[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"navigation":3,"url-settings":86,"updates-\u002Fupdates\u002Fnano-models":671,"updates-related-\u002Fupdates\u002Fnano-models":1724,"i-material-symbols:arrow-back-rounded":1725},{"id":4,"extension":5,"footer":6,"header":73,"meta":83,"stem":84,"__hash__":85},"navigation\u002Fdata\u002Fshared\u002Fnavigation.yml","yml",{"brand":7,"columns":13,"legal":63},{"name":8,"tagline":9,"downloadCta":10},"Pieces","The memory layer for modern work.",{"label":11,"href":12},"All downloads","\u002Fdownloads",[14,26,39,51],{"title":15,"links":16},"Product",[17,20,23],{"label":18,"href":19},"Pieces Desktop","\u002F",{"label":21,"href":22},"Pieces Enterprise","\u002Fenterprise",{"label":24,"href":25},"Pieces MCP","\u002Fmcp",{"title":27,"links":28},"Resources",[29,33,36],{"label":30,"href":31,"external":32},"Documentation","url:docs.home",true,{"label":34,"href":35},"Blog","\u002Fblog",{"label":37,"href":38,"external":32},"GitHub","url:github.org",{"title":40,"links":41},"Community",[42,45,48],{"label":43,"href":44,"external":32},"Discord","url:social.discord",{"label":46,"href":47,"external":32},"X \u002F Twitter","url:social.x",{"label":49,"href":50,"external":32},"LinkedIn","url:social.linkedin",{"title":52,"links":53},"Company",[54,57,60],{"label":55,"href":56},"About","\u002Fabout",{"label":58,"href":59},"Updates","\u002Fupdates",{"label":61,"href":62},"Contact","\u002Fcontact",[64,67,70],{"label":65,"href":66},"Privacy Policy","\u002Flegal\u002Fprivacy",{"label":68,"href":69},"Refund Policy","\u002Flegal\u002Frefund",{"label":71,"href":72},"Terms of Service","\u002Flegal\u002Fterms",{"links":74,"signIn":75,"contact":78,"cta":80},[],{"label":76,"href":77},"Manage account","url:portal.home",{"label":79,"href":62},"Get in touch",{"label":81,"href":82},"Download","url:routes.downloads",{},"data\u002Fshared\u002Fnavigation","XlSALC7vYCXHcmcXnjJViwIZKsqnQWF4bn4o9RrcSBE",{"id":87,"extension":5,"links":88,"meta":668,"stem":669,"__hash__":670},"urlSettings\u002Fdata\u002Fshared\u002Furls.yml",[89,93,97,101,105,109,113,117,121,125,129,133,137,141,145,149,153,157,161,165,169,173,177,181,185,189,192,196,200,204,208,212,216,220,224,228,232,236,240,244,248,252,256,260,264,268,272,276,280,284,288,292,296,300,303,307,311,314,318,322,326,330,334,338,342,346,350,354,358,362,366,370,374,378,382,386,390,394,398,402,406,410,414,418,422,426,430,434,438,442,446,450,454,458,461,465,469,473,477,481,485,488,491,494,497,501,505,508,512,516,520,524,528,532,536,540,544,548,552,556,559,563,567,571,575,579,582,585,588,592,595,599,603,607,611,614,618,622,625,628,631,635,639,643,646,649,652,656,660,664],{"key":90,"label":91,"href":92},"downloads.desktop","Desktop download page","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fdesktop\u002Fdownload",{"key":94,"label":95,"href":96},"downloads.macOS.dmgArm64","macOS DMG Apple Silicon","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fpieces_for_x\u002Fdmg-arm64\u002Fdownload",{"key":98,"label":99,"href":100},"downloads.macOS.dmgIntel","macOS DMG Intel","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fpieces_for_x\u002Fdmg\u002Fdownload",{"key":102,"label":103,"href":104},"downloads.macOS.pkg","macOS PKG","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fmacos_packaging\u002Fpkg\u002Fdownload",{"key":106,"label":107,"href":108},"downloads.macOS.universal","Desktop macOS Universal","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fpieces_for_x\u002Fmacos-universal\u002Fdownload",{"key":110,"label":111,"href":112},"downloads.macOS.pkgArm64","Desktop macOS PKG Apple Silicon","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fmacos_packaging\u002Fpkg-pfd-arm64\u002Fdownload",{"key":114,"label":115,"href":116},"downloads.macOS.pkgIntel","Desktop macOS PKG Intel","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fmacos_packaging\u002Fpkg-pfd\u002Fdownload",{"key":118,"label":119,"href":120},"downloads.windows.appinstaller","Windows App Installer","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fappinstaller\u002Fpieces_for_x.appinstaller",{"key":122,"label":123,"href":124},"downloads.windows.exe","Windows EXE","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fpieces_for_x\u002Fwindows-exe\u002Fdownload",{"key":126,"label":127,"href":128},"downloads.windows.msix","Windows MSIX","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fpieces_for_x\u002Fwindows-msix\u002Fdownload",{"key":130,"label":131,"href":132},"downloads.windows.suiteManager","Windows Suite Manager","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fpieces_suite_windows\u002Fappinstaller\u002Fdownload",{"key":134,"label":135,"href":136},"downloads.linux.flatpakRepo","Linux Flatpak repository","https:\u002F\u002Fbuilds.pieces.app\u002Fpieces-flatpak-repo\u002Fpieces-flatpak.flatpakrepo",{"key":138,"label":139,"href":140},"downloads.linux.snapDesktop","Linux Snap Desktop","https:\u002F\u002Fsnapcraft.io\u002Fpieces-for-developers",{"key":142,"label":143,"href":144},"downloads.linux.snapPiecesOS","Linux Snap PiecesOS","https:\u002F\u002Fsnapcraft.io\u002Fpieces-os",{"key":146,"label":147,"href":148},"downloads.linux.snap","Desktop Linux Snap package","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fpieces_for_x\u002Fsnap\u002Fdownload",{"key":150,"label":151,"href":152},"downloads.piecesOS.macOS.dmgArm64","PiecesOS macOS DMG Apple Silicon","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fos_server\u002Fdmg-arm64\u002Fdownload",{"key":154,"label":155,"href":156},"downloads.piecesOS.macOS.dmgIntel","PiecesOS macOS DMG Intel","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fos_server\u002Fdmg\u002Fdownload",{"key":158,"label":159,"href":160},"downloads.piecesOS.macOS.universal","PiecesOS macOS