- I-LangGraph ivumela ukudalwa kwama-ejenti e-AI kusetshenziswa umshini wesimo osuselwe kugrafu, okuvumela izihibe zokucabanga eziphindaphindayo.
- Iphethini ye-ReAct ihlukanisa ama-ejenti kuma-chatbot ngokuvumela umjikelezo wokucabanga, ukwenza ngamathuluzi, nokubuka imiphumela.
- Ukuthuthukiswa kuqala ngokusebenzisa abasizi abakhelwe ngaphambilini ukuze kwenziwe i-prototyping esheshayo kuya ekwakhiweni kwe-StateGraph ngesandla ukuze kulawulwe i-granular.
- Ama-ejenti alungele ukukhiqiza adinga ukugxila okunesu ekucaciseni kwemibhalo, ukuhlolwa kwesimo, kanye nokuqapha komuntu ngaphakathi kwe-loop.
Wake wazibuza ukuthi kwenzekani ngempela ngaphansi kwe-hood lapho i-AI ingagcini nje ngokuxoxa kodwa empeleni yenza izinto zenziwe ? Ngenkathi ama-chatbot ajwayelekile elungele imibuzo nezimpendulo ezisheshayo, umlingo wangempela uyenzeka uma ungena endaweni yama-ejenti e-AI. Lezi zinhlelo azigcini nje ngokuthumela umbhalo; ziyacabanga ngezinkinga , zikhethe amathuluzi afanele omsebenzi, futhi ziphindaphinde kuze kube yilapho umsebenzi usufeziwe. Kungukusuka kumuntu ophendulayo olula kuya kumsebenzi wedijithali, ikhono eliyigolide njengamanje emhlabeni we-AI Engineering.
Ukuze lokhu kube ngokoqobo, ithimba elisekela iLangChain lisinike iLangGraph , uhlaka olukuvumela ukuthi uklame la ma-ejenti njengamagrafu. Esikhundleni somugqa oqondile ukusuka ku-A kuya ku-B, ungakha ama-loop namagatsha ayinkimbinkimbi , okuvumela i-AI ukuthi icabange, yenze, ibone umphumela, bese icabanga futhi. Uma ufuna ukukhuphuka kusukela eziqondisweni eziyisisekelo kuya emisebenzini yokusebenza eyinkimbinkimbi nenesimo ku-Python, ufike endaweni efanele.
Yini Ehlukanisa I-AI Agent ne-Chatbot?
Cabanga nge-chatbot evamile njengomshini wokuthengisa: ufaka isicelo, futhi ikunikeza impendulo echazwe kusengaphambili noma ekhiqizwe. Kodwa-ke, i-ejenti ifana kakhulu nomsizi womuntu siqu onethuluzi lamathuluzi . Isebenzisa i-Large Language Model (LLM) njengobuchopho bayo obuphakathi ukuze inqume ukuthi yisiphi isinyathelo okufanele isithathe ngokulandelayo. Le nqubo ivame ukubizwa ngokuthi iphethini ye-ReAct (Ukubonisana Nokusebenza) , lapho imodeli ijikeleza khona ngomjikelezo wokucabanga ngomsebenzi, ibiza ithuluzi—njengokusesha iwebhu noma umbuzo wedathabheyisi—futhi ihlaziya umphumela ukuze inqume ukuthi idinga ukuzama enye into noma ukuthi inempendulo yokugcina.
Ukuthola Ukuqonda ku-LangGraph
I-LangGraph akuyona nje enye ilabhulali; iyindlela yokwenza imodeli ye-ejenti logic njengomshini wesimo . Kuchungechunge lwendabuko, uqhubeka ngokulandelana. Ku-LangGraph, uchaza isimo esabiwe (ngokuvamile umlando wemiyalezo) esihamba ngama-node ahlukahlukene. Lawa ma-node ngokuyisisekelo “yizindawo zokusebenza” lapho i-logic yenzeka khona, futhi imiphetho “yizindlela” ezinquma ukuthi ukugeleza kuya kuphi ngokulandelayo. Leli khono lokusingatha imijikelezo yingakho nje kuyisinqumo sama-ejenti adinga ukuzilungisa noma ukuziphindaphinda.
Kuye ngokuthi ufuna ukulawula okungakanani, kunezindlela ezimbili zokwakha. Ithrekhi esheshayo isebenzisa imisebenzi eyakhelwe ngaphambilini ephatha ukuphakamisa okunzima kuwe. Indlela esetshenziswa ngesandla ihilela ukwakha i- StateGraph node nge-node , efanelekile uma udinga ukunemba kokuhlinzwa mayelana nendlela i-ejenti eziphatha ngayo.
