Get started

v2026.8.1: Memory

Memory now lets an eligible personal Claw recall relevant context from that agent's other private conversations, including what mattered immediately before a reset, while visible workflows let you search indexed sources, inspect how memory is working, import supported history, and remove attributable derived memory. Recall stays within the same agent's private conversations and respects explicit isolation and access policy.

Built-in Memory owns the core search and recall path, with a supported Doctor migration from QMD. LanceDB, Memory Wiki, external embedding services, MEMORY.md, and USER.md still have distinct roles, and forgetting derived memory does not erase the original conversation or copies outside OpenClaw.

Upgrading to built-in Memory

Built-in Memory now owns the core search and recall path. If you use QMD, run openclaw doctor --fix to remove retired QMD settings, carry forward supported extra paths and any session indexing you explicitly enabled, preserve compatible rows already in the agent database, and rebuild the index from canonical Markdown.

The migration carries supported data into a different core, so QMD-only reranking, query expansion, and cross-agent transcript search are retired. Malformed structures, incompatible vector dimensions, and data without a safe owner remain stopped for repair.

Sources and complete change list

Improvements

Retire QMD and use builtin memory exclusively #120936
Add a pre-cutover check for missing local embedding setup #123575

Bug fixes

Recover gateway startup from legacy Memory Core index conflicts #108652
Prevent quadratic Memory Core migrations #109355
Coordinate QMD writers with SQLite leases #109636
Recover Memory Core from same-file legacy index conflicts #110216
Restore upgrades for pre-provenance agent databases #115144
Prevent malformed Active Memory timestamps from blocking migration #103116
Let Memory Core upgrades preserve compatible canonical cache rows #107243
Let Doctor repair legacy memory-index schemas #115154
Recover Memory Core upgrades with divergent legacy journals #115855
Keep memory databases compatible with rollback #116003
Keep memory diagnostics usable during degraded upgrades #116074
fix(memory): preserve session recall when upgrading from QMD #130016
Repair legacy memory-search settings without losing tuning #68685
Prevent oversized workspace files from crashing doctor repairs #101448
Remove false legacy-memory warnings for empty sidecars #114661
fix(memory): show canonical-session migration recovery instead of provider error #119061

Documentation

Document automatic migration from QMD memory #121911

Finding and recalling past context

On eligible personal setups, your Claw can recall relevant context from that same agent's other private conversations by default, including what mattered immediately before you reset the session. Recall remains limited to that agent's private conversations; groups, channels, shared aliases, other agents, deleted history, and policy-blocked sources stay out, and an explicit direct-message isolation setting still wins.

Built-in search now understands filenames, full and partial Unicode paths, and configured extra paths, broadens thin strict matches, and keeps keyword results available when an optional embedding provider cannot start. While memory or session content is being rebuilt, each search stays on one stable published index, and later searches use its replacement only after publication instead of waiting behind routine maintenance or mixing generations. Search stays within the configured roots and agent boundaries, while required embedding providers fail closed.

Sessions without a configured reset policy now remain open across days, and durable reset or compaction markers explain visible history changes. SQLite-backed chats on web, macOS, iOS, and Android can rewind to a user message, fork the conversation, and switch among preserved branches. Rewinding changes the transcript branch, but it does not undo files, sent messages, or other tool side effects.

Sources and complete change list

Improvements

Let personal agents remember across private conversations #100140
Add per-agent Memory Wiki vault isolation #103349
Start configured local embedding services on demand #104306
Find memories by filename and path #104449
Preserve interrupted-turn context and reopen search-hit history #105831
Default cross-conversation recall on for personal installs #110597
Rewind or fork a chat from any user message #110660
Switch between preserved transcript branches from chat #110857
Add rewind and fork actions to native chat bubbles #110886
Keep sessions continuous by default #111140
Add native session branch switching with safe offline queues #112056
Add Android chat rewind, fork, and branch switching #112284
Improve conversational memory recall #121196
Add glob patterns to extra memory paths #121209
Enable diverse builtin memory results by default #121224
Show durable reset and compaction boundaries in chat history #122222
Support embedding-only managed llama.cpp servers #125383
fix(llama): support embedding-only managed setup #130883
Add opt-in persistent sessions for webhook hooks #75918
Show llama.cpp acceleration and runtime diagnostics #104310
Add recall-specific fast mode for Active Memory #108043
Avoid duplicate embeddings during filtered memory recall #125735
Skip unchanged QMD session export rebuilds #77158
Prepare active-memory recall context once before launch 061ccb1

