What supermemoryai/supermemory shipped
Written by FoxPlug from the repository's own releases; not affiliated with Supermemory. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/supermemoryai/supermemory; Supermemory did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of November 2, 2026
What shipped
- Next.js was bumped from 16.3.8 to 16.4.1 for the web app, patching an unauthenticated remote code execution advisory, a Server Actions denial of service advisory and MCP SDK advisories. Pull request #1845
- The web app was switched to OpenNext's static-assets cache so production deploys no longer fail while uploading the R2 cache, and OpenNext was bumped so next 16.4.1 stops returning 500 on every page. Pull request #1862
- DM Sans was dropped from the web app so production builds stop fetching Google Fonts URLs that next/font/google under Turbopack cannot parse. Pull request #1866
- The MCP graph tools now read the v5 namespace graph feed instead of the legacy v3 documents route, paging by cursor. Pull request #1871
- The docs were synced with the cursor pagination of the v5 list endpoint, which ships in the TypeScript SDK from 5.1.1. Pull request #1870
Why it matters
This week production deploy failures were addressed, and the security patch for the Next.js remote code execution advisory is now able to go live. The MCP graph tools also moved off the last legacy call to the v5 namespace graph feed, with paging by cursor. Developers get a more reliable build and deploy path and docs that match the v5 list endpoint's cursor pagination.
Changelog entry
- Bumped next in apps/web from 16.3.8 to 16.4.1, patching GHSA-673q-mjm6-r5xr and GHSA-2464-fhr4-w9xw (unauthenticated RCE), GHSA-p3m5-59mp-jv5r (Server Actions DoS) and MCP SDK advisories. Pull request #1845
- Switched apps/web to OpenNext's static-assets cache, which needs no R2, so production deploys stop failing on the R2 cache upload. Pull request #1862
- Bumped OpenNext to 1.20.15 so next 16.4.1 stops returning 500 on every page. Pull request #1862
- Dropped DM Sans from apps/web so next build stops fetching Google Fonts URLs that Turbopack's next/font/google cannot parse. Pull request #1866
- MCP graph tools now read the v5 namespace graph feed at POST /ns/{namespace}/graph instead of POST /v3/documents/documents. Pull request #1871
- GetGraphDocuments pages by cursor in batches of 100. Pull request #1871
- Documented cursor pagination for the v5 list endpoint, shipping in the TypeScript SDK from 5.1.1. Pull request #1870
This week: patched the Next.js RCE advisory and MCP SDK advisories, fixed production deploys with OpenNext's static-assets cache, dropped DM Sans to unblock builds, moved the MCP graph tools to the v5 namespace graph feed, and synced docs with v5 cursor pagination.
This week: the web app's Next.js pin moved to 16.4.1 to patch an unauthenticated RCE advisory, a Server Actions denial of service advisory and MCP SDK advisories. Production deploys were unblocked by switching the web app to OpenNext's static-assets cache and bumping OpenNext so next 16.4.1 stops returning 500 on every page. DM Sans was dropped so builds stop fetching Google Fonts URLs that next/font/google cannot parse. The MCP graph tools now read the v5 namespace graph feed with cursor paging instead of the legacy v3 documents route. The docs were synced with cursor pagination for the v5 list endpoint, shipping in the TypeScript SDK from 5.1.1.
Week of October 26, 2026
What shipped
- The document listing and single document tools now return the existing dreaming status, and the add memory response explains that the document was accepted with memories extracted in the background. Pull request #1844
- The hosted worker version was bumped from 1.0.0 to 1.0.1 to follow the earlier API-key validation change, keeping the manifest, lockfile, server metadata, HTTP discovery response and app metadata in sync. Pull request #1858
- Dependency updates moved happy-dom to 21.0.4 and the zod 4 copies to 4.6.7, closing two dependency alerts. Pull request #1843
- A full docs drift audit compared the live code against the live docs across API and specs, concepts and guides, SDKs, coding-agent plugins, framework tools and pricing. Pull request #1855
- The Cursor, OpenCode and Muse Code docs pages now point to the console Agents page for managing, changing access to, or disconnecting the plugin. Pull request #1857
- The migration guide for Mem0 now names its dashboard without linking to it, matching how the site treats other named products. Pull request #1840
Why it matters
This week fixed a misleading confirmation message: adding a memory now reports that the document was accepted and extraction continues in the background, and the tools expose the existing dreaming status. The hosted worker also got a patch bump to stay consistent with the earlier API-key validation change, while dependency updates closed two alerts. Docs work covered a full drift audit and small fixes, and a new AGENTS.md gives coding agents correct project instructions in one place.