Universal","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fos_server\u002Fmacos-universal\u002Fdownload",{"key":162,"label":163,"href":164},"downloads.piecesOS.macOS.pkgArm64","PiecesOS macOS PKG Apple Silicon","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fmacos_packaging\u002Fpkg-pos-launch-only-arm64\u002Fdownload",{"key":166,"label":167,"href":168},"downloads.piecesOS.macOS.pkgIntel","PiecesOS macOS PKG Intel","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fmacos_packaging\u002Fpkg-pos-launch-only\u002Fdownload",{"key":170,"label":171,"href":172},"downloads.piecesOS.windows.appinstaller","PiecesOS Windows App Installer","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fappinstaller\u002Fos_server.appinstaller",{"key":174,"label":175,"href":176},"downloads.piecesOS.windows.exe","PiecesOS Windows EXE","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fos_server\u002Fwindows-exe\u002Fdownload",{"key":178,"label":179,"href":180},"downloads.piecesOS.windows.msix","PiecesOS Windows MSIX","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fos_server\u002Fwindows-msix\u002Fdownload",{"key":182,"label":183,"href":184},"downloads.piecesOS.linux.snap","PiecesOS Linux Snap package","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fos_server\u002Fsnap\u002Fdownload",{"key":186,"label":187,"href":188},"downloads.combined.macOS.appleSilicon","Combined macOS installer Apple Silicon","https:\u002F\u002Fbuilds.pieces.app\u002Fstages\u002Fproduction\u002Fmacos_packaging\u002Fpkg-arm64\u002Fdownload",{"key":190,"label":191,"href":104},"downloads.combined.macOS.intel","Combined macOS installer Intel",{"key":193,"label":194,"href":195},"downloads.guides.macOS","macOS installation guide","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmeet-pieces\u002Fmacos-installation-guide",{"key":197,"label":198,"href":199},"downloads.guides.windows","Windows installation guide","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmeet-pieces\u002Fwindows-installation-guide",{"key":201,"label":202,"href":203},"downloads.guides.linux","Linux installation guide","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmeet-pieces\u002Flinux-installation-guide",{"key":205,"label":206,"href":207},"downloads.guides.linuxFlatpak","Linux Flatpak installation guide","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmeet-pieces\u002Flinux-installation-guide\u002Fflatpak",{"key":209,"label":210,"href":211},"downloads.guides.linuxSnap","Linux Snap installation guide","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmeet-pieces\u002Flinux-installation-guide\u002Fsnap",{"key":213,"label":214,"href":215},"downloads.guides.piecesOS","PiecesOS manual installation","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fcore-dependencies\u002Fpieces-os\u002Fmanual-installation",{"key":217,"label":218,"href":219},"extensions.chrome","Chrome extension","https:\u002F\u002Fchrome.google.com\u002Fwebstore\u002Fdetail\u002Fpieces-save-code-snippets\u002Figbgibhbfonhmjlechmeefimncpekepm",{"key":221,"label":222,"href":223},"extensions.firefox","Firefox add-on","https:\u002F\u002Faddons.mozilla.org\u002Fen-US\u002Ffirefox\u002Faddon\u002Fpieces-save-code-from-the-web\u002F",{"key":225,"label":226,"href":227},"extensions.edge","Edge add-on","https:\u002F\u002Fmicrosoftedge.microsoft.com\u002Faddons\u002Fdetail\u002Fpieces-save-code-snippet\u002Fhglfimcdgonaeeobjckfdabcldfidmim",{"key":229,"label":230,"href":231},"extensions.vscode","VS Code extension","https:\u002F\u002Fmarketplace.visualstudio.com\u002Fitems?itemName=MeshIntelligentTechnologiesInc.pieces-vscode",{"key":233,"label":234,"href":235},"extensions.visualStudio","Visual Studio extension","https:\u002F\u002Fmarketplace.visualstudio.com\u002Fitems?itemName=MeshIntelligentTechnologiesInc.PiecesVisualStudio",{"key":237,"label":238,"href":239},"extensions.jetbrains","JetBrains plugin","https:\u002F\u002Fplugins.jetbrains.com\u002Fplugin\u002F17328-pieces--save-search-share--reuse-code-snippets",{"key":241,"label":242,"href":243},"extensions.obsidian","Obsidian plugin","https:\u002F\u002Fobsidian.md\u002Fplugins?id=pieces-for-developers",{"key":245,"label":246,"href":247},"extensions.sublime","Sublime package","https:\u002F\u002Fpackagecontrol.io\u002Fpackages\u002FPieces",{"key":249,"label":250,"href":251},"extensions.neovim","Neovim plugin","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fplugin_neo_vim",{"key":253,"label":254,"href":255},"extensions.jupyterlab","JupyterLab plugin","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fjupyterlab-pieces",{"key":257,"label":258,"href":259},"extensions.cli","Pieces CLI","https:\u002F\u002Fpypi.org\u002Fproject\u002Fpieces-cli\u002F",{"key":261,"label":262,"href":263},"docs.home","Documentation home","https:\u002F\u002Fdocs.pieces.app",{"key":265,"label":266,"href":267},"docs.getStarted","Get started docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmeet-pieces",{"key":269,"label":270,"href":271},"docs.api","API docs","https:\u002F\u002Fdocs.pieces.app\u002Fapi",{"key":273,"label":274,"href":275},"docs.desktop.overview","Desktop overview","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fdesktop",{"key":277,"label":278,"href":279},"docs.desktop.onboarding","Desktop onboarding","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fdesktop\u002Fonboarding",{"key":281,"label":282,"href":283},"docs.desktop.timeline","Desktop timeline docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fdesktop\u002Ftimeline",{"key":285,"label":286,"href":287},"docs.desktop.summaries","Desktop summaries docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fdesktop\u002Fsingle-click-summaries",{"key":289,"label":290,"href":291},"docs.desktop.search","Desktop conversational search docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fdesktop\u002Fconversational-search",{"key":293,"label":294,"href":295},"docs.desktop.drive","Desktop drive docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fdesktop\u002Fdrive",{"key":297,"label":298,"href":299},"docs.desktop.account","Desktop account settings docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fdesktop\u002Fconfiguration\u002Faccount",{"key":301,"label":302,"href":92},"docs.desktop.download","Desktop download docs",{"key":304,"label":305,"href":306},"docs.piecesOS.overview","PiecesOS overview docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fcore-dependencies",{"key":308,"label":309,"href":310},"docs.piecesOS.details","PiecesOS details docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fcore-dependencies\u002Fpieces-os",{"key":312,"label":313,"href":215},"docs.piecesOS.install","PiecesOS install docs",{"key":315,"label":316,"href":317},"docs.piecesOS.quickMenu","PiecesOS quick menu docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fcore-dependencies\u002Fpieces-os\u002Fquick-menu",{"key":319,"label":320,"href":321},"docs.piecesOS.storage","On-device storage docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fcore-dependencies\u002Fon-device-storage",{"key":323,"label":324,"href":325},"docs.piecesOS.troubleshooting","PiecesOS troubleshooting docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fcore-dependencies\u002Fpieces-os\u002Ftroubleshooting",{"key":327,"label":328,"href":329},"docs.mcp.overview","MCP overview docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp",{"key":331,"label":332,"href":333},"docs.mcp.cursor","MCP Cursor docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fcursor",{"key":335,"label":336,"href":337},"docs.mcp.vscode","MCP VS Code docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fvs-code",{"key":339,"label":340,"href":341},"docs.mcp.claudeDesktop","MCP Claude Desktop docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fclaude-desktop",{"key":343,"label":344,"href":345},"docs.mcp.claudeCode","MCP Claude Code docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fclaude-code",{"key":347,"label":348,"href":349},"docs.mcp.claudeCowork","MCP Claude Cowork docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fclaude-cowork",{"key":351,"label":352,"href":353},"docs.mcp.githubCopilot","MCP GitHub Copilot docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fgithub-copilot",{"key":355,"label":356,"href":357},"docs.mcp.goose","MCP Goose docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fgoose",{"key":359,"label":360,"href":361},"docs.mcp.windsurf","MCP Windsurf docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fwindsurf",{"key":363,"label":364,"href":365},"docs.mcp.zed","MCP Zed docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fzed",{"key":367,"label":368,"href":369},"docs.mcp.jetbrains","MCP JetBrains docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fjetbrains-ides",{"key":371,"label":372,"href":373},"docs.mcp.continueDev","MCP Continue docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fcontinue-dev",{"key":375,"label":376,"href":377},"docs.mcp.cline","MCP Cline docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fcline",{"key":379,"label":380,"href":381},"docs.mcp.raycast","MCP Raycast docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fraycast",{"key":383,"label":384,"href":385},"docs.mcp.rovoDevCli","MCP Rovo Dev CLI docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Frovo-dev-cli",{"key":387,"label":388,"href":389},"docs.mcp.openaiCodexCli","MCP OpenAI Codex CLI docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fopenai-codex-cli",{"key":391,"label":392,"href":393},"docs.mcp.googleGeminiCli","MCP Google Gemini CLI docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fgoogle-gemini-cli",{"key":395,"label":396,"href":397},"docs.mcp.amazonQ","MCP Amazon Q docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Famazon-q-developer",{"key":399,"label":400,"href":401},"docs.mcp.chatgptDev","MCP ChatGPT Developer Mode docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fchatgpt-developer-mode",{"key":403,"label":404,"href":405},"docs.mcp.openclaw","MCP OpenClaw docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fopenclaw",{"key":407,"label":408,"href":409},"docs.mcp.mcpRemote","MCP Remote docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fmcp-remote",{"key":411,"label":412,"href":413},"docs.mcp.ngrok","MCP ngrok docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmcp\u002Fngrok-setup",{"key":415,"label":416,"href":417},"docs.troubleshooting.macOS","macOS troubleshooting docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmeet-pieces\u002Ftroubleshooting\u002Fmacos",{"key":419,"label":420,"href":421},"docs.troubleshooting.windows","Windows troubleshooting docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmeet-pieces\u002Ftroubleshooting\u002Fwindows",{"key":423,"label":424,"href":425},"docs.troubleshooting.linux","Linux troubleshooting docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fmeet-pieces\u002Ftroubleshooting\u002Flinux",{"key":427,"label":428,"href":429},"docs.privacy","Privacy and security docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fprivacy-security-your-data",{"key":431,"label":432,"href":433},"docs.support","Support docs","https:\u002F\u002Fdocs.pieces.app\u002Fproducts\u002Fsupport",{"key":435,"label":436,"href":437},"portal.home","Pieces portal","https:\u002F\u002Fportal.pieces.app",{"key":439,"label":440,"href":441},"site.home","Website home","https:\u002F\u002Fpieces.app",{"key":443,"label":444,"href":445},"site.about","About page","https:\u002F\u002Fpieces.app\u002Fabout",{"key":447,"label":448,"href":449},"site.features","Features