Bug fixes

Harden embedding batch output and error handling #103472
Isolate LanceDB memories by agent #103799
Restore Active Memory recall with SQLite sessions #105255
Route Active Memory recall through subscription-authenticated Claude CLI #106840
Block cross-agent Memory Wiki reads from sandboxed sub-agents #111463
fix(memory-core): keep memory search usable during dirty sync #112214
Prevent broad memory searches from freezing the gateway #113359
Prevent overlapping memory embedding providers #113471
Restore prompt images when rewinding or forking chats #113945
Keep Codex transcript mirroring responsive as history grows #115070
Fall back to keyword memory search when embeddings cannot start #115397
Reuse one Active Memory recall per run #115627
Index active memory from SQLite while preserving searchable archives #117334
Restore CJK recall intent in Active Memory #117419
Warn when memory search results may be stale #117706
Keep Ollama embedding credentials on their selected host #118753
Isolate memory fallback provider credentials #118759
Isolate LanceDB embedding credentials per agent #118767
Keep private root memory out of shared chats #119198
Recall private conversation context after a session reset #122051
Prevent silent context loss during resets and memory search #125534
Bound LanceDB memory prompts and reload embedding credentials #125567
fix(memory): isolate sqlite-vec KNN from the event loop #128078
fix: preserve conversation after memory flush exhaustion #132010
Restore conversation history to Codex prompt hooks #101752
Stream Gemini embedding batch output to reduce memory use #102974
Keep memory batch failure accounting accurate #103117
Preserve Unicode at QMD memory snippet limits #103996
Preserve memory-search settings in isolated cron runs #104196
Isolate local embedding services by runtime #104741
Keep Gemini memory search available after version upgrades #104759
fix(active-memory): bound timeout circuit breakers #105316
Make LanceDB memory initialization safe across processes #105896
Preserve distinct runs in long native chat histories #107965
Keep Windows QMD cleanup from blocking the Gateway #109265
Isolate memory SecretRef failures by agent #109977
Move Memory Wiki compiled state into plugin-owned SQLite storage #110167
Reject malformed embedding batch UTF-8 #110502
Reject malformed UTF-8 in memory-host JSON responses #111279
Keep Gemini embedding batch waits within the configured timeout #111674
Keep OpenAI embedding batch polling within its timeout #111675
Stop repeated memory auto-recall stalls during embedding outages #112927
Preserve QMD memory hits when document IDs are stale #113515
Avoid full index rebuilds after empty memory searches #114183
Stop Memory Core retries immediately on cancellation #114300
Preserve QMD memory hits when command output is noisy #115023
Keep memory annotations out of recall and search text #115719
Reconcile stale memory full-text indexes safely #115775
fix(active-memory): reject assistant chitchat from recall summary (#84034) #115803
Preserve Windows memory session ownership across path casing #115851
Recover queued Memory Core sync work after SQLite failures #115923
Warm Active Memory before the first channel turn #115936
Reject malformed and empty embedding vectors #117200
Keep targeted memory indexing on one session snapshot #117247
Keep CLI compaction on rotated transcripts #117250
Honor LM Studio embedding JIT and canonical model identity #117320
Avoid redundant compaction after reset boundaries #117400
Recover memory freshness after transcript restores #117548
Keep OpenAI memory embedding caches stable across upgrades #117557
Accept command-local agent selection for Memory Wiki #117943