Changelog entry
- Document listing and single document tools now return dreaming status; add memory says the document was accepted and memories are extracted in the background. Pull request #1844
- Bumped the hosted worker from 1.0.0 to 1.0.1, keeping manifest, lockfile, server metadata, discovery response and app metadata in sync. Pull request #1858
- Bumped happy-dom to 21.0.4 and zod 4 copies to 4.6.7 to fix two dependency alerts. Pull request #1843
- Full docs drift audit across API and specs, concepts and guides, SDKs, coding-agent plugins, framework tools and pricing. Pull request #1855
- Cursor, OpenCode and Muse Code docs pages now point to the console Agents page for managing or disconnecting the plugin. Pull request #1857
- Mem0 migration guide now names the dashboard without linking it. Pull request #1840
This week: tool responses now say when a document is accepted and memories are extracted in the background, and document tools expose dreaming status. The hosted worker moved to 1.0.1, two dependency alerts closed, and docs went through a full drift audit.
This week's changes to the project: the add memory tool now reports what actually happened, that the document was accepted and memories are extracted in the background, and the document listing and single document tools expose the existing dreaming status. The hosted worker was bumped to 1.0.1 to stay in sync with the earlier API-key validation change, and dependency updates moved happy-dom to 21.0.4 and the zod 4 copies to 4.6.7. On the docs side, a full drift audit compared the live code against the live docs across API and specs, concepts and guides, SDKs, coding-agent plugins, framework tools and pricing. An AGENTS.md file was added at the repo root with instructions for coding agents, and the Mem0 migration guide now names that dashboard without linking it.
Week of October 12, 2026
What shipped
- Forget_memory now forgets memories by exact ID from
search_memoryinstead of semantic matching, cutting calls to about half a second instead of hitting the 30s timeout. Pull request #1820 - Supermemory Local 0.0.9 shipped with prebuilt Docker images for Linux amd64 and arm64, improved uploads and namespace operations, and API v5 support. Release
- Supermemory Local 0.0.9 now requires localhost requests and asks users to back up their store and keep their existing directory and API key before upgrading. Release
- @supermemory/tools 2.3.1 backports a fix so model-supplied containerTag stays inside the configured scope on the 2.x line. Pull request #1824
- @supermemory/tools 3.1.0 makes ai an optional peer covering versions 5, 6 and 7, so apps on newer AI SDK releases no longer get a second copy. Pull request #1794
- Text-only
forget_memorycalls now forget only exact matches, avoiding the old LLM matching path. Pull request #1820 - Docker installation and safe upgrade instructions were added for Supermemory Local, with a setup example, a self-hosting Docker page and a sidebar entry. Pull request #1801
- Docs now cover system.dreamingStatus, returned on v5 document reads and list items as dreaming or done, separate from system.status. Pull request #1797
Why it matters
This week fixed several MCP and tools problems that made ordinary calls fail or time out, and tightened scope handling in the 2.x tools line. It also added Docker-based installation for Supermemory Local with upgrade guidance, and brought the v5 docs in line with what the API actually accepts. Users on both the server and the Python wrappers get fixes that previously required workarounds.
Changelog entry
- Forget_memory forgets memories by exact ID from
search_memory, about 0.5s on production instead of the 30s timeout. Pull request #1820 - Supermemory Local 0.0.9 released with prebuilt Docker images for Linux amd64 and arm64, improved uploads, namespace operations, search scoping, CLI commands and API v5 support. Release
- Supermemory Local 0.0.9 requires localhost requests; back up your store and keep your existing directory and API key before upgrading. Release
- @supermemory/tools 2.3.1 backport keeps model-supplied containerTag inside the configured scope. Pull request #1824
- @supermemory/tools 3.1.2 bundles 8 community fixes adapted to v5. Pull request #1798
- @supermemory/tools 3.1.0 makes ai an optional peer covering versions 5, 6 and 7. Pull request #1794
- @supermemory/tools 3.1.1 saves AI SDK 7 images and keeps the provider peer optional. Pull request #1796
- Docker installation and safe upgrade instructions added for Supermemory Local, without a server-source checkout or registry login. Pull request #1801
- Docs document v5 system.dreamingStatus on document reads and list items, separate from system.status. Pull request #1797
- Docs use permanent redirects for the self-hosting and MCP paths. Pull request #1821
- MCP logs expected auth rejections at warn and reserves error for auth outages. Pull request #1822
Supermemory this week: forget_memory by exact ID runs in about 0.5s instead of the 30s timeout, Local 0.0.9 ships prebuilt Docker images for amd64 and arm64, and tools 2.3.1 keeps containerTag inside the configured scope.
