Ukugijima ucwaningo olujulile lwe-AI Agents e-Amazon Bedrock Agentcore

0 0 6 minutes read

Ukugijima ucwaningo olujulile lwe-AI Agents e-Amazon Bedrock Agentcore

Ama-ejenti we-AI avela ngaphezu kwabasizi abayisisekelo abasebenza emisebenzini enamandla amakhulu angahlela, abagxeki, futhi ahlanganyele namanye ama-ejenti ukuxazulula izinkinga eziyinkimbinkimbi. Ama-ejenti ajulile – Uhlaka olusanda kwethulwa owakhelwe kuLanggraph-Letha la makhono okuphila, okunika amandla ukugeleza kwemisebenzi enemisebenzi eminingi evuselela amandla e-Real-World Team Gynamics. Inselele, noma kunjalo, awenzi nje ukwakha ama-ejenti anjalo kodwa futhi ibagijimela ngokuthembekile nangokuphepha ekukhiqizweni. Yilapho lapho kungena khona i-Amazon Bedrock Agentcore Agentcore. Ngokuhlinzeka ngenhloso ephephile, engenasici – eyakhelwe ama-AI ama-Agents namathuluzi, isikhathi sokugijima kwenza sikwazi ukufaka ama-ejenti ajulile ezikalini zebhizinisi ngaphandle kokuphatha izingqalasizinda.

Kulokhu okuthunyelwe, sibonisa ukuthi singathumela kanjani ama-ejenti ajulile ku-Agentcore Runtime. Njengoba kukhonjisiwe kulokhu okulandelayo, i-Agentcore Runtime Scates noma yimuphi umenzeli futhi ihlinzeka ngokwehlukaniswa kabusha nge-microvm entsha yeseshini ngayinye entsha.

Kuyini i-agentcore ye-Amazon Bedrock?

I-Amazon Bedrock AgentCore yomabili uhlaka lwezinhlaka-agnostic kanye nemodeli-agnostic, ikunikeza ukuguquguquka kokuhambisa nokusebenzisa ama-agents athuthukile we-AI ngokuphephile nangesilinganiso. Noma ngabe wakha ama-ejenti ama-strands, uCrewai, uLanggraph, iLlayinidex, noma olunye uhlaka – futhi lubagijima kwimodeli enkulu yolimi (LLM) -Agentcore ihlinzeka ngengqalasizinda ukuze ibasekele. Izinsizakalo zalo ezishukumisayo ziyenzelwe injongo yokulayisha ama-ejenti ashukumisayo, ngamathuluzi wokunweba amandla e-agent kanye nezilawuli ezidingekayo ekusetshenzisweni komkhiqizo. Ngokunciphisa ukuphakanyiswa okusindayo okungapheli kokwakha nokuphatha ingqalasizinda ekhethekile ye-ejenti, i-agentcore ikuvumela ukuthi ulethe uhlaka lwakho oluncamile nemodeli bese uhambisa ngaphandle kwekhodi yokubhala kabusha.

I-Amazon Bedrock Agentcore inikeza amandla aphelele amakhono aklanyelwe ukuguqula ama-agent agent prototymes abe amasistimu alungele ukukhiqizwa. Lokhu kufaka imemori ephikelelayo yokugcina umongo ngaphakathi nasezingxoxweni ezikhona, ukufinyelela ama-API akhona usebenzisa imodeli ye-Protocol (MCP) nezinqubo zokubhekisisa zeWebhu kanye nokuhlolwa okujulile kwezinqubo zokubonisana kwe-Agent. Kulokhu okuthunyelwe, sigxile ngqo kwingxenye ye-Agentcore Runtime.

Amandla asemqondweni we-AgentCore Runtime

I-AgentCore Runtime ihlinzeka nge-Serverless, Indawo Yokubamba Evikelekile eyenzelwe ngqo imithwalo ye-Agentic. Iphawula ikhodi esitsheni esingasindi ngesikhombisi esilula, esingaguquki, okwenza kube lula ukulungele ama-ejenti okusebenzisa ama-multimor, kanye nokusebenza okusetshenziselwa ukusetshenziswa okususelwa ekusetshenzisweni okunezinyusa kuphela lapho kusebenza khona – hhayi ngenkathi Ilinde izimpendulo ze-LLM noma izimpendulo zamathuluzi. Iseshini ngayinye yomsebenzisi isebenza ngokuhlukaniswa okuphelele ngaphakathi kwemishini emincane yokuzinikezela imishini (ama-Microvms), ukugcina ukuphepha nokusiza ukuvikela ukungcoliswa kweseshini yeseshini phakathi kokusebenzisana kwe-agent. Isikhathi sokugijima sisebenza ngohlaka oluningi (ngokwesibonelo: I-Langgraph, Crewaph, imicu, nokunye) kanye nakulabo abahlinzeki bemodeli ye-Agent, ngenkathi bekwazi ukubonwa kwenhlangano ebanzi, kanye nokutholakala okuhlangene kwimvelo ebanzi ye-Agentcore nge-SDK eyodwa nge-SDK eyodwa.

