Kwenzekeni
I-Google imemezele ukuthi i-Antigravity SDK manje isisekela ukwenziwa kwemodeli yendawo ye-AI, okuvumela onjiniyela bakhe izinhlelo zokusebenza ezisebenza ngokungaxhunyiwe ku-inthanethi ngokuphelele. Isibuyekezo sifaka ukusekelwa kokuqala kwemodeli ye-Gemma 4 26B A4B kusetshenziswa isikhathi sokusebenza se-Google AI Edge's LiteRT. Lokhu kunika amandla ukusetshenziswa kwezinsiza ze-GPU yendawo kanye ne-RAM kwimisebenzi ye-ejenti, efana nokuhlolwa kwekhodi nokukhiqiza izinsiza, ngaphandle kokuthembela ekuxhumekeni kwamafu. I-SDK ihlanganisa nokusekelwa kwe-plug-and-play kwamaseva endawo ahambisana ne-OpenAI afana ne-Ollama, i-LM Studio, ne-vLLM.
I-Google yethule ukusekelwa kwamamodeli asendaweni e-AI ku-Antigravity SDK, ithuluzi elivumela abathuthukisi ukuthi bakhe izinhlelo zokusebenza ezinekhono elifanayo le-ejenti elinika amandla i-Google Antigravity. Lesi sibuyekezo sivumela ukwenziwa kokugeleza komsebenzi kwe-ejenti ungaxhunyiwe ku-inthanethi ngokuphelele, kusetshenziswa izinsiza zezingxenyekazi zekhompuyutha zasendaweni ezifana ne-GPU ne-RAM.
Ukuqaliswa kokuqala kufaka ukusekelwa kwemodeli ye-Gemma 4 26B A4B, elungiselelwe isikhathi sokusebenza se-Google AI Edge's LiteRT. I-Google ithi lokhu kulungiselelwa kusebenzisa kahle amakhono omshini wendawo ukuze kuthuthukiswe lokho okungalethwa kudivayisi. Isimemezelo sihlanganisa isampula yeskripthi se-Python, agy_sample.py, sokusiza onjiniyela baqale ngalokhu kugeleza komsebenzi kwasendaweni.
I-SDK futhi ihlinzeka ngosekelo olungenamthungo kunoma iyiphi iseva ehambisana ne-OpenAI, okuhlanganisa i-Ollama, i-LM Studio, ne-vLLM, nge-LocalOpenAIAgentConfig. Lokhu kuvumelana nezimo kuvumela onjiniyela ukuthi bahlole i-backends yendawo ehlukile ngenkathi begcina i-orchestration ye-ejenti engashintshi, amathuluzi, nokugeleza komsebenzi.
I-Google ibonise ukusetshenziswa kwalokhu kusetha ngedemo eyingxube lapho imodeli yomakhi wezakhiwo esekelwe efwini (Gemini 3.8 Flash) yasebenza njengomhleli, kuyilapho uquqaba lwasendaweni lwezenzakalo ze-Gemma 4 26B lusingathe imisebenzi yokuhlolwa kwekhodi nemisebenzi yokuchibiyela. Esinye isibonelo sibonise i-ejenti yendawo ngokuzenzakalelayo yakha ukuqapha kwensiza yokugcina ebuyekezwa bukhoma isebenzisa imitapo yolwazi ye-Python efana ne-psutil nokucebile, konke ngaphandle kokufinyelela ku-inthanethi.
Imininingwane yomthombo: developers.googleblog.com ↗
Kungani kubalulekile
Lesi sibuyekezo sehlisa umgoqo wokuba onjiniyela basebenzise ama-agent e-AI ezindaweni lapho ubumfihlo bedatha, ukubambezeleka, noma ukuxhumeka kwe-inthanethi kuyizingqinamba ezibalulekile. Ngokusekela ukusetshenziswa kwasendaweni, i-SDK ivumela izakhiwo ezixubile lapho amamodeli wamafu aphatha ukuhlela okusezingeni eliphezulu kuyilapho amamodeli endawo enza umsebenzi wokuphakamisa osindayo wekhompyutha kudivayisi. Le ndlela inganciphisa izindleko zamathokheni futhi iqinisekise ukuthi idatha ebucayi ihlala emshinini womsebenzisi. Ukwenziwa ngcono okuqondile kwe-LiteRT kanye ne-Gemma 4 26B kunikeza indlela ephathekayo, ehloliwe yokuthuthukiswa kwe-AI enqenqemeni, idlulela ngale kwemibono ye-AI yendawo ecatshangelwayo iye kumathuluzi asebenzayo, ahlanganisiwe ngaphakathi kwe-ecosystem yonjiniyela yenkampani enkulu yezobuchwepheshe.