This week in Supermemory: forget_memory now forgets memories by exact ID taken from search_memory, running in about half a second on production instead of hitting the 30s timeout. Supermemory Local 0.0.9 is out with prebuilt Docker images for Linux amd64 and arm64, improved uploads, namespace operations and API v5 support. On the tools side, 2.3.1 backports a fix that keeps model-supplied containerTag inside the configured scope, 3.1.2 bundles 8 community fixes adapted to v5, and 3.1.0 makes ai an optional peer so apps on newer SDK releases avoid a duplicate copy. Docker installation and safe upgrade instructions are now in the docs, and the v5 docs were aligned with the API across ingestion, search, connectors and self-hosting.
Week of October 5, 2026
What shipped
- @supermemory/tools 3.0 now calls the v5 API across every integration and retires @supermemory/ai-sdk, with config taking one namespace. Pull request #1773
- The server now runs on the v5 API and supermemory 5.0, with the namespace as the first argument on every tool. Pull request #1783
- Pre-snake_case tool names return as deprecated aliases so cached clients stop failing on the old names. Pull request #1782
- Removal requests now have a separate tool with its own schemas and destructive safety annotations. Pull request #1790
- The agent framework integration is prepared for 2.0.0 on the Python SDK. Pull request #1786
- The memory graph adds opt-in stacked nodes and emphasized connections, so the console no longer carries a patch. Pull request #1792
- Wrapper packages now cap the Python SDK below 5 so existing installs keep calling the 3.x API. Pull request #1774
- The docs add a versioned v5 API reference with legacy and latest versions organized by workflow. Pull request #1690
- A migration guide documents every migrated workflow, request and response change, filter conversion and rollout step. Pull request #1691
- SDK snippets across 50 pages move to the shipped v5 call shape and the namespace rename is finished. Pull request #1772
Why it matters
This week the project moved onto the v5 API and supermemory 5.0, and the tools package went to 3.0 with no aliases, so call shapes and config change for integrations. The migration guide, versioned API reference and rewritten snippets exist so you can upgrade deterministically, and the wrapper cap keeps older installs working while they catch up.
Changelog entry
- @supermemory/tools 3.0 calls the v5 API across every integration and retires @supermemory/ai-sdk; config takes one namespace. Pull request #1773
- The MCP server moves to the v5 API and supermemory 5.0 with the namespace as the first argument. Pull request #1783
- Pre-snake_case tool names are kept as deprecated aliases for clients with a cached tool list. Pull request #1782
- Removal requests have a separate tool with its own description, schemas and destructive annotations. Pull request #1790
- The agent framework integration is prepared for 2.0.0 on the Python SDK. Pull request #1786
- Wrapper packages pin the Python SDK below 5 and bump patch versions. Pull request #1774
- The memory graph gains opt-in stacked nodes and emphasized connections. Pull request #1792
- Docs add a versioned v5 API reference with legacy and latest versions. Pull request #1690
- Docs add a V3/V4 to V5 migration guide. Pull request #1691
- SDK snippets across 50 pages move to the shipped v5 call shape. Pull request #1772
- A one-line migrate tip appears on every v5 entry point. Pull request #1779
- Docs cover the paginated v5 namespace list and both delete outcomes. Pull request #1769
Tools 3.0 and the server now run on v5, with the namespace as the first argument and a separate tool for removals. Old tool names come back as deprecated aliases so cached clients stop failing. Migration guide and versioned API reference are out.
This week the MCP server moved to the v5 API and supermemory 5.0, with the namespace as the first argument on every tool, and @supermemory/tools 3.0 brings every integration onto v5 while retiring @supermemory/ai-sdk. Removals now have a separate tool with its own schemas and destructive annotations, and the pre-snake_case tool names return as deprecated aliases so cached clients stop failing. The wrapper packages pin the Python SDK below 5 so existing installs keep working. On the docs side, a versioned v5 API reference, a migration guide and rewritten snippets across 50 pages support the upgrade.
Week of September 28, 2026
What shipped
- Added supermemory-livekit, a Python plugin that gives LiveKit Agents voice sessions persistent memory by loading caller profiles and related memories before each reply. Pull request #1702
- Published documentation disclosing that server-v0.0.8 binds all interfaces with unsafe implicit local authentication and warns operators against exposing it to untrusted networks. Pull request #1711
- Upgraded Better Auth to patched version 1.7.6 and applied root dependency security changes. Pull request #1719
- Patched root workspace security vulnerabilities including Next 16.3.3, Drizzle 0.45.2, Vite 7.3.5 and related packages. Pull request #1724
- Fixed out-of-credits error messaging to direct users to the billing console and prevented these errors from inflating error rates. Pull request #1753
- Refreshed documentation design to use console design tokens including Geist, neutral greys, updated radii and borders in light and dark mode. Pull request #1715
- Added team invitation guide documenting how to invite team members with role and access choices. Pull request #1754
- Instrumented MCP server with PostHog analytics sending only allowlisted metadata with personless user IDs and disabled schema injection. Pull request #1712
- Documented advanced PDF extraction availability on Scale and Enterprise plans including page-by-page OCR and automatic tier selection. Pull request #1728
- Enabled conversation ids in PostHog MCP tool calls to group related calls instead of creating separate sessions for each call. Pull request #1718
Why it matters
This week brings security patches across the entire dependency stack, a new LiveKit integration for voice sessions, and significant documentation improvements. The disclosure of self-hosted security behavior and telemetry practices reflects commitment to transparency for operators deploying Supermemory.