Isibonelo sangempela somhlaba: Ukuhlanganiswa okujulile kwama-ejenti

Kulesi post sizosebenzisa isibonelo esanda kukhishwa ama-Agents Entents Isibonelo ekusebenzeni kwe-Agentcore Runtime-ekhombisa nje ukuthi kuthatha umzamo omncane kanjani ukuthola izinto ezintsha zakamuva zisebenza.

Ukuqaliswa kwesampula kumdwebo owedlule kufaka phakathi:

A umenzeli wokucwaninga lokho kwenza ukusesha okujulile kwe-inthanethi kusetshenziswa i-tavily api
A U-Crique Agent lokho kubuyekezwa futhi kunikeza impendulo emibikweni eyenziwe
A I-Orchestrator eyinhloko ophatha ukugeleza komsebenzi futhi ahlabele imisebenzi yefayela

Ama-ejenti ajulile asebenzisa ukuphathwa kwesimo seLanggraph ukudala uhlelo lwe-ejenti ehlukahlukene nge:

Ukuhlelwa Kwemisebenzi Yakhelwe ngaphakathi ngokusebenzisa a write_todos Ithuluzi elisiza ama-ejenti ahlukanise izicelo eziyinkimbinkimbi
Uhlelo lwefayela le-Virtual lapho ama-ejenti angafunda / abhale khona amafayela ukuze alondoloze umongo kokusebenzisana
Isakhiwo se-Sub-Agent Ukuvumela ama-ejenti akhethekile ukuthi asuswe emisebenzini ethile ngenkathi egcina ukwahlukaniswa umongo
Ukubonisana kabusha kabusha ngemikhawulo ephezulu yokuphinda (ngaphezulu kwe-1 000) ukuphatha ukugeleza okunzima, okuma-multi-step

Lo mbuso unika amandla ama-ejenti ajulile ukuthi aphathe imisebenzi yocwaningo edinga ukuhlanganiswa okuningi kwemibuthano yolwazi, ukuhlanganiswa, nokuhlaziywa kwamaphoyinti okuhlanganisa asemqoka ku-Agentcore yethu. Ubuhle buwukulula kwayo – sidinga kuphela ukwengeza imigqa embalwa yekhodi ukwenza i-agentcore ehambisanayo ye-ejenti:

# 1. Import the AgentCore runtime
from bedrock_agentcore.runtime import BedrockAgentCoreApp
app = BedrockAgentCoreApp()

# 2. Decorate your agent function with @app.entrypoint
@app.entrypoint
async def langgraph_bedrock(payload):
    # Your existing agent logic remains unchanged
    user_input = payload.get("prompt")
    
    # Call your agent as before
    stream = agent.astream(
        {"messages": [HumanMessage(content=user_input)]},
        stream_mode="values"
    )
    
    # Stream responses back
    async for chunk in stream:
        yield(chunk)

# 3. Add the runtime starter at the bottom
if __name__ == "__main__":
    app.run()

Yilokho kuphela! Ukuqalwa kwemodeli ye-Code-Model, ukuhlanganiswa kwe-API, kanye ne-Agent Logic – kuhlala njengoba kwakunjalo. I-AgentCore iphatha ingqalasizinda ngenkathi i-ejenti yakho ibhekana nobuhlakani. Le ndlela yokuhlanganisa isebenzela izinhlaka eziningi ze-ejenti yePython, okwenza uhlaka lwe-agentcore ngempela-agnostic.

Ukuphamba esikhathini sokugijimisana kwe-AgentCore: Isinyathelo-ngesinyathelo

Ake sihambe ngenqubo yangempela yokuhambisa usebenzisa i-Agentcore Starter ToolKit, okube lula kakhulu ukuhamba komsebenzi okwenziwayo.

Izimfuneko

Ngaphambi kokuthi uqale, qiniseka ukuthi unayo:

Python 3.10 noma ngaphezulu
Iziqinisekiso ze-AWS ezilungiselelwe
Kufakwe i-Amazon Bedrock Agentcore SDK efakiwe

Isinyathelo 1: Izimvume ze-IAM

Kunama-Workity ama-AWS amabili ahlukene kanye nokuphathwa kokufinyelela (IAM) Izimvume okudingeka uzicabangele lapho uthumela i-ejenti ngesikhathi sokugijimisana kwe-AgentTime-Indima oyisebenzisayo ukudala izinsizakusebenza ze-AlentCore kanye neqhaza le-ejenti elidinga ukugijima ngesikhathi sokusebenza kwe-agentcore. Ngenkathi indalo yokugcina manje ingenziwa ngokuzenzakalela yi-Agentcore Starter Toolkit (auto_create_execution_role=True), okwedlule kumele kuchazwe njengoba kuchazwe kwizimvume ze-IAM zesikhathi sokusebenza kwe-Agentcore.