page","https:\u002F\u002Fpieces.app\u002Ffeatures",{"key":451,"label":452,"href":453},"site.plugins","Plugins page","https:\u002F\u002Fpieces.app\u002Fplugins",{"key":455,"label":456,"href":457},"site.contact","Contact page","https:\u002F\u002Fpieces.app\u002Fcontact",{"key":459,"label":58,"href":460},"site.updates","https:\u002F\u002Fpieces.app\u002Fupdates",{"key":462,"label":463,"href":464},"site.news","News","https:\u002F\u002Fpieces.app\u002Fnews",{"key":466,"label":467,"href":468},"site.events","Community events","https:\u002F\u002Fpieces.app\u002Fcommunity\u002Fevents",{"key":470,"label":471,"href":472},"site.userStories","User stories","https:\u002F\u002Fpieces.app\u002Fuser-stories",{"key":474,"label":475,"href":476},"site.academy","Academy","https:\u002F\u002Fpieces.app\u002Flearn\u002Facademy",{"key":478,"label":479,"href":480},"site.support","Website support","https:\u002F\u002Fpieces.app\u002Fsupport",{"key":482,"label":483,"href":484},"site.standup","Standup","https:\u002F\u002Fpieces.app\u002Fstandup",{"key":486,"label":34,"href":487},"site.blog","https:\u002F\u002Fcode.pieces.app\u002Fblog",{"key":489,"label":43,"href":490},"social.discord","https:\u002F\u002Fdiscord.gg\u002Fgetpieces",{"key":492,"label":46,"href":493},"social.x","https:\u002F\u002Fx.com\u002Fgetpieces",{"key":495,"label":496,"href":493},"social.twitter","Twitter",{"key":498,"label":499,"href":500},"social.instagram","Instagram","https:\u002F\u002Fwww.instagram.com\u002Fgetpieces\u002F",{"key":502,"label":503,"href":504},"social.tiktok","TikTok","https:\u002F\u002Fwww.tiktok.com\u002F@getpieces",{"key":506,"label":49,"href":507},"social.linkedin","https:\u002F\u002Fwww.linkedin.com\u002Fcompany\u002Fgetpieces\u002F",{"key":509,"label":510,"href":511},"social.youtube","YouTube","https:\u002F\u002Fyoutube.com\u002F@getpieces",{"key":513,"label":514,"href":515},"github.org","GitHub organization","https:\u002F\u002Fgithub.com\u002Fpieces-app",{"key":517,"label":518,"href":519},"github.support","GitHub support","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fsupport",{"key":521,"label":522,"href":523},"github.issues","GitHub issues","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fsupport\u002Fissues",{"key":525,"label":526,"href":527},"github.discussions","GitHub discussions","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fsupport\u002Fdiscussions",{"key":529,"label":530,"href":531},"github.documentation","GitHub documentation","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fdocumentation",{"key":533,"label":534,"href":535},"github.opensource","GitHub open source","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fopensource",{"key":537,"label":538,"href":539},"github.sdks.python","Python SDK","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fpieces-os-client-sdk-for-python",{"key":541,"label":542,"href":543},"github.sdks.typescript","TypeScript SDK","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fpieces-os-client-sdk-for-typescript",{"key":545,"label":546,"href":547},"github.sdks.dart","Dart SDK","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fpieces-os-client-sdk-for-dart",{"key":549,"label":550,"href":551},"github.sdks.kotlin","Kotlin SDK","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fpieces-os-client-sdk-for-kotlin",{"key":553,"label":554,"href":555},"github.plugins.obsidian","Obsidian plugin repository","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fobsidian-pieces",{"key":557,"label":558,"href":255},"github.plugins.jupyterlab","JupyterLab plugin repository",{"key":560,"label":561,"href":562},"github.plugins.sublime","Sublime plugin repository","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fplugin_sublime",{"key":564,"label":565,"href":566},"github.plugins.neovim","Neovim plugin repository","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fplugin_neovim",{"key":568,"label":569,"href":570},"github.cliAgent","CLI agent repository","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fcli-agent",{"key":572,"label":573,"href":574},"github.mcpDart","MCP Dart repository","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fmcp_dart",{"key":576,"label":577,"href":578},"github.awesomePieces","Awesome Pieces repository","https:\u002F\u002Fgithub.com\u002Fpieces-app\u002Fawesome-pieces",{"key":580,"label":581,"href":66},"legal.privacyPolicy","Privacy policy",{"key":583,"label":584,"href":69},"legal.refundPolicy","Refund policy",{"key":586,"label":587,"href":72},"legal.terms","Terms",{"key":589,"label":590,"href":591},"legal.security","Legal security","https:\u002F\u002Fpieces.app\u002Flegal\u002Fsecurity",{"key":593,"label":594,"href":511},"videos.youtubeChannel","YouTube channel",{"key":596,"label":597,"href":598},"videos.gettingStartedDesktop","Getting started desktop video","https:\u002F\u002Fyoutu.be\u002FdUr1lRM_TYk",{"key":600,"label":601,"href":602},"videos.snippetDiscoveryxR","Snippet discovery video","https:\u002F\u002Fyoutu.be\u002FG6vb1USw-30",{"key":604,"label":605,"href":606},"sales.bookACall","Book a sales call","https:\u002F\u002Fcalendar.app.google\u002FWVUDtUfNy5Vst3sH7",{"key":608,"label":609,"href":610},"sales.enterprise","Enterprise form","https:\u002F\u002Fgetpieces.typeform.com\u002Fto\u002FaVQFTvpE",{"key":612,"label":613,"href":527},"sales.feedback","Feedback discussions",{"key":615,"label":616,"href":617},"sales.earlyAccess","Early access form","https:\u002F\u002Fgetpieces.typeform.com\u002Fearlyaccess",{"key":619,"label":620,"href":621},"sales.supportEmail","Support email","mailto:support@pieces.app",{"key":623,"label":624,"href":19},"routes.home","Home route",{"key":626,"label":627,"href":56},"routes.about","About