 

package/dist/context-engine.js

@@ -1626,6 +1626,44 @@ export function buildContextEngineFactory(runtime, cfg, logger = console) {
                 : {}),
         };
     }
+    async function injectContinuityContext(params) {
+        try {
+            // Use a natural-language query that semantically matches the
+            // pointer record text ("Previous session continuity — ...").
+            // Fetch enough results so the exact ID match isn't crowded out
+            // by stronger semantic hits in the user collection.
+            const continuityHits = await params.client.searchTextCollections({
+                collections: [resolveUserCollection(params.userId)],
+                text: "previous session context continuity",
+                k: 8,
+                excludeByCollection: {},
+            });
+            const continuityHit = continuityHits.results?.find((r) => r.id === "__session_continuity__");
+            if (!continuityHit)
+                return null;
+            let meta = {};
+            if (continuityHit.metadataJson && continuityHit.metadataJson.length > 0) {
+                try {
+                    meta = JSON.parse(new TextDecoder().decode(continuityHit.metadataJson));
+                }
+                catch { /* metadata parse failed, use empty */ }
+            }
+            const summaryId = meta.summary_id;
+            if (!summaryId)
+                return null;
+            const expanded = await params.client.expandSummary({
+                sessionId: meta.session_id ?? params.sessionId,
+                summaryId,
+                maxDepth: 2,
+            });
+            if (!expanded.text)
+                return null;
+            return '<continuity_context>\nThe following is a summary of the previous session. Use it for context about what was discussed before the reset.\n' + expanded.text + '\n</continuity_context>';
+        }
+        catch {
+            return null;
+        }
+    }
     async function runCompaction(args) {
         const request = buildCompactSessionRequest(args);
         try {
@@ -1857,7 +1895,18 @@ export function buildContextEngineFactory(runtime, cfg, logger = console) {
                     emitDebug: true,
                 });
                 const assembled = normalizeAssembleResult(resp, args.messages);
-                let enforced = enforceTokenBudgetInvariant(await augmentWithExactRecall(assembled, {
+                const continuityContext = await injectContinuityContext({
+                    client,
+                    userId,
+                    sessionId,
+                    logger,
+                    tokenBudget: args.tokenBudget,
+                    systemPromptAddition: assembled.systemPromptAddition,
+                });
+                const withContinuity = continuityContext
+                    ? { ...assembled, systemPromptAddition: appendSystemPromptAddition(assembled.systemPromptAddition, continuityContext) }
+                    : assembled;
+                let enforced = enforceTokenBudgetInvariant(await augmentWithExactRecall(withContinuity, {
                     queryText: strippedPrompt || (messages[messages.length - 1]?.content ?? ""),
                     userId,
                     sessionId,