Isinyathelo 2: Faka i-wrapper kumenzeli wakho

Njengoba kukhonjisiwe kuma-ejenti ajulile ajulile, engeza okungeniswa kwe-agententcore kanye nomhlobiso kwikhodi yakho ye-ejenti ekhona.

Isinyathelo 3: Sebenzisa ukusebenzisa i-Agentcore Starter Toolkit

I-Starter Toolkit ihlinzeka ngenqubo yokuhambisa izinto ezintathu:

from bedrock_agentcore_starter_toolkit import Runtime

# Step 1: Configure
agentcore_runtime = Runtime()
config_response = agentcore_runtime.configure(
    entrypoint="hello.py", # contains the code we showed earlier in the post
    execution_role=role_arn, # or auto-create
    auto_create_ecr=True,
    requirements_file="requirements.txt",
    region="us-west-2",
    agent_name="deepagents-research"
)

# Step 2: Launch
launch_result = agentcore_runtime.launch()
print(f"Agent deployed! ARN: {launch_result['agent_arn']}")

# Step 3: Invoke
response = agentcore_runtime.invoke({
    "prompt": "Research the latest developments in quantum computing"
})

Isinyathelo 4: Kwenzekani ngemuva kwezigcawu

Uma uqala ukuthunyelwa, i-Starter Kit ngokuzenzakalelayo:

Kwakha ifayela le-docker elenziwe kahle Nge-Python 3.13-Slim Base Image kanye ne-OpenterElemetry Instrucation
Wakha isitsha sakho ngokuncika kusuka requirements.txt
Kwakha i I-Amazon Elastic Choineer Registry (Amazon ECR) indawo yokugcina impahla (if auto_create_ecr=True) futhi icindezela isithombe sakho
Ukuphamba ku-Agentcore Runtime futhi iqaphe isimo sokuhanjiswa
Ilungiselela ukuxhumana kanye nokuqashelwa Nge-Amazon CloudWatch kanye ne-AWS X-ray ukuhlanganiswa

Yonke inqubo imvamisa ithatha imizuzu engama-2-3, ngemuva kwalokho i-ejenti yakho isilungele ukuphatha izicelo esikalini. Iseshini ngasinye esisha sethulwa endaweni yaso ye-Agentcore Runtime ye-Agentshore entsha, ukulondolozela ukuhlukaniswa okuphelele kwemvelo.

I-Starter Kit ikhiqiza ifayela lokucushwa (.bedrock_agentcore.yaml) Lokho kuthumba izilungiselelo zakho zokuhanjiswa, okwenza kuqonde ukuthumela kabusha noma ukubuyekeza umenzeli wakho ngokuhamba kwesikhathi.

Ukukhipha i-ejenti yakho esetshenzisiwe

Ngemuva kokuthunyelwa, unezinketho ezimbili zokucela umenze wakho:

Inketho 1: Kusetshenziswa i-Start Kit (kukhonjiswe kusinyathelo 3)

response = agentcore_runtime.invoke({
    "prompt": "Research the latest developments in quantum computing"
})

Inketho 2: Usebenzisa iBoto3 SDK ngqo

import boto3
import json

agentcore_client = boto3.client('bedrock-agentcore', region_name="us-west-2")
response = agentcore_client.invoke_agent_runtime(
    agentRuntimeArn=agent_arn,
    qualifier="DEFAULT",
    payload=json.dumps({
        "prompt": "Analyze the impact of AI on healthcare in 2024"
    })
)

# Handle streaming response
for event in response['completion']:
    if 'chunk' in event:
        print(event['chunk']['bytes'].decode('utf-8'))

Ama-ejenti ajulile asebenza

Njengoba ikhodi ikhipha e-Bedrock Agentcore Runtime, i-ejenti ye-ejenti yama-ejenti yama-orchestrates akhethekile akhethekile ama-sub-agents – ngalinye linenhloso yalo, ngokushesha, ukufinyelela kwamathuluzi-ukuxazulula imisebenzi eyinkimbinkimbi. Kulokhu, i-Orchestrator Prompt (research_instructions) Ibeka uhlelo:

Bhala umbuzo oya kumbuzo.txt
Abalandeli ku-One noma ngaphezulu ocwaningweni – Izingcingo ze-Agent (ngayinye ngesihloko esisodwa) Esebenzisa Ithuluzi le-Intanethi
Synthesize okutholakele ku-Final_report.md
Shayela i-critique-ejenti ukuhlola izikhala nesakhiwo
Ngokuzithandela ubheke emuva ocwaningweni / ekuhlelweni kuze kube yilapho ikhwalithi iyahlangana