Isinyathelo Ngesinyathelo: Ukwakha I-ejenti Yakho Yokuqala
Ngaphambi kokuthi ungene kukhodi, kuzodingeka ukuthi setha indawo yakho. Bamba i- langgraph futhi langchain-openai amaphakheji nge-pip. Uzodinga futhi ukhiye we-API ovela kumhlinzeki onjenge-OpenAI noma i-Anthropic, ogcinwe ngokuphephile njengokuguquguquka kwemvelo. Uma unesabelomali esincane noma ukhetha ubumfihlo, empeleni ungakusebenzisa lokhu endaweni usebenzisa i-Ollama, uma nje ihadiwe yakho ingakwazi ukusingatha umthwalo.
Ukuchaza Ithuluzi Lomenzeli
I-ejenti ilungile kuphela njengamathuluzi ayo. Ku-Python, la mathuluzi ayimisebenzi nje ehlanganiswe ne- @umhlobisi wamathuluzi kusukela langchain_core.toolsNansi isosi eyimfihlo: i-docstringI-LLM ifunda incazelo yomsebenzi ukuze ithole ukuthi kufanele isetshenziswe nini. Uma incazelo yakho ingacacile, i-ejenti izodideka; uma icacile ngekristalu, i-ejenti izoba neqiniso eliqondile ekukhetheni kwayo amathuluzi.
Ukuhlanganisa i-Logic
Kulabo abafuna imiphumela esheshayo, create_react_agent umsebenzi usindisa impilo. Umane uwunike imodeli, uhlu lwamathuluzi, kanye nesaziso sesistimu, bese yakha ngokuzenzakalelayo i-ReAct loopNgaphansi kwe-hood, lokhu kudala igrafu ene-node yokucabanga, i- ToolNode ukuze kwenziwe, kanye nomkhawulo onemibandela onquma ukuthi uzobuyela ebuchosheni noma uzosonga izinto. Ube usushayela ucingo .compile() ukuguqula lo mdwebo ube yi-ejenti ephilayo.
Ukwenziwa Nokuqashelwa
Ukuze ugingqike ibhola, sebenzisa .invoke() indlela enesichazamazwi esiqukethe imiyalezo yakho. Ukuze ubone ukuthi kwenzekani ngempela ekhanda le-ejenti, kunconywa kakhulu ukuthi uxhumeke LangSmithLokhu kunikeza ithuba lokuthola umkhondo wokucabanga, okwenza kube lula kakhulu lungisa iphutha lapho i-ejenti iphume khona endleleni ngesikhathi somsebenzi oyinkimbinkimbi wezinyathelo eziningi.
Amathiphu Ochwepheshe Kwabameleli Bezinga Lokukhiqiza
Ukusuka ku-demo epholile uye kumkhiqizo othembekile kudinga ukupholisha okwengeziwe. Okokuqala, ukuhlola kubalulekile ukuze kukhunjulwe; kuvumela umenzeli ukuthi akhumbule umongo phakathi kwezingxoxo ezahlukene. Okwesibili, ngezenzo ezibaluleke kakhulu, kufanele usebenzise izisusa zomuntu ngaphakathi kwe-loop , umise igrafu ukuze umuntu wangempela akwazi ukunikeza ukukhanya okuluhlaza. Okokugcina, qaphela ukubambezeleka kanye nezindleko , njengoba yonke i-loop kugrafu ingenye ikholi ye-API ehlanganisa isikhathi nemali.
Imibuzo Evamile Yabasanda Kufika
Akudingeki ube yi-PhD yokufunda ngomshini ukuze wenze lokhu. Umsebenzi omningi empeleni Uhlelo lwe-Python kanye ne-orchestration logic; usebenzisa amamodeli akhona, awuwaqeqeshi kusukela ekuqaleni. Uma unquma phakathi komsizi owakhiwe ngaphambilini kanye negrafu yesandla, khumbula lokho create_react_agent kuhle kakhulu kuma-80% wamacala okusetshenziswa, kuyilapho incwadi yemiyalelo Igrafu Yesimo ilungele lapho udinga ukwenziwa okuhambisanayo noma i-logic yokuzama kabusha eyinkimbinkimbi.
Ukuqonda kahle ukugeleza kwesimo, ama-node, kanye nemiphetho kukuvumela ukuthi uguqule i-LLM elula ibe uhlelo oluguquguqukayo olukwazi ukwenza izinto ngokuzimela . Ngokuhlanganisa izincazelo zamathuluzi ezicacile, ukuphathwa kwememori okuhlakaniphile, kanye nohlaka olufanele lokuhlelwa, ungakha abasizi be-AI abangakhulumi nje kuphela kodwa empeleni bahambe phambili ezinhlelweni zokusebenza zangempela.