Keep QMD running after filesystem watcher errors #118171
Honor keyed agent rosters in Memory Core #118310
Surface memory-wiki backend read failures #118449
Keep unrelated session transcripts out of Voice Call fast context #118498
Honor configured memory providers in CLI searches #119186
Restore delegated context-engine compaction session identity #120047
Report persisted vector index state in memory status #120048
fix(active-memory): keep recall errors out of prompt context #120658
Keep session reconciliation off the memory-search path #120837
fix(memory): apply temporal decay to dated files in memory subdirectories #121103
Preserve memory diversity with project relevance #121608
Keep multimodal memory indexing limited to extra paths #121627
Remove stale Memory Core session rows safely at startup #121665
Restore configured context engines for scheduled turns #122457
fix(memory-wiki): keep default vaults in configured state directory #122591
Retain transcript updates during active memory sync #124024
Keep local memory healthy after a cancelled search #124051
Stop cancelled memory searches before returning results #124118
Stop timed-out LanceDB recall before returning results #124143
Preserve complete startup memory after short file reads #124597
Preserve memory indexes when source scans fail #125319
Warn when combined memory search omits the memory corpus #125500
Preserve conversation history when context assembly fails #125714
Distinguish missing memory files from empty reads #125979
Codex transcript mirrors survive matching redaction rules #126245
Preserve partial results and corpus outcomes in memory tools #126530
Detect offline memory source drift in status #126535
Keep Memory guidance aligned with configured sources #126552
Enforce memory embedding deadlines after event-loop stalls #126667
Bound Memory Wiki model context #126779
fix: prevent stale transcript projections from publishing #126947
fix(tui): recover assistant responses after reconnecting to replacement runs #128221
fix(embeddings): reject empty vectors from compatible providers #128480
fix(google): support stable Gemini Embedding 2 request contracts #128716
fix(llama-cpp): preserve managed local embedding batch capacity #128721
fix(tui): keep foreign session resets from clearing active chats #128723
fix(google): support all documented Gemini embedding dimensions #129038
Preserve document order for indexed embedding responses #129227
fix(tui): prevent stale session actions from mutating replacement conversations #129681
fix(memory): recover indexing when embedding providers reject oversized batches #129927
fix(android): restore rewind and fork for attachment-only messages #129951
fix(tui): preserve run and session ownership across async events #130147
fix(embeddings): preserve provider endpoint query parameters #130211
fix(tui): isolate replacement sessions from stale async results #130227
fix(memory): stop false Copilot index mismatch warnings #130603
fix(memory): keep indexing after memory folder replacement #130698
fix(google): preserve gateway queries in embedding batches #130768
fix(memory): honor mixed-case embedding header overrides #130774
fix(doctor): preserve local memory diagnosis #130984
fix(memory): bound the embedding cache with a built-in default #131121
fix(active-memory): restore interactive recall and preserve completed results #131302
fix(memory): ignore vector debt for FTS-only indexes #131469
fix(tui): retain reset confirmation after history refresh #131557
fix(llama): honor configured embedding model during setup #132557
fix(memory): preserve wiki corpus in direct agent scopes #133234
fix(memory): report only failed index refreshes #133239
