superlocalmemory 4.1.20 → 4.1.22

package/.claude-plugin/marketplace.json

@@ -9,7 +9,7 @@
       "author": {
         "name": "Qualixar"
       },
-      "description": "Local-first agent memory + reversible context compression and KV cache, as an MCP server. 38-tool code profile with graph intelligence.",
+      "description": "Local-first agent memory + reversible context compression and KV cache, as an MCP server. Serves the 56-tool full profile by default; set SLM_MCP_PROFILE=code for the 38-tool code profile.",
       "homepage": "https://github.com/qualixar/superlocalmemory",
       "keywords": [
         "memory",
@@ -21,7 +21,7 @@
       "license": "AGPL-3.0-or-later",
       "name": "superlocalmemory",
       "source": "./plugin",
-      "version": "4.1.20"
+      "version": "4.1.22"
     },
     {
       "author": {
@@ -39,7 +39,7 @@
       "license": "AGPL-3.0-or-later",
       "name": "superlocalmemory-codex",
       "source": "./codex-plugin",
-      "version": "4.1.20"
+      "version": "4.1.22"
     }
   ]
 }

package/CHANGELOG.md

@@ -5,6 +5,223 @@ All notable changes to SuperLocalMemory will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [4.1.22] — Faster recall from the first second, memories that stay findable, deletes you can trust
+
+### Added
+
+- **Filter by exact tags.** `recall`, `search`, `list` and `slm recall/list`
+  take a tag filter, over MCP, HTTP and the daemon proxy. A tag filter searches
+  inside the tagged memories and says why a result is empty.
+- **Switch the embedding model without losing recall.** `slm embedder`
+  re-indexes in the background, keeps the previous vectors, gives way to
+  recall while it works, resumes after a crash and can roll back.
+- **Model recommendations for this computer.** A model catalogue with current
+  defaults and each local model's result in SLM's extraction test; the setup
+  wizard offers your installed Ollama models, best first, and `slm models` and
+  the dashboard suggest installed and recommended models.
+- **Every recall says whether its answer was checked**: supported,
+  unsupported or unjudged, and why.
+- **`slm db integrity` and `slm db repair`** show store health in five parts
+  and fix leftover rows, erased words, unfinished deletes and memories that
+  lost their searchable fact, with receipts and undo. A repair never brings
+  back anything that was erased, deleted or withheld.
+- **`slm corrections overtaken`** lists corrections SLM had only proposed
+  itself that a delete, replace or edit overtook; `restore-overtaken` puts one
+  back.
+- Older memories in an upgraded store are classified by kind once, on this
+  device, in the background.
+- SuperLocalMemory installs as a Cursor-format plugin for Grok Bot, with a
+  low-RAM profile for shared bot hosts and two skills for bots sharing one
+  computer.
+
+### Changed
+
+- **Recall is about 2.6 times faster, with the same answers.** On a copy of a
+  real store, typical recall went from 849 ms to 321 ms (slowest 5%: 1.2 s to
+  0.49 s), and on 124 held-out questions from 923 ms to 340 ms (slowest 5%:
+  2.0 s to 0.77 s). Every speed change returns the same top ten memories as
+  before on all 184 test questions.
+- **Fast from the first second after a start, and while you save.** The store
+  is read into memory once at start, graph rankings and stored vectors are
+  prepared in separate background processes, background work yields to a
+  running recall at fine grain, requests no longer open the database to check
+  the login setting, and several slow store queries now use indexes. On a
+  22,000-memory store with the answer check on, a recall every second and a
+  save every half second: recalls slower than 3 s went from 13 of 318 to 7 of
+  297, with none in the first minute after a start (3 before), and saves slower
+  than 1.5 s from 7 of 56 to 1 of 87.
+- **Background enrichment keeps up.** Saved memories finish enrichment while
+  you keep recalling, instead of waiting for a quiet moment that never came:
+  about six times as many per minute under steady recall.
+- **A stated rule answers a yes/no permission question.** "May the agent
+  publish without approval?" is answered by a memory that says "Never publish
+  until the owner approves", with the same answer-check threshold as before;
+  the answer says it came from a rule. A rule about one named thing does not
+  answer a general question.
+- A memory that says it corrects an earlier one keeps that suggestion when the
+  model's check is unsure.
+- Local models copy the extraction prompt's example into your memories no
+  more, and invent fewer dates and numbers (llama3.2: sources with a fact the