route",{"key":629,"label":630,"href":12},"routes.downloads","Downloads route",{"key":632,"label":633,"href":634},"routes.thanks","Post-install thanks route","\u002Finstall\u002Fthanks",{"key":636,"label":637,"href":638},"routes.pricing","Pricing route","\u002Fpricing",{"key":640,"label":641,"href":642},"cta.freeTrial","Start free trial CTA (Pieces Pro) — used by \u002Fcampaigns\u002F* landing pages","https:\u002F\u002Fcampaigns.pieces.app\u002F",{"key":644,"label":645,"href":22},"routes.enterprise","Enterprise route",{"key":647,"label":648,"href":62},"routes.contact","Contact route",{"key":650,"label":651,"href":59},"routes.updates","Updates route",{"key":653,"label":654,"href":655},"routes.migrationError","Migration error route","\u002Fmigration\u002Ferror",{"key":657,"label":658,"href":659},"routes.authenticated","Post-login authenticated route","\u002Fauth\u002Fsigned-in",{"key":661,"label":662,"href":663},"routes.signed-out","Signed out route","\u002Fauth\u002Fsigned-out",{"key":665,"label":666,"href":667},"routes.authentication-error","Authentication error route","\u002Fauth\u002Ferror",{},"data\u002Fshared\u002Furls","2AvfSWI6bZACH5Kr1ZS3txMLjteRjCjbDR_ul1uRZ9Q",{"id":672,"title":673,"body":674,"category":15,"date":1710,"description":1711,"draft":1712,"extension":1713,"icon":1714,"iconColor":1715,"iconTone":1716,"image":1717,"imageAlt":1716,"meta":1718,"navigation":32,"ogImage":1716,"ogImageAlt":1716,"path":1719,"relatedResources":1716,"seo":1720,"stem":1721,"subtitle":1722,"summary":1716,"__hash__":1723},"updates\u002Fupdates\u002Fnano-models.md","Nano-Models power LTM-2.5",{"type":675,"value":676,"toc":1688},"minimark",[677,698,701,704,707,710,713,716,719,724,736,820,824,830,833,838,841,949,953,956,996,999,1003,1006,1010,1027,1031,1048,1052,1069,1080,1084,1091,1221,1225,1228,1232,1235,1348,1353,1379,1384,1409,1413,1416,1513,1518,1535,1540,1563,1584,1588,1593,1600,1605,1614,1619,1631,1636,1656,1660,1673,1676,1679,1682,1685],[678,679,680,681,688,689,693,694,697],"p",{},"In the pursuit of building ",[682,683,687],"a",{"href":684,"rel":685},"https:\u002F\u002Fpieces.app\u002Ffeatures\u002Flong-term-memory",[686],"nofollow","long-term Artificial Memory"," at the OS level, understanding ",[690,691,692],"em",{},"when"," a user wants to retrieve information is just as crucial as ",[690,695,696],{},"what"," they want.",[678,699,700],{},"In the early days, every step of that retrieval pipeline, from intent classification through span extraction, normalization, enrichment, relevance scoring, formatting, and upload, ran through cloud-hosted LLMs.",[678,702,703],{},"That meant 8–11 preprocessing tasks before touching the memory store, another 2–4 post-processing tasks afterward, and finally a round-trip to a remote model to compose the answer.",[678,705,706],{},"The result?",[678,708,709],{},"Cumulative latency that drags time-to-first-token into the seconds, accuracy hurdles at each stage, user data exposed in transit, and token bills that balloon with every query.",[678,711,712],{},"Our breakthrough with LTM-2.5: two purpose-built on-device nano-models that offload temporal understanding entirely to local hardware — one for interpreting the user's temporal intent, the other for extracting the precise time span(s) implied by their language.",[678,714,715],{},"These specialized models are the result of extensive knowledge distillation from larger foundation models, quantized and pruned to run efficiently on consumer hardware.",[678,717,718],{},"Now, the entire 10–15 step pipeline lives on-device, preserving privacy, slashing costs, and taking the deterministic retrieval of long-term context down from seconds to milliseconds in latency.",[720,721,723],"h2",{"id":722},"when-to-leverage-the-temporal-model","When to leverage the temporal model",[678,725,726,727,731,732,735],{},"Our pipeline depends on two critical steps: determining ",[728,729,730],"strong",{},"intent"," first, then generating one or more ",[728,733,734],{},"time ranges"," representative of the user's natural-language query:",[737,738,739,756],"table",{},[740,741,742],"thead",{},[743,744,745,751],"tr",{},[746,747,748],"th",{},[728,749,750],{},"Use Case",[746,752,753],{},[728,754,755],{},"Description",[757,758,759,770,780,790,800,810],"tbody",{},[743,760,761,767],{},[762,763,764],"td",{},[728,765,766],{},"Content Retrieval",[762,768,769],{},"Fetching past events (\"What was I working on just now?\")",[743,771,772,777],{},[762,773,774],{},[728,775,776],{},"Action \u002F Scheduling",[762,778,779],{},"Setting reminders or appointments (\"Remind me in two hours\")",[743,781,782,787],{},[762,783,784],{},[728,785,786],{},"Future Information \u002F Planning",[762,788,789],{},"Forecasting or \"next week\" inquiries (\"What am I doing tomorrow afternoon?\")",[743,791,792,797],{},[762,793,794],{},[728,795,796],{},"Current Status",[762,798,799],{},"Real-time checks (\"What am I doing right now?\")",[743,801,802,807],{},[762,803,804],{},[728,805,806],{},"Temporal – General",[762,808,809],{},"Ambiguous or loosely specified time references (\"Show me last week around Friday evening\")",[743,811,812,817],{},[762,813,814],{},[728,815,816],{},"Non-Temporal",[762,818,819],{},"Queries without a time component (\"Explain the concept of recursion.