fix(memory): honor configured primary search result limits #133254
Retry memory embeddings when providers reject dimensions #69707
Preserve relevance for vector-only multimodal memory results #92196
Restore same-agent access to unmapped live transcripts #92261
Preserve useful keyword-only memory results without displacing stronger matches #92524
Let slow QMD memory searches use their configured timeout #95757
Keep reset and deleted sessions searchable without a restart #96132
Keep custom Ollama memory indexes tied to the correct endpoint #97059
Keep local semantic memory working after Homebrew Node upgrades #99318
Handle missing files correctly in allowed extra paths 2d06d46
fix(github-copilot): add timeout to embedding model discovery (#102886) ff15856
Reject queued local embedding requests during worker shutdown #102451
Keep truncated QMD errors valid Unicode #102547
Keep Active Memory log truncation Unicode-safe #102551
Keep Active Memory search queries UTF-16 safe #102621
Keep Active Memory emoji truncation UTF-16 safe #102877
Accept max thinking in Active Memory configuration #103614
Skip provider setup for blank memory searches #103728
Keep memory-host redacted token hints Unicode-safe #103813
Accept leading-zero lengths in Memory Host JSON responses #105916
Preserve Unicode text in fragmented QMD output #107263
Treat memory provider none as intentional FTS-only mode #107778
Stop stalled Obsidian helpers after 10 seconds #109232
Preserve thinking and verbose preferences across session rollover #109788
Keep Voyage embedding batch waits within their timeout #111673
Stop Memory Wiki doctor errors for keyed agents #112541
Restore global-session search for non-default agents #113919
Fully disable derived memory full-text indexes #115685
Reject malformed QMD memory search results #115705
Stop memory reads from offering phantom continuation pages #115722
Return cleanly for blank memory searches #116034
Stop embedding work after client disconnects #116146
Suggest reindexing for a missing QMD index #116958
Reject empty embedding requests before provider setup #117615
Keep Memory Wiki search within protected recall visibility #118265
fix(memory): score LIKE-fallback keyword hits as 0 instead of 1 #120603
fix(memory): a memory_search that hits its own deadline reports a broken embedding provider #121073
Reject invalid extra memory paths and trim test facades #121696
Accept Windows casing for explicit memory files #121976
Return a clear wiki_get error for malformed lookup input #122549
report clear memory indexing outcomes #123863
Report indexed SQLite sessions truthfully #124834
Reject unknown memory agent selections with runnable guidance #126570
fix(sessions): exclude archived trajectory artifacts #126912
fix(memory): enforce canonical SecretRef resolution #127699
fix(memory-wiki): ignore structural markers in query text #128105
fix(doctor): suppress skipped memory warnings for all agents #128209
fix(memory): search warns about stale index during session catch-up #128894
fix(lmstudio): preserve embedding index across API key rotation #129076
fix(openai): preserve embedding index across proxy key rotation #129177
fix(openai): preserve Azure embedding index across key rotation #129516
fix(memory): partition Mistral and DeepInfra embedding caches #129669
fix(ollama): preserve namespaced embedding query instructions #129706
fix(memory): suppress wiki not-registered warning for corpus=all (#129866) #130088
fix(doctor): align memory provider guidance #130756
fix(memory): indexing fails when provider API key uses a SecretRef #131351
Fix ordered LanceDB queries with narrowed output columns #133120
fix: honor embedding dimensions when memory search is disabled #133158