package/dist/index.js

@@ -28144,6 +28144,38 @@ function buildContextEngineFactory(runtime, cfg, logger = console) {
       ...typeof cfg.continuityPriorContextTokens === "number" ? { continuityPriorContextTokens: cfg.continuityPriorContextTokens } : {}
     };
   }
+  async function injectContinuityContext(params) {
+    try {
+      const continuityHits = await params.client.searchTextCollections({
+        collections: [resolveUserCollection(params.userId)],
+        text: "previous session context continuity",
+        k: 8,
+        excludeByCollection: {}
+      });
+      const continuityHit = continuityHits.results?.find(
+        (r) => r.id === "__session_continuity__"
+      );
+      if (!continuityHit) return null;
+      let meta = {};
+      if (continuityHit.metadataJson && continuityHit.metadataJson.length > 0) {
+        try {
+          meta = JSON.parse(new TextDecoder().decode(continuityHit.metadataJson));
+        } catch {
+        }
+      }
+      const summaryId = meta.summary_id;
+      if (!summaryId) return null;
+      const expanded = await params.client.expandSummary({
+        sessionId: meta.session_id ?? params.sessionId,
+        summaryId,
+        maxDepth: 2
+      });
+      if (!expanded.text) return null;
+      return "<continuity_context>\nThe following is a summary of the previous session. Use it for context about what was discussed before the reset.\n" + expanded.text + "\n</continuity_context>";
+    } catch {
+      return null;
+    }
+  }
   async function runCompaction(args) {
     const request3 = buildCompactSessionRequest(args);
     try {
@@ -28376,8 +28408,17 @@ function buildContextEngineFactory(runtime, cfg, logger = console) {
           emitDebug: true
         });
         const assembled = normalizeAssembleResult(resp, args.messages);
+        const continuityContext = await injectContinuityContext({
+          client,
+          userId,
+          sessionId,
+          logger,
+          tokenBudget: args.tokenBudget,
+          systemPromptAddition: assembled.systemPromptAddition
+        });
+        const withContinuity = continuityContext ? { ...assembled, systemPromptAddition: appendSystemPromptAddition(assembled.systemPromptAddition, continuityContext) } : assembled;
         let enforced = enforceTokenBudgetInvariant(
-          await augmentWithExactRecall(assembled, {
+          await augmentWithExactRecall(withContinuity, {
             queryText: strippedPrompt || (messages[messages.length - 1]?.content ?? ""),
             userId,
             sessionId,
@@ -36079,7 +36120,7 @@ function createMemoryDescribeTool(getClient, getSessionId = () => void 0, logger
     }
   };
 }
-function createMemoryExpandTool(getClient, getSessionKey, logger = console) {
+function createMemoryExpandTool(getClient, getSessionKey, logger = console, getSessionId = () => void 0) {
   return {
     name: "memory_expand",
     label: "Memory Expand",
@@ -36092,7 +36133,7 @@ function createMemoryExpandTool(getClient, getSessionKey, logger = console) {
       if (summaryIds.length === 0) throw new Error("memory_expand requires at least one summaryId");
       const maxDepth = readNum(params, "maxDepth", { integer: true, min: 0 }) ?? 1;
       let maxTokens = readNum(params, "maxTokens", { integer: true }) ?? MAX_EXPAND_TOKENS;
-      const sessionId = readStr(params, "sessionId") ?? "";
+      const sessionId = readStr(params, "sessionId") ?? getSessionId() ?? "";
       const sessionKey = getSessionKey();
       if (sessionKey) {
         const grantedTokens = consumeSubagentBudget(sessionKey, maxTokens);
@@ -39976,7 +40017,8 @@ function register(api) {
     api.registerTool?.((ctx) => {
       const getClient = runtimeOrNull.getClient;
       const getSessionKey = () => ctx.sessionKey;
-      return createMemoryExpandTool(getClient, getSessionKey, logger);
+      const getSessionId = () => ctx.sessionId;
+      return createMemoryExpandTool(getClient, getSessionKey, logger, getSessionId);
     }, { names: ["memory_expand"] });
     api.registerTool?.((ctx) => {
       const getClient = runtimeOrNull.getClient;

package/dist/tools/memory-recall.d.ts

@@ -67,7 +67,7 @@ export declare function createMemoryDescribeTool(getClient: ClientGetter, getSes
     };
     execute: (_toolCallId: string, rawParams: unknown) => Promise<ToolResult<MemoryDescribeDetails>>;
 };
-export declare function createMemoryExpandTool(getClient: ClientGetter, getSessionKey: () => string | undefined, logger?: LoggerLike): {
+export declare function createMemoryExpandTool(getClient: ClientGetter, getSessionKey: () => string | undefined, logger?: LoggerLike, getSessionId?: () => string | undefined): {
     name: string;
     label: string;
     description: string;

package/dist/tools/memory-recall.js

@@ -187,7 +187,7 @@ export function createMemoryDescribeTool(getClient, getSessionId = () => undefin
         },
     };
 }
-export function createMemoryExpandTool(getClient, getSessionKey, logger = console) {
+export function createMemoryExpandTool(getClient, getSessionKey, logger = console, getSessionId = () => undefined) {
     return {
         name: "memory_expand",
         label: "Memory Expand",
@@ -204,7 +204,7 @@ export function createMemoryExpandTool(getClient, getSessionKey, logger = consol
                 throw new Error("memory_expand requires at least one summaryId");
             const maxDepth = readNum(params, "maxDepth", { integer: true, min: 0 }) ?? 1;
             let maxTokens = readNum(params, "maxTokens", { integer: true }) ?? MAX_EXPAND_TOKENS;
-            const sessionId = readStr(params, "sessionId") ?? "";
+            const sessionId = readStr(params, "sessionId") ?? getSessionId() ?? "";
             // Subagent budget gate: if this is a subagent, check remaining expansion budget.
             const sessionKey = getSessionKey();
             if (sessionKey) {