\")",[720,821,823],{"id":822},"temporal-range-generation","Temporal range generation",[678,825,826,827,829],{},"Once we've determined that a query requires temporal memory access, we need to precisely identify ",[728,828,692],{}," to search in the user's activity timeline.",[678,831,832],{},"This is where our second nano-model comes into play:",[834,835,837],"h3",{"id":836},"range-types-and-boundaries","Range types and boundaries",[678,839,840],{},"The temporal span predictor handles several distinct types of time references:",[737,842,843,867],{},[740,844,845],{},[743,846,847,852,857,862],{},[746,848,849],{},[728,850,851],{},"Range Type",[746,853,854],{},[728,855,856],{},"Example Query",[746,858,859],{},[728,860,861],{},"Generated Span",[746,863,864],{},[728,865,866],{},"Search Strategy",[757,868,869,885,901,917,933],{},[743,870,871,876,879,882],{},[762,872,873],{},[728,874,875],{},"Point-in-time",[762,877,878],{},"\"Show me what I was doing at 2pm yesterday\"",[762,880,881],{},"Single timestamp with narrow context window",[762,883,884],{},"Precise timestamp lookup with small buffer",[743,886,887,892,895,898],{},[762,888,889],{},[728,890,891],{},"Explicit period",[762,893,894],{},"\"What emails did I receive between Monday and Wednesday?\"",[762,896,897],{},"Clearly defined start\u002Fend boundaries",[762,899,900],{},"Bounded range search with exact limits",[743,902,903,908,911,914],{},[762,904,905],{},[728,906,907],{},"Implicit period",[762,909,910],{},"\"What was I working on last week?\"",[762,912,913],{},"Inferred start\u002Fend based on cultural\u002Fcontextual norms",[762,915,916],{},"Automatically expanded to appropriate calendar boundaries",[743,918,919,924,927,930],{},[762,920,921],{},[728,922,923],{},"Relative recent",[762,925,926],{},"\"What was I just doing?\"",[762,928,929],{},"Short window counting backward from current time",[762,931,932],{},"Recency-prioritized retrieval with adaptive timespan",[743,934,935,940,943,946],{},[762,936,937],{},[728,938,939],{},"Fuzzy historical",[762,941,942],{},"\"Show me that article I read about quantum computing last summer\"",[762,944,945],{},"Broad date range with lower confidence boundaries",[762,947,948],{},"Expanded search space with relevance decay at boundaries",[834,950,952],{"id":951},"optimizing-the-temporal-search-space","Optimizing the temporal search space",[678,954,955],{},"The model doesn't just identify time boundaries — it also generates crucial metadata about search strategy:",[957,958,959,966,981,987],"ul",{},[960,961,962,965],"li",{},[728,963,964],{},"Confidence scores"," for timespan boundaries (enabling better retrieval when dates are ambiguous)",[960,967,968,971,972,976,977,980],{},[728,969,970],{},"Periodicity hints"," for recurring events (distinguishing ",[973,974,975],"code",{},"\"my Monday meeting\""," from ",[973,978,979],{},"\"last Monday's meeting\"",")",[960,982,983,986],{},[728,984,985],{},"Time-zone awareness"," for properly interpreting references when users travel",[960,988,989,992,993,980],{},[728,990,991],{},"Contextual weighting"," that prioritizes activity density over raw timestamps (e.g., for ",[973,994,995],{},"\"when I was working on the Smith project\"",[678,997,998],{},"This specialized temporal range extraction eliminates the need to scan the entire memory corpus for each query, dramatically reducing both computational load and latency while improving retrieval precision.",[720,1000,1002],{"id":1001},"intention-differentiation-edge-cases","Intention differentiation & edge cases",[678,1004,1005],{},"Ensuring we route queries correctly between retrieval and planning:",[834,1007,1009],{"id":1008},"retrieval-vs-planning","Retrieval vs. planning",[957,1011,1012,1020],{},[960,1013,1014,1017,1018],{},[973,1015,1016],{},"\"What was I working on just now?\""," → ",[728,1019,766],{},[960,1021,1022,1017,1025],{},[973,1023,1024],{},"\"What am I doing tomorrow afternoon?\"",[728,1026,786],{},[834,1028,1030],{"id":1029},"broad-vs-specific","Broad vs. specific",[957,1032,1033,1041],{},[960,1034,1035,1017,1038,1040],{},[973,1036,1037],{},"\"Show me last week around Friday evening\"",[728,1039,766],{}," with a loose span",[960,1042,1043,1017,1046],{},[973,1044,1045],{},"\"Plan my weekend for next Friday evening\"",[728,1047,786],{},[834,1049,1051],{"id":1050},"temporal-vs-non-temporal","Temporal vs. non-temporal",[957,1053,1054,1061],{},[960,1055,1056,1017,1059],{},[973,1057,1058],{},"\"What was the website I was just looking at?\"",[728,1060,766],{},[960,1062,1063,1017,1066,1068],{},[973,1064,1065],{},"\"Explain the concept of recursion.\"",[728,1067,816],{}," (no memory lookup)",[678,1070,1071,1072,1075,1076,1079],{},"By clearly distinguishing ",[728,1073,1074],{},"temporal retrieval"," (pulling historical context) from ",[728,1077,1078],{},"temporal reference"," (scheduling or future-oriented intent), our on-device pipeline avoids misrouted cloud calls, cuts latency to the millisecond level, and maintains top-tier accuracy without sacrificing privacy or incurring hidden costs.",[720,1081,1083],{"id":1082},"examples-scenarios","Examples & scenarios",[678,1085,1086,1087,1090],{},"Below are representative user queries, each fed into our pipeline along with the user's local time in UTC (e.g. ",[973,1088,1089],{},"2025-04-17T16:43:02.151857+00:00","):",[1092,1093,1094,1128,1163,1192],"ol",{},[960,1095,1096,1099],{},[728,1097,1098],{},"Recent Activity Retrieval",[957,1100,1101,1107,1113,1122],{},[960,1102,1103,1106],{},[728,1104,1105],{},"Query:"," \"Could you tell me what I was just doing?\"",[960,1108,1109,1112],{},[728,1110,1111],{},"Classifier (23 ms):"," Content Retrieval",[960,1114,1115,1118,1119],{},[728,1116,1117],{},"Span Predictor (102 ms):"," ",[973,1120,1121],{},"2025-04-17T16:37:05.603Z – 2025-04-17T16:43:02.151857Z",[960,1123,1124,1127],{},[728,1125,1126],{},"Showcases:"," precise on-device extraction of the last few minutes of activity",[960,1129,1130,1133],{},[728,1131,1132],{},"Future Planning (Nuanced Task)",[957,1134,1135,1140,1146,1154],{},[960,1136,1137,1139],{},[728,1138,1105],{}," \"I will go to the store tomorrow.