Documentation

Explain overlapping memory recall sources #118299

Importing memories and conversation history

Importing is separate from the QMD-to-built-in upgrade. You can bring supported memory from Codex, Claude Code, or Hermes into an agent workspace through the Control UI, first-run setup, or Ask OpenClaw while leaving the source alone and not sweeping in credentials, settings, skills, or arbitrary provider files. If the destination already contains conflicting content, replacement has to be reviewed explicitly.

Old conversations follow their own preview-first path. The CLI shows what it would stage before it writes, the Control UI reports bounded-batch progress, and material owned by that import can be rolled back and applied again. Large histories use bounded or indexed processing, Memory Wiki preserves human notes through supported imports, and older history without complete ownership tracking may need to be rescanned. Staged material still does not become durable memory until dreaming or an explicit promotion chooses it.

Sources and complete change list

Improvements

Import Codex and Claude Code memory from the Control UI #106406
Offer memory imports during onboarding #108977
Backfill retained session transcripts into memory staging #115162
Backfill memory from past sessions in Control UI #115266
Import local agent memories from Ask OpenClaw #115376
Lower Memory Core startup memory use #119372

Bug fixes

Preserve Memory Wiki notes when imported pages are pruned #102294
Bound large SQLite session-history reads #108851
Harden memory migration path resolution and file copying #109314
Preserve Memory Wiki pages during temporary source outages #111034
Preserve Memory Wiki edits during ChatGPT import rollback #116517
Preserve Memory Wiki notes when markers are damaged #117104
Protect Memory Wiki notes and private recall boundaries #118375
Keep Gateway RPCs responsive after startup #119710
Keep large chat histories from blocking startup #124994
Keep large CLI-bound chat histories responsive #126201
fix(agents): wait for deferred transcript projection before reopening the session #130930
Prevent Memory Wiki source ingest from losing content during concurrent writes #104209
Harden memory migration plan selection and apply safety #109323
Redact Claude CLI history before Control UI merge #115597
Make memory session backfill complete and reversible #115926
Preserve Memory Core transcript-ingestion checkpoints #117293
Prevent duplicate Claude CLI replies after reload #125030
fix(memory-wiki): persist ChatGPT import run record before compiling the vault #126487
Stop duplicate Memory Wiki imports during polling #91828
Make legacy session backfill rollback re-applicable #116004
fix(gateway): chat history shows injected skill instructions on claude-cli sessions #130269
fix(gateway): images sent to Claude CLI sessions show as raw cache-path links in chat history #130463

Reviewing, creating, and forgetting memories

Settings now has a Memory destination for live status, Dreams, configuration, indexed-source search and browsing, add-on controls, and embedding readiness, so you can see what your Claw has available and open safe workspace memory files. Memory Wiki overview and Imported Insights dashboards now read published snapshots instead of reparsing the whole vault for every request, keeping large views responsive and showing when a rebuild is running or a manual compile is required. Session and legacy sources remain snippet-only, and add-on changes remain read-only unless the connection has admin authorization.

Eligible observations can be consolidated into MEMORY.md, durable directives can live in USER.md, standing intents can wait for the right event, and project memories stay scoped to the project that produced them. Automatic memory remains bounded and provenance-gated, so content derived from network or restricted sessions keeps an untrusted origin and stays out of automatic context even though an explicit search can still surface it. This provenance protection applies to newly tracked material, while older untracked files retain their existing classification.

openclaw memory forget lets you preview a purge by session, hook source, or participant, then remove attributable derived memory and stop the selected session from being pulled back in by backfill, indexing, or dreaming. An interrupted purge can be retried, and newly attributable consolidation or backfill diary claims are removed with the selected session. The purge follows recorded provenance, so the original transcript, older lineage-free notes, other agents' stores, direct or external writes, exports, and backups may remain. Review the preview before applying it.

Sources and complete change list

Improvements

Add a dedicated Memory settings page #114037
Add provenance-gated memory and default-on dreaming #114819
Give Memory a live overview, Dreams, and Settings #115383
Add memory search and browsing to Settings #115419
Isolate curated memory by project #115438
Add inline controls and readiness checks to Memory settings #115557
Retain recent project memory scopes within a session #115731
feat(memory): session provenance, admission policy, and openclaw memory forget #130151
Add operator-scoped memory search to the gateway #115346
Rename Memory Palace to Memory Wiki and document it #115954
Align the Memory page and simplify agent selection #115884