+  source did not support went from 36 to 22 of 102; gemma3:4b from 1 to 0;
+  qwen2.5:7b 6 before, 7 now).
+- A fresh install (and CI) starts in Mode A.
+- `slm --json` and MCP tool responses are no longer pretty-printed.
+
+### Fixed
+
+- **A note that looked like an older one no longer loses its searchable fact.**
+  Enrichment deleted a new memory's only fact when it resembled another
+  memory's, so notes that differed only in a number or a declared kind could
+  not be found a moment after saving. `slm db repair` gives existing ones back.
+- **A question that mentions a time like "recently" still finds an older
+  memory it matches**, instead of returning nothing.
+- **A memory you save stays findable while SLM's own guess that it updates an
+  older one waits for review.** Approving the guess retires the older memory;
+  rejecting it keeps both. In one real store this made 406 saved memories
+  findable again; how often the right memory comes first is unchanged.
+- **Saves are not refused when the computer briefly stalls**, and a save
+  acknowledged just before the service was killed is enriched as soon as it
+  restarts.
+- **Deleting and erasing.** Erasing a memory removes its words everywhere
+  (keyword index, journal, event previews, entity summaries, logs, soft-forget
+  copies, vectors kept for a model switch). Erasing a person also erases the
+  memories they corrected, and erasing a whole profile no longer stops partway.
+  A delete refused at the last moment no longer leaves a memory unfindable.
+  Deleting a profile moves its memories to default, still findable, linked and
+  correctable.
+- A save sees the people and things it just added, so it does not create them
+  twice; a cancelled save leaves no trace in recall's lookups.
+- A background computation that times out, or whose service was stopped, no
+  longer keeps running.
+- Two recalls at the same moment are both reranked.
+- Measurements like 2004.6 ms are kept as measurements, never turned into
+  dates.
+- Bounded Loops runs with a node that never ran (an approval or a join) are no
+  longer rejected, and a gate verdict must be a real true or false.
+- The first memory call after a fresh start waits for SLM to finish starting;
+  a tool call no longer waits 60 s when two starts race; stopping a daemon
+  stuck in start-up no longer hangs.
+- Windows: SLM notices a stopped daemon at once, and Codex and Claude Code
+  hooks run SLM's own command.
+- Linux hosts with no GPU install CPU-only torch.
+- The remote listener closes a connection that floods data before its first
+  request; which cached answers get re-checked can no longer be predicted from
+  outside this install.
+- Plugin listings state the tool counts they really serve; the Hermes plugin
+  reaches the new `slm models`, `slm corrections` and `slm embedder` commands.
+
+### Known issues
+
+- On very large stores under heavy saving, a few recalls can still take longer
+  than 3 s while the store is busy committing a large write.
+
+
+## [4.1.21] — Recall that knows your project, and memories that stay findable
+
+### Added
+
+- **Recall knows your project** (#150). Memories saved for the project you are
+  working in rank ahead of cross-project trivia; `project=` filters to one
+  project, and a filter that matched nothing says so instead of returning an
+  empty list. Recall trace explains the same order recall returns.
+- **Saved views and summaries** (#113). Save a recall query as a named view and
+  re-run it from the CLI (`slm view`), MCP or the dashboard, with the same
+  answer from each. Daily, session and project summaries trace every line back
+  to the memories it came from.
+- **`slm provider set custom`** (#112): point Mode B or Mode C at your own
+  OpenAI-compatible server (llama.cpp, vLLM, LM Studio) from the CLI, with or
+  without a key. Its connection test goes through the same no-redirect gate as
+  every other request.
+- **Measure answer quality on your own memories.** Label your own questions and
+  see whether the right memory comes first, and set each answer check's
+  threshold from your own results instead of a default.
+
+### Changed
+
+- **The same question returns the same memories every time.** Linking
+  memories through the people, projects and things they share used to stop at
+  a time limit, so a busy machine could return different memories for the same
+  question. It now does a fixed amount of work in a fixed order, and a
+  background fill of the entity index after the upgrade keeps it fast. Until
+  that fill finishes, a few minutes after the first start, recall uses the
+  exact scan, which can be slower on a busy machine.