\"",[960,1141,1142,1145],{},[728,1143,1144],{},"Classifier (21 ms):"," Temporal – General",[960,1147,1148,1118,1151],{},[728,1149,1150],{},"Span Predictor:",[690,1152,1153],{},"N\u002FA",[960,1155,1156,1158,1159,1162],{},[728,1157,1126],{}," correctly ",[728,1160,1161],{},"not"," generating a past time-range for future intentions—an essential nuance",[960,1164,1165,1168],{},[728,1166,1167],{},"\"Just\" Retrieval Consistency",[957,1169,1170,1175,1180,1187],{},[960,1171,1172,1174],{},[728,1173,1105],{}," \"What was the website I was just looking at?\"",[960,1176,1177,1112],{},[728,1178,1179],{},"Classifier (22 ms):",[960,1181,1182,1118,1185],{},[728,1183,1184],{},"Span Predictor (108 ms):",[973,1186,1121],{},[960,1188,1189,1191],{},[728,1190,1126],{}," consistent span output across semantically similar \"just\" queries",[960,1193,1194,1197],{},[728,1195,1196],{},"Long-Range Historical Query",[957,1198,1199,1204,1208,1216],{},[960,1200,1201,1203],{},[728,1202,1105],{}," \"What was I working on last year around Thanksgiving?\"",[960,1205,1206,1112],{},[728,1207,1179],{},[960,1209,1210,1118,1213],{},[728,1211,1212],{},"Span Predictor (88 ms):",[973,1214,1215],{},"2024-11-01T00:00:00Z – 2024-11-30T23:59:59.999999Z",[960,1217,1218,1220],{},[728,1219,1126],{}," broad date-range generation for loosely specified historical periods",[720,1222,1224],{"id":1223},"benchmarks","Benchmarks",[678,1226,1227],{},"Tested on an Apple M1 Max (32 GB) under heavy load (30+ tabs, video, IDEs, messaging) to simulate real-world conditions:",[834,1229,1231],{"id":1230},"classification-results","Classification results",[678,1233,1234],{},"This table compares how well each model identifies the correct temporal intent label for a given sample.",[737,1236,1237,1271],{},[740,1238,1239],{},[743,1240,1241,1246,1251,1256,1261,1266],{},[746,1242,1243],{},[728,1244,1245],{},"Model Name",[746,1247,1248],{},[728,1249,1250],{},"Accuracy",[746,1252,1253],{},[728,1254,1255],{},"F1 (W)",[746,1257,1258],{},[728,1259,1260],{},"Prec (W)",[746,1262,1263],{},[728,1264,1265],{},"Recall (W)",[746,1267,1268],{},[728,1269,1270],{},"Samples\u002FSec",[757,1272,1273,1291,1310,1329],{},[743,1274,1275,1278,1281,1283,1286,1288],{},[762,1276,1277],{},"nano-temporal-intent (TIME Intent)",[762,1279,1280],{},"0.9930",[762,1282,1280],{},[762,1284,1285],{},"0.9931",[762,1287,1280],{},[762,1289,1290],{},"544.41",[743,1292,1293,1296,1299,1302,1305,1307],{},[762,1294,1295],{},"gemini-1.5-flash-002",[762,1297,1298],{},"0.8241",[762,1300,1301],{},"0.8384",[762,1303,1304],{},"0.8834",[762,1306,1298],{},[762,1308,1309],{},"9.14",[743,1311,1312,1315,1318,1321,1324,1326],{},[762,1313,1314],{},"gpt-4o",[762,1316,1317],{},"0.8634",[762,1319,1320],{},"0.8470",[762,1322,1323],{},"0.8698",[762,1325,1317],{},[762,1327,1328],{},"9.40",[743,1330,1331,1334,1337,1340,1343,1345],{},[762,1332,1333],{},"meta-llama\u002FLlama-3.2-3B-Instruct",[762,1335,1336],{},"0.4604",[762,1338,1339],{},"0.4094",[762,1341,1342],{},"0.4080",[762,1344,1336],{},[762,1346,1347],{},"92.43",[678,1349,1350],{},[728,1351,1352],{},"Legend: Classification Models",[957,1354,1355,1361,1367,1373],{},[960,1356,1357,1360],{},[728,1358,1359],{},"nano-temporal-intent (TIME Intent):"," Our on-device nano-model for intent classification—ultra-lightweight and lightning-fast inference.",[960,1362,1363,1366],{},[728,1364,1365],{},"gemini-1.5-flash-002:"," Google's mid-tier large language model via API; good accuracy but higher latency and cost.",[960,1368,1369,1372],{},[728,1370,1371],{},"gpt-4o:"," OpenAI's flagship multimodal LLM; strong performance at premium compute and pricing.",[960,1374,1375,1378],{},[728,1376,1377],{},"meta-llama\u002FLlama-3.2-3B-Instruct:"," A 3 billion-parameter open-weights LLM; lower accuracy but faster than cloud LLMs.",[678,1380,1381],{},[728,1382,1383],{},"Legend: Classification Metrics",[957,1385,1386,1392,1403],{},[960,1387,1388,1391],{},[728,1389,1390],{},"Accuracy:"," Proportion of samples for which the top-prediction matches the true class.",[960,1393,1394,1396,1397,1396,1399,1402],{},[728,1395,1255],{},", ",[728,1398,1260],{},[728,1400,1401],{},"Recall (W):"," Weighted F1-score, precision, and recall across all intent classes (accounts for class imbalances).",[960,1404,1405,1408],{},[728,1406,1407],{},"Samples\u002FSec:"," Number of inference calls the model can process per second. (higher is better)",[720,1410,1412],{"id":1411},"span-prediction-results","Span prediction results",[678,1414,1415],{},"This table measures how precisely each model extracts the correct time-span from text.",[737,1417,1418,1445],{},[740,1419,1420],{},[743,1421,1422,1426,1431,1436,1441],{},[746,1423,1424],{},[728,1425,1245],{},[746,1427,1428],{},[728,1429,1430],{},"E.C.O. Rate",[746,1432,1433],{},[728,1434,1435],{},"Avg IoU",[746,1437,1438],{},[728,1439,1440],{},"Exact Match",[746,1442,1443],{},[728,1444,1270],{},[757,1446,1447,1464,1481,1497],{},[743,1448,1449,1452,1455,1458,1461],{},[762,1450,1451],{},"nano-temporal-span-pred (TIME Range)",[762,1453,1454],{},"0.9450",[762,1456,1457],{},"0.9201",[762,1459,1460],{},"0.8659",[762,1462,1463],{},"785.39",[743,1465,1466,1469,1472,1475,1478],{},[762,1467,1468],{},"gemini-1.5-pro-002",[762,1470,1471],{},"0.2065",[762,1473,1474],{},"0.1865",[762,1476,1477],{},"0.1684",[762,1479,1480],{},"9.35",[743,1482,1483,1485,1488,1491,1494],{},[762,1484,1314],{},[762,1486,1487],{},"0.1767",[762,1489,1490],{},"0.1611",[762,1492,1493],{},"0.1535",[762,1495,1496],{},"9.47",[743,1498,1499,1501,1504,1507,1510],{},[762,1500,1333],{},[762,1502,1503],{},"0.1725",[762,1505,1506],{},"0.1640",[762,1508,1509],{},"0.1517",[762,1511,1512],{},"62.02",[678,1514,1515],{},[728,1516,1517],{},"Legend: Span Models",[957,1519,1520,1526],{},[960,1521,1522,1525],{},[728,1523,1524],{},"nano-temporal-span-pred (TIME Range):"," On-device span extractor optimized for low latency and high