Bug fixes

Protect long-term memory during failed automatic promotions #108397
Restore per-agent Memory Core dream diaries #115069
Keep dreaming consolidation within project boundaries #115721
Quarantine memory writes from network-tainted agent turns #115818
Preserve user-authored headings during MEMORY.md compaction #116057
Preserve user notes during MEMORY.md budget compaction #116180
Mark post-network provider memory writes as untrusted #118760
Keep synthetic handoffs out of trusted owner memory #118987
Correct false memory save and forget claims #120989
Publish Memory Core dreaming reports atomically #122343
Revoke retained memory tools after memory is disabled #125393
Enforce turn tool policy during automatic memory recall #126482
Preserve memory trust across session capture and dreaming #126489
fix(memory): respect provenance in automatic context #127469
fix(memory): prevent forgotten session content from returning #130451
fix(memory): forgotten content survives retries and consolidation #131179
fix(memory-wiki): publish dashboard snapshots #132296
Keep emoji intact in startup memory and heartbeat prompts #102483
Keep compaction and memory snippets Unicode-safe #102542
Prevent partial Memory Wiki failures during concurrent writes #103408
fix(memory-core): dreaming narrative subagent receives full generic prompt instead of minimal #106404
Prevent raw output from breaking Tokenjuice compaction #107705
Preserve linked MEMORY.md files during automatic promotion #108921
Keep heartbeat replies out of new Dreaming memories #109403
Keep invalid recall timestamps out of dreaming rankings #110537
Preserve provider routing during safeguard compaction #112416
Bound compaction-planning worker payloads #112593
Keep memory-maintenance turns out of chat #112942
Keep media presentation markers out of LanceDB memories #113145
Correct the Dreaming toggle confirmation #113960
Remove dangling Memory Core recall records during repair #114419
Recover memory dreaming Cron after startup failure #115174
Give Memory settings stable path-based tab URLs #115515
Promote durable memories at calibrated defaults #115715
Make Memory engine switching clear and reliable #115750
Preserve network-content taint across transcript runtimes #115850
Stabilize memory-core dreaming schedules and cleanup #115925
Remove duplicate Memory Dreaming jobs after copied installs #115986
Honor user timezone in session-memory files #116136
Prevent memory flush from ending the parent session #116198
fix(memory-core): filter junk topics from REM phase extraction (#111923) #117248
Prevent stalled Git lookups from hanging agent runs #117346
Keep Dreams data with the selected agent #117538
Clean up interrupted Dreaming sessions after restart #117885
Keep valid memory diagnostics ahead of malformed timestamps #118749
Keep memory-flush appends from ending in bridge errors #120404
Keep replies running when memory flush preparation fails #121157
Preserve the previous Memory Wiki lint report on write failure #122568
CLI and Gateway agent turns now flush memory #124964
Keep session-memory filename generation tool-free #125394
Keep Dream Diary data scoped to the active agent #125936
fix(memory-core): restore MEMORY.md when the in-place fallback write fails midway #126486
fix(memory-core): read dreaming narrative from the terminal reply instead of racing the session store (#123360) #127184
fix(agents): stop cancelled compactions before model requests #128215
fix(agents): prevent model requests after context processing is cancelled #128607
fix(compaction): share model-aware memory maintenance budgets #130072
fix(memory): stop suggesting unavailable scheduling tools #130128
fix(memory): preserve line endings when forgetting entries #130943
fix(memory): forgotten facts survive in consolidation history #131248
fix(memory): forget diary claims after session backfill #131626
fix(memory): preserve session excerpts across resets #132757
Save session memory on automatic rollover #61675
Prevent nested compaction from locking its own transcript #88919
Honor the configured model for session-memory slugs #97007
Contain forged role lines in session-memory records #98574
Preserve valid Unicode in Memory Core dreaming snippets #101946
Keep Talk memory snippets valid at emoji boundaries #102477
Keep truncated memory snippets valid around emoji #102478
Preserve emoji when truncating dream-diary context #102524
Keep memory-wiki terminal truncation Unicode-safe #103178
Strictly parse Dreaming integer settings #107619
Reject invalid agent targets in memory doctor calls #115365
Preserve complete Unicode characters during memory promotion #115591
Preserve descriptive session-memory filenames after manual resets #116094
Keep Dreams diary navigation visible while scrolling #118054
Let memory-backed CLI commands exit naturally #118500
Return validation errors for malformed wiki_apply input #123050
Report failed Memory Core repairs and skipped promotions accurately #124770
Keep the Memory hub header stable across tabs #125568
Stop warning for expected non-owner memory-tool gating #126679
test(agents): bind process registry cleanup to its module instance #129456
Show total signals in memory promotion audit output #87590

Documentation

docs(memory): add memory provenance and deletion concepts page #130278
Align proactivity and memory docs with shipped behavior #112295
docs(memory): document shipped turn-taint propagation for network tool output #129837

Was this useful?