Changelog entry
- Feat(livekit): add persistent memory for LiveKit Agents Pull request #1702
- Docs: disclose current self-hosted security and telemetry behavior Pull request #1711
- Fix(auth): upgrade Better Auth to patched 1.7.6 Pull request #1719
- Fix(deps): patch root workspace security vulnerabilities Pull request #1724
- Fix(mcp): clear out-of-credits error and keep it out of error rate Pull request #1753
- Docs: refresh the documentation design Pull request #1715
- Docs: add team invitation guide with console screenshots Pull request #1754
- Feat(mcp): instrument server with metadata-only PostHog analytics Pull request #1712
- Docs: explain advanced PDF extraction availability Pull request #1728
- Feat(mcp): group PostHog MCP tool calls by conversation id Pull request #1718
- Fix(docs): remove vulnerable ZIP extractor and patch tar via Mintlify Pull request #1727
- Fix: docs icons in dark mode Pull request #1735
Ship day: LiveKit voice agent memory, Better Auth security patch, advanced PDF extraction docs, and dependency security fixes across the stack.
This week we shipped persistent memory for LiveKit Agents voice sessions, security patches across the dependency stack including Better Auth 1.7.6, refreshed documentation design, improved team invitation workflow, and PostHog analytics instrumentation for better MCP observability. We also published transparency disclosures about self-hosted security behavior and documented advanced PDF extraction capabilities on Scale and Enterprise plans.
Week of September 14, 2026
What shipped
- Memory graph 0.2.4 released with fixes for theme rendering, initial node layout, and settling behavior. Pull request #1686
- Public MCP tools now use
snake_casenaming, andsearch_memoryno longer fetches full profile data on every call. Pull request #1665 - Billing and Gmail documentation updated to reflect Max plan ($100/month with $130 in monthly credits) and correct feature availability across plans. Pull request #1685
- Muse Code plugin documentation added, covering marketplace install, file-based auth, lifecycle hooks, and MCP tools. Pull request #1674
- Added 35 documentation redirects for stale URLs identified in Google Search Console Pages report. Pull request #1673
- Graph styling restored with configurable dots, and incoming nodes now automatically settle and fit without manual intervention. Pull request #1684
- Eve integration guide added and integration icons refreshed with updated plugin ordering. Pull request #1671
- CI auto-fix workflow now restricted to same-repository pull requests from human actors. Pull request #1663
- Documentation build process now runs on isolated Node 22 runtime for local development compatibility. Pull request #1692
Why it matters
The memory graph got visual and performance improvements while the plugin ecosystem expanded with Muse Code documentation. Documentation accuracy received significant attention through billing updates and SEO fixes, and internal CI workflows were refined to prevent unintended automated actions.
Changelog entry
- Memory graph 0.2.4 released with theme fixes, initial node fit, and node settling improvements Pull request #1686
- MCP public tools now use
snake_casenaming;search_memoryreturns memories only, profile data moved toget_profilePull request #1665 - Billing and Gmail documentation updated to reflect Max plan and current feature entitlements Pull request #1685
- Muse Code plugin documentation added with marketplace install, auth, lifecycle hooks, and MCP tools Pull request #1674
- 35 documentation redirects added for stale /docs URLs Pull request #1673
- Graph console theme and dots restored; incoming nodes automatically settle and fit Pull request #1684
- Eve integration guide added; integration icons refreshed and plugin order updated Pull request #1671
- CI auto-fix workflow restricted to same-repository pull requests from human actors Pull request #1663
- Documentation development now runs on isolated Node 22 runtime Pull request #1692
This week: memory graph 0.2.4 with theme and layout fixes, MCP tools now using snake_case, and 35 doc redirects for stale URLs. Billing docs updated with Max plan details.
Weekly update: memory graph 0.2.4 shipped with theme restoration, automatic node settling, and improved initial layout. MCP tools refactored to snake_case naming and search_memory optimized to reduce unnecessary data fetching. Documentation expanded with Muse Code plugin guide, updated billing information for Max plan, and 35 redirects addressing stale URLs from GSC audit.