IoU.",[960,1527,1528,1396,1530,1396,1532,1534],{},[728,1529,1468],{},[728,1531,1314],{},[728,1533,1377],{}," LLMs & SLMs performing span extraction via API calls.",[678,1536,1537],{},[728,1538,1539],{},"Legend: Span Metrics",[957,1541,1542,1547,1552,1558],{},[960,1543,1544,1546],{},[728,1545,1430],{}," (Exact Coverage Overlap): Fraction of predicted spans that exactly match the gold span boundaries.",[960,1548,1549,1551],{},[728,1550,1435],{}," (Intersection-over-Union): Average overlap ratio between predicted and true spans.",[960,1553,1554,1557],{},[728,1555,1556],{},"Exact Match:"," Strict percentage of samples where predicted span text equals ground truth.",[960,1559,1560,1562],{},[728,1561,1407],{}," Span-prediction throughput on the benchmark hardware. (higher is better)",[1564,1565,1566],"blockquote",{},[678,1567,1568,1571,1572,1575,1576,1579,1580,1583],{},[728,1569,1570],{},"We observed SLMs running in the cloud"," on H100 GPU with vLLM incur $0.018 – $1.90 per run and took 15-25 min of compute time — our cascade delivers structured time-spans offline in ",[728,1573,1574],{},"milliseconds",", with ",[728,1577,1578],{},"zero API cost"," and ",[728,1581,1582],{},"full data privacy",".",[720,1585,1587],{"id":1586},"why-it-matters","Why it matters",[678,1589,1590],{},[728,1591,1592],{},"🏗️ Architectural Specialization",[678,1594,1595,1596,1599],{},"Breaking monolithic LLMs into nano-models for classification vs. span prediction yields ",[728,1597,1598],{},"massive gains"," in both accuracy and speed.",[678,1601,1602],{},[728,1603,1604],{},"🌐 Edge-First AI",[678,1606,1607,1608,1613],{},"Offline inference ",[682,1609,1612],{"href":1610,"rel":1611},"https:\u002F\u002Fpieces.app\u002Fblog\u002Foffline-ai",[686],"keeps sensitive data on-device"," — critical for medical, defense, and privacy-focused applications.",[678,1615,1616],{},[728,1617,1618],{},"💡 Energy & Cost Efficiency",[678,1620,1621,1626,1627,1630],{},[682,1622,1625],{"href":1623,"rel":1624},"https:\u002F\u002Fpieces.app\u002Fblog\u002Fsmall-language-models-outshine-large-language-models-enterprise-users",[686],"Eliminate token fees"," and slash compute budgets. This is the future of ",[728,1628,1629],{},"sustainable",", scaled AI on laptops, wearables, and IoT.",[678,1632,1633],{},[728,1634,1635],{},"🔬 Research Frontiers",[957,1637,1638,1644,1650],{},[960,1639,1640,1643],{},[728,1641,1642],{},"Task-specific:"," distillation, quantization, and final pruning for modular pipelines",[960,1645,1646,1649],{},[728,1647,1648],{},"Adaptive orchestration",": dynamic model selection based on compute availability",[960,1651,1652,1655],{},[728,1653,1654],{},"Hardware\u002Fsoftware:"," co-design for ultra-efficient inference",[720,1657,1659],{"id":1658},"conclusion","Conclusion",[678,1661,1662,1663,1396,1666,1669,1670,1583],{},"This nano-temporal pipeline is one of approximately 11 nano-models we're weaving into LTM-2.5 to make long-term memory formation and retrieval across your entire OS ",[728,1664,1665],{},"blazingly fast",[728,1667,1668],{},"highly accurate",", and ",[728,1671,1672],{},"privacy-first",[678,1674,1675],{},"Innovation isn't about bigger models — it's about smarter, specialized models that deliver tangible benefits in real-world applications.",[678,1677,1678],{},"By focusing on modular, purpose-built AI systems that run entirely on-device, we're redefining what's possible for intelligent, responsive computing that respects user privacy while dramatically reducing cost and latency.",[678,1680,1681],{},"We can't wait to share more as we push the boundaries of on-device AI in the world of OS-level Long-Term Memory.",[678,1683,1684],{},"Lastly, I would be remiss if I didn't mention the obvious: none of this would be possible without the incredible creativity, dedication, and perseverance from the team behind Pieces.",[678,1686,1687],{},"I’ll close with a special shout out to our ML team and a extra special shout out to Antreas Antoniou and Sam Jones for believing in the approach and turning these first-principal theories into breakthroughs ✨",{"title":1689,"searchDepth":1690,"depth":1690,"links":1691},"",2,[1692,1693,1698,1703,1704,1707,1708,1709],{"id":722,"depth":1690,"text":723},{"id":822,"depth":1690,"text":823,"children":1694},[1695,1697],{"id":836,"depth":1696,"text":837},3,{"id":951,"depth":1696,"text":952},{"id":1001,"depth":1690,"text":1002,"children":1699},[1700,1701,1702],{"id":1008,"depth":1696,"text":1009},{"id":1029,"depth":1696,"text":1030},{"id":1050,"depth":1696,"text":1051},{"id":1082,"depth":1690,"text":1083},{"id":1223,"depth":1690,"text":1224,"children":1705},[1706],{"id":1230,"depth":1696,"text":1231},{"id":1411,"depth":1690,"text":1412},{"id":1586,"depth":1690,"text":1587},{"id":1658,"depth":1690,"text":1659},"2025-04-17T00:00:00.000Z","In the pursuit of building long-term Artificial Memory at the OS level, understanding when a user wants to retrieve information is just as crucial as what they want.",false,"md","material-symbols:memory-rounded","#D500F9",null,"https:\u002F\u002Fstorage.googleapis.com\u002Fpieces-marketing-website\u002Fimages\u002Fblog\u002Fnano-models\u002Fhero.png",{},"\u002Fupdates\u002Fnano-models",{"title":673,"description":1711},"updates\u002Fnano-models","Discover the latest breakthrough in AI with Nano-Models as the Pieces Team unveils LTM‑2.5. A game-changer for AI-powered development!","_teYoSnPg1HhTqrrunf02Au1y8nXEmksX3CxD3xBR0U",[],{"left":1726,"top":1726,"width":1727,"height":1727,"rotate":1726,"vFlip":1712,"hFlip":1712,"body":1728},0,24,"\u003Cpath fill=\"currentColor\" d=\"m7.825 13l4.9 4.9q.3.3.288.7t-.313.7q-.3.275-.7.288t-.7-.288l-6.6-6.6q-.15-.15-.213-.325T4.426 12t.063-.375t.212-.325l6.6-6.6q.275-.275.688-.275t.712.275q.3.3.3.713t-.3.712L7.825 11H19q.425 0 .713.288T20 12t-.288.713T19 13z\"\u002F>"]