zg Uyasebenza

zg (zvec-grep) iyithuluzi elihlanganisiwe lokusesha elivulekile le-Zvec AI, eligxile ekusebenzeni endaweni futhi elakhelwe onjiniyela nama-AI Agent. Leli thuluzi lihlanganisa ukufanisa okunembile kwe-ripgrep, ukusesha kwamagama kwe-BM25 kanye nokusesha kwe-vector semantic kube yinto eyodwa, lixazulule igebe lamagama phakathi kwenhloso yolimi lwemvelo nendlela okuqukethwe kubhalwe ngayo ngempela kukhodi noma emibhalweni, futhi lenze kube nokugeleza okuphelele kokusesha kusukela ekuhloleni okungacacile kuya ekutholeni indawo ngokuqondile. I-zg isekela ukuskena amafayela, i-Embedding, ukukhonjwa nokusesha, konke kwenziwa ngokuzenzakalelayo kudivayisi yendawo. Iza namamodeli alula asebenza kudivayisi, ayidingi i-GPU noma isevisi yangaphandle, futhi ihlanganisa ubumfihlo nokusebenza kahle.

Landela i-WeChat bese uphendula ngokuthi “开源” ukuze ujoyine iqembu lokuxhumana lamaphrojekthi e-AI avulekile
Ukusesha kwe-semantic: thola okuqukethwe okuhlobene kukhodi nasemibhalweni ngokwenhloso yolimi lwemvelo, ngaphandle kokuqagela amagama angukhiye kusenesikhathi.
Ukusesha kwe-BM25: hlela izingxenye zombhalo ngokuhambisana kwazo usebenzisa isisindo esisekelwe ebuningini bamagama, ukuze ugxile ngokushesha emiphumeleni efanele kakhulu.
Ukusesha okuxubile: hlanganisa ukubuyiswa kwe-semantic vector nezindlela eziningi ze-BM25, usebenzise i-RRF ukuhlanganisa, ususe okuphindaphindiwe futhi uhlele kabusha.
Ukufanisa okunembile kwe-ripgrep: qinisekisa ngokuphelele izimpawu ezaziwayo, amagama emisebenzi, okhiye bokumisa noma amaphethini ombhalo.
Ukukhipha okuqukethwe kwamafomethi amaningi: hlaziya ngokuzenzakalelayo amazinga ezimpawu zekhodi, ukwakheka kwezahluko zemibhalo namafayela edatha, ugcine imethadatha yendawo.
I-Embedding yasendaweni kudivayisi: inamamodeli alula angu-11 akhelwe ngaphakathi, asebenza ngokuzenzakalelayo kudivayisi, ngaphandle kwe-GPU noma ukulayisha idatha.
Ukugcinwa kwenkomba okushumekiwe: isebenzisa isizindalwazi se-Zvec esisebenza ngaphakathi kwenqubo ukuze igcine izinkomba ze-vector ne-BM25, ngaphandle kokuthumela isevisi ezimele.
Ukuvuselelwa kwenkomba okwandayo: ngemva kokwakhiwa kokuqala, icubungula kuphela okuqukethwe okushintshile; inkomba ingaphinde isetshenziswe isikhathi eside emibuzweni nakwimisebenzi yama-Agent.
Izindlela ezimbili zokungena, i-CLI + MCP: inikeza umugqa womyalo kubantu, futhi isebenzisa iphrothokholi ye-MCP ukuze isetshenziswe ama-Agent afana ne-Codex ne-Cursor.
Ukuhlanganiswa okuzenzakalelayo kwama-Agent: i-zg install ithola ngokuzenzakalelayo ama-Agent afakiwe kule khompyutha bese iqedela ukumiswa kwe-MCP.

Ukufaka: faka ithuluzi le-zg emhlabeni wonke usebenzisa i-npm install -g @zvec/zvec-grep.
Ukumisa: sebenzisa i-zg install ukuze ithole ngokuzenzakalelayo ama-Agent afakiwe kule khompyutha, afana ne-Codex ne-Cursor, bese iqedela ukuhlanganiswa kwe-MCP.
Ukwakha inkomba: ngena kumkhombandlela we-repository oqondisiwe, bese usebenzisa i-zg index ukwakha izinkomba zasendaweni ze-semantic ne-BM25 zekhodi nemibhalo.
Ukusesha nge-CLI: kutheminali sebenzisa i-zg query --human "Umbuzo wakho" ukuze useshe okuqukethwe kwasendaweni ngolimi lwemvelo.
Ukusesha nge-Agent: buza ngokuqondile ngolimi lwemvelo ku-Agent exhunyiwe; i-Agent izobiza i-zg nge-MCP ngokuzenzakalelayo ukuze iqedele ukusesha.

