docs: add memory index identity plan

docs/concepts/active-memory.md

@@ -683,7 +683,7 @@ The most important fields are:
 | `config.thinking`            | `"off" \| "minimal" \| "low" \| "medium" \| "high" \| "xhigh" \| "adaptive" \| "max"`                | Advanced thinking override for the blocking memory sub-agent; default `off` for speed                                                                                                                                                                    |
 | `config.promptOverride`      | `string`                                                                                             | Advanced full prompt replacement; not recommended for normal use                                                                                                                                                                                         |
 | `config.promptAppend`        | `string`                                                                                             | Advanced extra instructions appended to the default or overridden prompt                                                                                                                                                                                 |
-| `config.timeoutMs`           | `number`                                                                                             | Hard timeout for the blocking memory sub-agent, capped at 120000 ms                                                                                                                                                                                      |
+| `config.timeoutMs`           | `number`                                                                                             | Hard timeout for the blocking memory sub-agent, capped at 300000 ms. Higher values can delay the visible reply.                                                                                                                                          |
 | `config.setupGraceTimeoutMs` | `number`                                                                                             | Advanced extra setup budget before the recall timeout expires; defaults to 0 and is capped at 30000 ms. See [Cold-start grace](#cold-start-grace) for v2026.4.x upgrade guidance                                                                         |
 | `config.maxSummaryChars`     | `number`                                                                                             | Maximum total characters allowed in the active-memory summary                                                                                                                                                                                            |
 | `config.logging`             | `boolean`                                                                                            | Emits active memory logs while tuning                                                                                                                                                                                                                    |
@@ -800,6 +800,10 @@ If active memory is too slow:
 - reduce recent turn counts
 - reduce per-turn char caps
 
+For slow local recall models, you can raise `config.timeoutMs` up to `300000`
+ms. This is an explicit trade-off: Active Memory may wait longer before the
+visible reply starts.
+
 ## Common issues
 
 Active Memory rides on the configured memory plugin's recall pipeline, so most

extensions/active-memory/config.test.ts

@@ -60,7 +60,7 @@ describe("active-memory manifest config schema", () => {
       value: {
         enabled: true,
         agents: ["main"],
-        timeoutMs: 120_000,
+        timeoutMs: 300_000,
       },
     });
 
@@ -102,7 +102,7 @@ describe("active-memory manifest config schema", () => {
       value: {
         enabled: true,
         agents: ["main"],
-        timeoutMs: 120_001,
+        timeoutMs: 300_001,
       },
     });
 

extensions/active-memory/index.test.ts

@@ -381,11 +381,11 @@ describe("active-memory plugin", () => {
   it("registers before_prompt_build with the configured recall timeout", () => {
     api.pluginConfig = {
       agents: ["main"],
-      timeoutMs: 90_000,
+      timeoutMs: 180_000,
     };
     plugin.register(api as unknown as OpenClawPluginApi);
 
-    expect(hookOptions.before_prompt_build?.timeoutMs).toBe(90_000);
+    expect(hookOptions.before_prompt_build?.timeoutMs).toBe(180_000);
   });
 
   it("registers before_prompt_build with explicit setup grace when configured", () => {
@@ -3231,10 +3231,10 @@ describe("active-memory plugin", () => {
     expectLinesToContain(warnLines, "before_prompt_build");
   });
 
-  it("honors configured timeoutMs values above the former 60 000 ms ceiling", async () => {
+  it("honors configured timeoutMs values above the former 120 000 ms ceiling", async () => {
     api.pluginConfig = {
       agents: ["main"],
-      timeoutMs: 90_000,
+      timeoutMs: 180_000,
       logging: true,
     };
     plugin.register(api as unknown as OpenClawPluginApi);
@@ -3250,13 +3250,13 @@ describe("active-memory plugin", () => {
     );
 
     const passedTimeoutMs = lastEmbeddedRunParams().timeoutMs;
-    expect(passedTimeoutMs).toBe(90_000);
+    expect(passedTimeoutMs).toBe(180_000);
   });
 
-  it("clamps timeoutMs above the 120 000 ms ceiling to the ceiling", async () => {
+  it("clamps timeoutMs above the 300 000 ms ceiling to the ceiling", async () => {
     api.pluginConfig = {
       agents: ["main"],
-      timeoutMs: 200_000,
+      timeoutMs: 400_000,
       logging: true,
     };
     plugin.register(api as unknown as OpenClawPluginApi);
@@ -3272,7 +3272,7 @@ describe("active-memory plugin", () => {
     );
 
     const passedTimeoutMs = lastEmbeddedRunParams().timeoutMs;
-    expect(passedTimeoutMs).toBe(120_000);
+    expect(passedTimeoutMs).toBe(300_000);
   });
 
   it("sanitizes active-memory log fields onto a single line", async () => {

extensions/active-memory/index.ts

@@ -39,6 +39,7 @@ const DEFAULT_CACHE_TTL_MS = 15_000;
 const DEFAULT_MAX_CACHE_ENTRIES = 1000;
 const CACHE_SWEEP_INTERVAL_MS = 1000;
 const DEFAULT_MIN_TIMEOUT_MS = 250;
+const MAX_ACTIVE_MEMORY_TIMEOUT_MS = 300_000;
 const DEFAULT_SETUP_GRACE_TIMEOUT_MS = 0;
 const DEFAULT_QUERY_MODE = "recent" as const;
 const DEFAULT_QMD_SEARCH_MODE = "search" as const;
@@ -894,7 +895,7 @@ function normalizePluginConfig(
       parseOptionalPositiveInt(raw.timeoutMs, DEFAULT_TIMEOUT_MS),
       DEFAULT_TIMEOUT_MS,
       minimumTimeoutMs,
-      120_000,
+      MAX_ACTIVE_MEMORY_TIMEOUT_MS,
     ),
     setupGraceTimeoutMs: clampInt(raw.setupGraceTimeoutMs, setupGraceTimeoutMs, 0, 30_000),
     queryMode:

