@smyslenny/agent-memory 5.0.2 → 5.1.0

package/docs/README-zh.md

@@ -4,7 +4,7 @@
 
 **完整文档请参阅 [README.md](../README.md)（英文）**
 
-当前版本：**v5.0.1**
+当前版本：**v5.0.2**
 
 ---
 
@@ -134,7 +134,6 @@ npx agent-memory status
 - [通用运行时集成](integrations/generic.md)
 - [OpenClaw 集成](integrations/openclaw.md)
 - [v3 → v4 迁移指南](migration-v3-v4.md)
-- [v5 设计文档 (DD-0018)](design/0018-v5-memory-intelligence.md)
 
 ## 许可证
 

package/docs/integrations/generic.md

@@ -157,8 +157,8 @@ MCP is a good fit when your host already has a tool abstraction.
 
 ### MCP tool list
 
-- `remember`
-- `recall`
+- `remember` — store a memory (supports provenance: `source_session`, `source_context`, `observed_at`)
+- `recall` — hybrid search (supports `related`, `after`, `before`, `recency_boost`)
 - `recall_path`
 - `boot`
 - `forget`
@@ -166,7 +166,8 @@ MCP is a good fit when your host already has a tool abstraction.
 - `status`
 - `ingest`
 - `reindex`
-- `surface`
+- `surface` — context-aware surfacing (supports `related`, `after`, `before`, `recency_boost`)
+- `link` — manually create or remove associations between memories
 
 See [examples/mcp-stdio](../../examples/mcp-stdio) for a minimal example.
 
@@ -276,6 +277,45 @@ Options:
 For generic runtimes, prefer **explicit ingest** first. Set
 `AGENT_MEMORY_AUTO_INGEST=0` unless you intentionally want watcher behavior.
 
+## v5 Features
+
+AgentMemory v5 adds intelligence capabilities that enhance the integration
+pattern described above. All features are backward-compatible.
+
+### Memory Links (F1)
+
+Memories are automatically linked to semantically related memories during
+writes. Use `related=true` in `recall` or `surface` to expand results with
+linked memories. The `link` tool allows manual link management.
+
+### Conflict Detection (F2)
+
+Write Guard now detects contradictions (negation, value changes, status
+changes) during writes. Conflicts are reported in the sync result. A
+**Conflict Override** rule ensures status updates (e.g. TODO → DONE) are not
+incorrectly deduplicated.
+
+### Temporal Recall (F3)
+
+`recall` and `surface` accept `after`, `before`, and `recency_boost` parameters
+for time-aware search. Time filtering happens at the SQL layer.
+
+### Passive Feedback (F4)
+
+When `recall` returns results, positive feedback is automatically logged for
+the top-3 hits. Rate-limited to 3 passive events per memory per 24 hours.
+
+### Semantic Decay (F5)
+
+The `tidy` phase detects stale content through keyword pattern matching
+(e.g. "in progress", "TODO:", "just now"). `identity` and `emotion` types
+are exempt.
+
+### Memory Provenance (F6)
+
+Memories can carry `source_session`, `source_context`, and `observed_at`
+metadata to track where and when they originated.
+
 ## Common mistakes
 
 - **Writing everything**: only store durable memory candidates

package/docs/integrations/openclaw.md

@@ -37,10 +37,10 @@ That means the Markdown workflow is still valid here, but it is now an
       "command": "node",
       "args": ["./node_modules/@smyslenny/agent-memory/dist/mcp/server.js"],
       "env": {
-        "AGENT_MEMORY_DB": "./agent-memory.db",
-        "AGENT_MEMORY_AGENT_ID": "noah",
+        "AGENT_MEMORY_DB": "~/.openclaw/workspace/agent-memory.db",
+        "AGENT_MEMORY_AGENT_ID": "my-agent",
         "AGENT_MEMORY_AUTO_INGEST": "1",
-        "AGENT_MEMORY_WORKSPACE": "/home/user/.openclaw/workspace"
+        "AGENT_MEMORY_WORKSPACE": "~/.openclaw/workspace"
       }
     }
   }
@@ -73,6 +73,46 @@ This keeps responsibilities separated:
 - AgentMemory is good for retrieval, dedup, surfacing, and lifecycle
 - OpenClaw is good at orchestration and scheduled host behavior
 
+## v5 Features for OpenClaw Users
+
+AgentMemory v5 adds six intelligence features that are particularly useful
+in the OpenClaw cron pipeline:
+
+### Memory Provenance (F6)
+
+When calling `remember` from your sync prompt, include provenance metadata:
+
+```
+remember(content="...", type="knowledge",
+  source_session="session-id",
+  source_context="extracted from 14:00 sync",
+  observed_at="2026-03-20T14:00:00+08:00")
+```
+
+### Temporal Recall (F3)
+
+Use `after`, `before`, and `recency_boost` to filter memories by time:
+
+```
+recall(query="deployment decisions", after="2026-03-01", recency_boost=true)
+surface(task="plan next sprint", after="2026-03-15", related=true)
+```
+
+### Memory Links (F1)
+
+The `link` tool allows manual association between memories. Use `related=true`
+in `recall` or `surface` to expand results with linked memories.
+
+### Conflict Detection (F2)
+
+Write Guard detects contradictions during writes (e.g., status changes from
+TODO to DONE). Conflicts are reported without blocking writes.
+
+### Passive Feedback (F4) & Semantic Decay (F5)
+
+`recall` automatically logs positive feedback for top hits. The `tidy` phase
+detects stale content patterns and accelerates decay for outdated memories.
+
 ## Example directory layout
 
 ```text
@@ -132,10 +172,9 @@ What changed in v4 is not support, but **positioning**:
 
 See [examples/openclaw](../../examples/openclaw) for:
 
-- setup notes
-- sample cron prompts
-- journal examples
+- setup notes and cron configuration
 - memory janitor Phase 5 template
+- journal and long-term memory examples
 
 ## Migration notes for v3 users
 
@@ -145,4 +184,5 @@ If you came from the old v3 README:
 - it has simply moved out of the project homepage
 - generic integration guidance now lives alongside it, not under it
 
-For release-level changes, see [v3 → v4 migration guide](../migration-v3-v4.md).
+For release-level changes, see [v3 → v4 migration guide](../migration-v3-v4.md)
+and the [README](../../README.md) for v5 features.

package/docs/migration-v3-v4.md

@@ -234,3 +234,18 @@ The biggest v4 migration is conceptual, not operational:
 
 That makes the project easier to evaluate, easier to adopt, and easier to plug
 into systems that do not look like OpenClaw at all.
+
+## What's next: v5
+
+AgentMemory **v5 (Memory Intelligence)** is now available. It adds six
+backward-compatible features on top of v4:
+
+- **Memory Links** — automatic semantic associations between memories
+- **Conflict Detection** — Write Guard detects contradictions during writes
+- **Temporal Recall** — `after`, `before`, and `recency_boost` for time-aware search
+- **Passive Feedback** — automatic positive feedback for accessed memories
+- **Semantic Decay** — stale content detection beyond pure time-based Ebbinghaus
+- **Memory Provenance** — `source_session`, `source_context`, `observed_at` metadata
+
+All v4 workflows continue to work unchanged. See the
+[README](../README.md) for the full v5 feature overview.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@smyslenny/agent-memory",
-  "version": "5.0.2",
+  "version": "5.1.0",
   "description": "Agent-native memory layer with lifecycle management for AI agents — SQLite-first, Write Guard, hybrid recall, MCP, CLI, and HTTP API.",
   "type": "module",
   "main": "dist/index.js",