Amandla okusebenzisa i-AI ye-agent endaweni abhekana nokukhathazeka okubalulekile okuphathelene nobumfihlo bedatha nokuphepha, njengoba ikhodi ebucayi nedatha akudingeki ukuthi idluliselwe kumaseva amafu angaphandle. Lokhu kusebenza ikakhulukazi ezindaweni zebhizinisi noma konjiniyela abangabodwana abasebenza ngolwazi lobunikazi noma oluyimfihlo.
Ngokusekela iphethini ye-hybrid Architect-Builder, i-SDK inikeza isixazululo esisebenzayo sokulinganisa isikali samamodeli wamafu nobumfihlo kanye nokusebenza kahle kwezindleko kwamamodeli endawo. Lesi sakhiwo sivumela imisebenzi yokuhlela eyinkimbinkimbi ukuthi isingathwe amamodeli amafu anamandla kuyilapho imisebenzi evamile noma ephakamisa izinto ezisindayo yenziwa endaweni, kuthuthukisa ukusetshenziswa kwamathokheni kanye nokunciphisa ukubambezeleka.
Isincomo esithile sezingxenyekazi zekhompuyutha ezingaphezu kuka-24GB we-VRAM noma inkumbulo ehlanganisiwe kusetha isisekelo esicacile sobuchwepheshe sabasebenzisi. Nakuba lokhu kunciphisa ukufinyeleleka ngokushesha emishinini ephezulu, kunikeza okuqondiwe okukhonkolo konjiniyela nabakhiqizi bezingxenyekazi zekhompuyutha abahlose ukusekela imithwalo yemisebenzi ye-AI ethuthukisiwe.
Ukuhlanganiswa kwe-LiteRT nokusekelwa kwamaseva ajwayelekile asendaweni afana ne-Ollama ne-vLLM kunciphisa ukushayisana konjiniyela asebewajwayele la mathuluzi. Lokhu kumiswa kusiza ukuhlanganisa ukwakheka kwezwe okuhlukene kokuthuthukiswa kwe-AI yasendaweni, kunikeze isixhumi esibonakalayo esihlanganisiwe se-orchestration ye-ejenti ngokunganaki ingemuva lemodeli yendawo.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
crm_get_transaction(id='4092').What most distinguishes an AI agent from a basic chatbot?
Ongakubuka ngokulandelayo
Onjiniyela kufanele baqaphe ikhosombe le-Antigravity Python SDK GitHub ukuze bathole izibuyekezo zokucushwa kwemodeli yendawo kanye namabhentshimakhi okusebenza. Ukusebenza okungokoqobo kwezidingo zehadiwe ezinconyiwe, ikakhulukazi imishini ene-VRAM engaphezu kuka-24GB noma inkumbulo ehlanganisiwe, kuzonquma ukufinyeleleka kwalezi zici ngesisekelo esibanzi sonjiniyela. Ukwengeza, ukwamukelwa kwephethini ye-Architect-Builder, ehlanganisa ukuhlelwa kwamafu nokusebenza kwasendaweni, kungase kube nomthelela ekutheni amasistimu e-AI ayingxubevange aklanywa kanjani kubhizinisi nezinhlelo zokusebenza zabathengi.
Ukusebenza nokuzinza kwemodeli ye-Gemma 4 26B A4B lapho isebenza kuzingxenyekazi zekhompuyutha zezinga lomthengi ezine-24GB+ VRAM enconyiwe kuzoba inkomba eyinhloko yento ewusizo yesici. Abamukeli bangaphambi kwesikhathi bangase babike ngokubambezeleka, ukuphathwa kwenkumbulo, kanye nekhwalithi yemiphumela ye-ejenti kuzimo zomhlaba wangempela.
Ukuvela kokusekela kwe-Antigravity SDK kwamanye amamodeli endawo kanye ne-backends kuzonquma ukuhambisana kwayo kwesikhathi eside. Uma i-Google inweba usekelo kumamodeli amancane noma asebenza kahle kakhulu, izimfuneko zezingxenyekazi zekhompuyutha zingase zehle, kunwetshwe isisekelo somsebenzisi esingaba khona.
Ukwamukelwa kokuhamba komsebenzi kwe-hybrid cloud-yasendaweni kuzilungiselelo zebhizinisi kuzoba umkhuba obalulekile wokuqapha. Uma le phethini ibonakala iphumelela ekugcineni ubumfihlo bedatha ngenkathi isebenzisa ubuhlakani bamafu, ingaba yisakhiwo esijwayelekile samathuluzi okuthuthukisa aqhutshwa yi-AI.
Impendulo yomphakathi nge-Antigravity Python SDK GitHub Issue Tracker izohlinzeka ngemininingwane ezinseleleni ezivamile kanye nezicelo zesici. Le ndlela yempendulo ibalulekile ekuqondeni imikhawulo yomhlaba wangempela wokusetshenziswa kwamanje kanye nokuqondisa izibuyekezo ezizayo.