Indlela yokusesha ehlanganisiwe: ihlanganisa ukutholwa kwe-semantic, ukuhlelwa kwe-BM25 nokufanisa okunembile kwe-ripgrep kube ukugeleza okukodwa okuphelele kokusesha, kusukela enhlosweni engacacile kuya ekuqinisekiseni izimpawu.
Idizayinelwe ama-Agent kusukela ekuqaleni: ihlanganiswa ngokujulile nama-Agent adumile ngephrothokholi ye-MCP, futhi ukuhlolwa kwangempela kukhombisa ukuncipha okungaphezu kuka-40% ekubizweni kwamathuluzi nasekusetshenzisweni kwama-Token.
Ubumfihlo obubeka indawo kuqala: ukuskena amafayela, i-Embedding, ukukhonjwa nokusesha kwenziwa ngokuzenzakalelayo kudivayisi, ngaphandle kokulayisha noma yikuphi okuqukethwe.
Amamodeli alula asebenza kudivayisi: anamamodeli angu-11 e-Embedding asebenza kudivayisi; imodeli ezenzakalelayo enamapharamitha angu-16M iyakwazi ukusebenza ngaphandle kwe-GPU noma i-API ekude.
Ukugcinwa okushumekiwe okungenaz依依: isebenzisa isizindalwazi se-Zvec esisebenza ngaphakathi kwenqubo ukuze yenze izinkomba ze-vector ne-BM25, ngaphandle kokuthumela noma ukugcina isevisi yesizindalwazi ezimele.
Ukusetshenziswa kabusha kwenkomba okwandayo: ngemva kokwakhiwa kokuqala, icubungula kuphela okuqukethwe okushintshile; inkomba efanayo ingabiwa futhi isetshenziswe isikhathi eside yi-CLI nemisebenzi eminingi yama-Agent.
Ukukhipha okuqukethwe okuhlelekile: ihlaziya ngokuzenzakalelayo amazinga ezimpawu zekhodi, ukwakheka kwezahluko zemibhalo nemethadatha, ukuze imiphumela yosesho ibe nendawo eqondakalayo futhi iqinisekiseke.
Ukuhlanganiswa kwama-Agent ngokuchofoza okukodwa: i-zg install ithola ngokuzenzakalelayo i-Codex, i-Cursor, i-Claude Code namanye ama-Agent akule khompyutha bese iqedela ukumiswa kwe-MCP.

Indawo yokugcina ye-GitHub: https://github.com/zvec-ai/zvec-grep

Ukuhlola inhloso yekhodi: unjiniyela noma i-Agent ichaza into yebhizinisi ngolimi lwemvelo, bese i-zg semantic search ithola ukuqaliswa okuhlobene kwe-hydratePreferences kukhodi noma “ukugunyazwa nokwamukelwa kwe-akhawunti” emibhalweni.
Ukuthola amaphutha nokuhlaziya imbangela eyisisekelo: uma kubhekwana namalogi amaphutha noma incazelo engacacile evela kumsebenzisi, i-Agent isebenzisa i-zg semantic search ukuthola ngokushesha izindlela zekhodi, amafayela okumisa nezincazelo zemibhalo ezihlobene, kuncishiswe ukusesha okuningi nokufundwa kwamafayela angahlobene.
Ukuqonda ikhodi ewela amafayela: lapho kuqondwa ukuqaliswa okuphelele komsebenzi othile kwikhodi enkulu, ukubuza ngolimi lwemvelo kuthola imisebenzi, izincazelo zezixhumi nezindlela zokubizana ezihlobene, eziwela amamojula nezilimi, ngaphandle kokuzulazula mathupha esihlahleni semikhombandlela.
Imibuzo nezimpendulo kusizindalwazi solwazi sendawo: ngemva kokukhomba imibhalo yezobuchwepheshe, izinto zocwaningo, amanothi namanye amafayela e-Markdown/ombhalo, sebenzisa umbuzo wolimi lwemvelo ukuze uthole izimpendulo ezinembile ngokushesha, kanye nefayela lomthombo nendawo yesahluko esithile.