extensions/active-memory/openclaw.plugin.json

@@ -39,7 +39,7 @@
         "type": "string",
         "enum": ["off", "minimal", "low", "medium", "high", "xhigh", "adaptive"]
       },
-      "timeoutMs": { "type": "integer", "minimum": 250, "maximum": 120000 },
+      "timeoutMs": { "type": "integer", "minimum": 250, "maximum": 300000 },
       "setupGraceTimeoutMs": { "type": "integer", "minimum": 0, "maximum": 30000 },
       "queryMode": {
         "type": "string",
@@ -123,7 +123,8 @@
       "help": "Optional explicit denylist of chat/user IDs. Sessions whose resolved conversation id matches the list are skipped even when the chat type is allowed. Applied after allowedChatIds."
     },
     "timeoutMs": {
-      "label": "Timeout (ms)"
+      "label": "Timeout (ms)",
+      "help": "Hard timeout for the blocking memory sub-agent. Default: 15000. Maximum: 300000. Higher values can delay the visible reply."
     },
     "setupGraceTimeoutMs": {
       "label": "Setup Grace Timeout (ms)",

scratch/83333-memory-index-identity-plan.md

@@ -0,0 +1,97 @@
+# Memory Index Identity Fix Plan
+
+Issue: https://github.com/openclaw/openclaw/issues/83333
+
+## Goal
+
+Make a memory index healthy only when its stored identity matches the active memory config and embedding provider.
+
+The current bad state is:
+
+- config now wants `ollama` / `nomic-embed-text`
+- SQLite metadata still describes an old OpenAI index
+- `memory status --deep` can show `Dirty: no`
+- search can silently return no useful vector matches
+
+## Production Invariant
+
+Every path that trusts the memory DB must answer the same question:
+
+> Does this index belong to the active memory provider, model, provider key, sources, scope, chunking, tokenizer, and vector state?
+
+If no, the index is stale. It must be dirty, invalid, or rebuilt. It must not be reported as clean.
+
+## Current Gap
+
+The explicit indexing path already checks provider/model/providerKey mismatch through `shouldRunFullMemoryReindex` in `extensions/memory-core/src/memory/manager-reindex-state.ts`.
+
+The status-only path does not. `MemoryIndexManager` reads `meta.vectorDims`, then `resolveInitialMemoryDirty` treats any existing metadata as clean for status managers:
+
+- `extensions/memory-core/src/memory/manager.ts`
+- `extensions/memory-core/src/memory/manager-status-state.ts`
+- `extensions/memory-core/src/cli.runtime.ts`
+
+That lets deep status combine new provider info with old index metadata.
+
+## Fix Shape
+
+1. Extract a shared index identity check.
+   - Reuse the comparison logic behind `shouldRunFullMemoryReindex`.
+   - Return a structured result: `valid`, `missing`, or `mismatched`.
+   - Include a short mismatch reason for status output and tests.
+
+2. Use that check when constructing status-only managers.
+   - Read stored metadata.
+   - Build expected identity from current settings and initialized provider when available.
+   - If identity does not match, set `dirty = true`.
+
+3. Make `memory status --deep` validate after provider initialization.
+   - Deep status probes initialize the provider.
+   - Re-run the identity check after that point.
+   - Never print `Dirty: no` when provider/model/providerKey metadata mismatches.
+
+4. Protect search bootstrap.
+   - Before trusting existing vector rows, validate index identity.
+   - If stale and sync is allowed, trigger the safe reindex path.
+   - If stale and sync cannot run, return a degraded/mismatch state instead of silently acting healthy.
+
+5. Keep rebuilds atomic.
+   - Provider/model identity changes should use the existing safe temp-DB reindex and swap.
+   - Do not patch mixed-provider rows in place.
+
+## Non-Goals
+
+- No Ollama special case.
+- No config migration.
+- No plugin API change unless implementation proves memory-core lacks required identity data.
+- No hardcoded vector dimension rules such as `1536 means OpenAI`.
+
+## Tests
+
+Add focused tests for:
+
+- status-only manager marks mismatched provider/model metadata dirty
+- `memory status --deep` does not report clean after OpenAI metadata with Ollama config
+- manual `memory index` still reuses the same identity check
+- search does not silently return a healthy empty vector result from a mismatched index
+
+Suggested commands:
+
+```sh
+node scripts/run-vitest.mjs \
+  extensions/memory-core/src/memory/manager-status-state.test.ts \
+  extensions/memory-core/src/memory/manager-reindex-state.test.ts \
+  extensions/memory-core/src/memory/manager-search.test.ts \
+  extensions/memory-core/src/cli.test.ts
+```
+
+## Optional Live Proof
+
+If feasible before landing, run a Docker or Crabbox smoke:
+
+1. Build a small OpenAI-style memory index.
+2. Switch `agents.defaults.memorySearch` to Ollama.
+3. Run `openclaw memory status --deep`.
+4. Confirm status is dirty or invalid, not clean with stale dimensions.
+
+Unit proof is enough for the core invariant. Live proof is useful for confidence, not a reason to special-case Ollama.