+- **Better first answers.** Memories a search actually found can no longer be
+  pushed out of the reranker's shortlist by loosely related neighbours, and
+  when the reranker clearly puts one memory first, recency and quality
+  adjustments can only break near-ties, not overturn it. On a large real
+  store: right memory first on 27 of 50 questions (was 23), and the right
+  memory in the top 3 on 17 of 97 held-out questions (was 11).
+- A question that resembles none of your memories no longer gets graph matches
+  started from unrelated memories, so the result is the same with and without
+  the vector extension.
+- **The semantic cache makes the same choice on every platform and every
+  run.** After a run of correct hits it serves a stored answer at similarity
+  0.98 and above every time, where before it skipped it at random (up to 43% of
+  the time at 0.98).
+- **Modes are named by what you get, not by a vendor** (#112), in setup, the
+  CLI and the dashboard.
+- **Remote access works on the key's profile for every tool**, whatever profile
+  this computer is using.
+- In Settings, ticking a Jev option while Laya is chosen asks whether to
+  switch; "Try it" shows where the time went; a memory kind below the shared
+  confidence threshold shows as unconfirmed.
+- The dashboard's live events show saved, captured, corrected and deleted
+  memories from every surface, and its filter offers only events SLM sends.
+- Every dashboard text meets readable contrast in light and dark themes.
+- The README leads with the answer check (Laya on your Mac, Jev on Windows,
+  Linux and macOS), lists platform support plainly, and brings back the
+  architecture diagram and the walkthrough video.
+
+### Fixed
+
+- **Background maintenance no longer moves memories you still use into the
+  archive**, where recall could not find them. On a large store most memories
+  had drifted there, and questions about them came back with the wrong answer
+  or none.
+- A question about a version number ("what changed in 4.1.20") finds the memory
+  about that version.
+- A new or changed link is used by the next recall instead of a cached answer.
+- Erasure: unfinished deletions are re-checked and closed in every profile,
+  back off instead of retrying forever, heal entity erasures, and an unreadable
+  vector store never counts as erased (PR #149, thanks @bgausden). `slm ops`
+  and the Ops Health page say what each entry is and what to do.
+- `slm ops resolve` rejects a bad operation ID cleanly instead of crashing
+  (#148); a broken daemon response is no longer reported as a bad ID.
+- Re-verified the Mode B / Ollama, daemon-start and migration fixes reported in
+  #128 and #125.
+- On a computer shared by several accounts, SLM never sends data to, starts
+  against, or adopts another account's daemon.
+- Restore points, backups, remote secrets, the answer-check key and config are
+  readable only by you, including files older versions left open.
+- Remote access closes a connection that stalls before TLS within 10 seconds.
+- Shutting down while another program holds the memory database no longer
+  fails; a save waiting for it is kept for the next start.
+- **Windows:** output to a pipe or file is UTF-8 and never crashes on a
+  character; the dashboard and SLM's files read and write UTF-8; process checks
+  never send Ctrl+C; profiles, restores, embedding migrations, backups and the
+  benchmark close their database files so they can finish; the code graph
+  records forward-slash paths; bounded-loops receipts import; the Answer Check
+  demo says plainly that it needs an Apple silicon Mac.
+- **macOS:** no native matrix factorization on the recall and cache paths can
+  corrupt process memory.
+- **Windows:** `slm serve stop` and the idle timeout shut the daemon down
+  cleanly instead of killing it, so it leaves no stale state behind; a model
+  download started by the daemon's self-repair stops when the daemon stops.
+- The Summaries picker lists projects saved on your memories, not only the
+  folders an agent worked in.
+- Each dashboard page opens at its first line instead of under the top bar.
+
+### Security
+
+- Installs and upgrades get fsspec 2026.6.0 or later, which closes a remote
+  code execution advisory (GHSA-27vj-qcqg-25rc) in a package SLM uses
+  indirectly.
+- `config.json` and the answer-check and feedback keys that an older release
+  left readable by other accounts are made private when SLM starts.
+
+Thanks to @bgausden for the erasure fix in PR #149 and for #148 and #150, to
+@tianjueak for #128, and to @unfall103-debug for #112, #113 and #125.
+
 ## [4.1.20] — Answers you can check, from any computer
 
 ### Added