package/docs/yaml/default-gguf.yaml

@@ -0,0 +1,52 @@
+# Vector Service — GGUF backend (recommended)
+#
+# Place at /etc/libravdbd/config.yaml, ~/.libravdbd/config.yaml,
+# or any path pointed to by LIBRAVDB_CONFIG.
+#
+# GGUF/llama.cpp is 3-5x faster than ONNX on CPU.
+# Requires libllama.dylib/.so and a .gguf model file.
+# Provision: scripts/provision.sh (shipped with the daemon release).
+#
+# All fields are optional. Env vars override YAML values.
+
+# --- storage ---
+db_path: ""
+
+# --- transport ---
+grpc_endpoint: ""            # unix:/path/to/sock or tcp:host:port
+
+# --- embedding ---
+embedding_backend: gguf      # gguf only — does not fall back to ONNX
+embedding_profile: nomic-embed-text-v1.5
+fallback_profile: bge-small-en-v1.5
+embedding_normalize: true
+embedding_dimensions: 768
+
+# --- gguf ---
+llama_lib_path: ""           # path to libllama.dylib/.so (empty = auto-resolve)
+                             # Linux example: /var/lib/libravdbd/models/llama/llama-linux-amd64/lib/libllama.so
+                             # macOS example: /opt/homebrew/opt/libravdbd/models/llama/llama-darwin-arm64/lib/libllama.dylib
+
+# --- summarizer ---
+summarizer_backend: bundled  # bundled (=extractive, always used regardless)
+
+# --- gating ---
+gating_threshold: 0.35
+gating_w1c: 0.35
+gating_w2c: 0.40
+gating_w3c: 0.25
+gating_w1t: 0.40
+gating_w2t: 0.35
+gating_w3t: 0.25
+gating_tech_norm: 1.5
+
+# --- timeouts ---
+embed_timeout: 5s
+drain_timeout: 25s
+
+# --- circuit breaker ---
+circuit_max_failures: 3
+circuit_cooldown: 60s
+
+# --- lifecycle ---
+lifecycle_journal_max_entries: 500

package/docs/yaml/default-onnx.yaml

@@ -0,0 +1,47 @@
+# Vector Service — ONNX backend (default fallback)
+#
+# Place at /etc/libravdbd/config.yaml, ~/.libravdbd/config.yaml,
+# or any path pointed to by LIBRAVDB_CONFIG.
+#
+# All fields are optional. Env vars override YAML values.
+
+# --- storage ---
+db_path: ""
+
+# --- transport ---
+grpc_endpoint: ""            # unix:/path/to/sock or tcp:host:port
+
+# --- embedding ---
+embedding_backend: bundled   # bundled (=gguf-first, onnx fallback), onnx-local, gguf, remote
+embedding_profile: nomic-embed-text-v1.5
+fallback_profile: bge-small-en-v1.5
+embedding_model_path: ""
+embedding_tokenizer_path: ""
+embedding_dimensions: 768
+embedding_normalize: true
+onnx_runtime_path: ""        # path to libonnxruntime.dylib/.so (empty = auto-resolve)
+onnx_device: ""              # auto, cpu, cuda, coreml, openvino, directml
+
+# --- summarizer ---
+summarizer_backend: bundled  # bundled (=extractive, always used regardless)
+
+# --- gating ---
+gating_threshold: 0.35
+gating_w1c: 0.35
+gating_w2c: 0.40
+gating_w3c: 0.25
+gating_w1t: 0.40
+gating_w2t: 0.35
+gating_w3t: 0.25
+gating_tech_norm: 1.5
+
+# --- timeouts ---
+embed_timeout: 5s
+drain_timeout: 25s
+
+# --- circuit breaker ---
+circuit_max_failures: 3
+circuit_cooldown: 60s
+
+# --- lifecycle ---
+lifecycle_journal_max_entries: 500

package/openclaw.plugin.json

@@ -2,7 +2,7 @@
   "id": "libravdb-memory",
   "name": "LibraVDB Memory",
   "description": "Persistent vector memory with three-tier hybrid scoring",
-  "version": "1.8.6",
+  "version": "1.8.8",
   "kind": [
     "memory",
     "context-engine"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@xdarkicex/openclaw-memory-libravdb",
-  "version": "1.8.6",
+  "version": "1.8.8",
   "type": "module",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",