Nakhu kusebenza:

Hlanza

Uma usuqedile, ungakhohlwa ukukhipha i-Agent Curening Isikhathi sokusebenza kwe-AgentCore ngokungeziwe kwindawo yokugcina impahla eyakhiwe ngesikhathi senqubo:

agentcore_control_client = boto3.client(
    'bedrock-agentcore-control', region_name=region )
ecr_client = boto3.client('ecr',region_name=region )
runtime_delete_response = agentcore_control_client.delete_agent_runtime(    agentRuntimeId=launch_result.agent_id,)
response = ecr_client.delete_repository(
    repositoryName=launch_result.ecr_uri.split('/')[1],force=True)

Ukugcina

I-Amazon Bedrock AgentCore imele i-paradigm shift ngendlela sifaka kanjani ama-AI ADERS. Ngokukhipha inkimbinkimbi yengqalasizinda lapho ugcina uhlaka kanye nokuguquguquka kwemodeli, i-agentcore inika amandla abathuthukisi ukuthi bagxile ekwakheni logic e-agent enobuhlakani kunokuba baphathe amapayipi okuthumela. Ukuhanjiswa kwama-ejenti wethu okujulile kukhombisa ukuthi izinhlelo eziyinkimbinkimbi, ezinemisebenzi eminingi ngokuhlanganiswa kwangaphandle kwe-API kungathunyelwa ngezinguquko zekhodi elincane. Inhlanganisela yokuphepha kwebanga lebhizinisi, ukubonwa okwakhelwe ngaphakathi, futhi ukusakaza kusengaphambili kwenza i-agentcore ikhetheke kakhulu ekukhiqizweni kwe-Agent Deploments. Ngokukhethekile ama-ejenti ocwaningo olujulile, i-agentcore inikezela ngamakhono alandelayo ahlukile ongawahlola:

I-AgentCore Runtime ingaphatha ukucubungula i-asynchronous kanye nokugijima isikhathi eside (kuze kube amahora angama-8) ama-ejenti. Imisebenzi ye-Asynchronous Vumela i-ejenti yakho ukuthi iqhubeke icubungule ngemuva kokuphendula iklayenti futhi ibambe imisebenzi esebenza isikhathi eside ngaphandle kokuvimba izimpendulo. Umenzeli wakho ongemuva wangemuva ungacwaningisisa amahora amaningi.
I-AgentCore Runtime isebenza ngememori ye-AgentCore, amakhono anika amandla anjengesakhiwo okutholakele kwangaphambilini, akhumbule ukukhethwa kokucwaninga, futhi alondoloze ingqikithi yokuphenya okuyinkimbinkimbi ngaphandle kokulahlekelwa inqubekela phambili phakathi kwamaseshini.
Ungasebenzisa isango le-AgentCore ukwelula ucwaningo lwakho olujulile ukufaka phakathi ukuqonda okuphathelene nezinsizakalo zebhizinisi nemithombo yedatha. Ngokudalula lezi zinsiza ezihlukanisiwe njengamathuluzi we-MCP, ama-ejenti akho angawasebenzisa ngokushesha futhi ahlanganise lokho ngolwazi olutholakala emphakathini.

Ukulungele ukuthumela ama-ejenti akho ukukhiqiza? Nakhu ukuthi ungaqala kanjani:

Faka i-Agentcore Starter Kit: pip install bedrock-agentcore-starter-toolkit
Umsebenzi wokulingi: Sebenzisa ikhodi yakho ngokulandela lesi sinyathelo ngesinyathelo umhlahlandlela.

Inkathi yabahlinzeki be-AI abalungele ukukhiqizwa lapha. Nge-AgentCore, uhambo olusuka kwiPrototype ukukhiqizwa alukaze lube mfushane.

Mayelana nababhali

Vadim Oneltchenko Ingabe ukwakhiwa kwezixazululo ze-SR. AI / ML Solutions othanda ukusiza amakhasimende ama-AWS asungula efwini. Ukuhlangenwe nakho kwakhe kwalo kwangaphambilini kwakuwumhlaba.

I-Eashan Kaushik Ingabe izakhiwo ezikhethekile zokwakha i-AI / ml ezinsizakalweni ze-Amazon Web. Uqhutshwa ngokwakha izixazululo zokusika ze-AI onqenqemeni ngenkathi ebeka phambili indlela yamakhasimende – i-Centerric eya emsebenzini wakhe. Ngaphambi kwale ndima, wathola ama-MS kwisayensi yekhompyutha evela eNyu Tandon School of Engineering. Ngaphandle komsebenzi, uyayijabulela ezemidlalo, ukuphakamisa, nokugijima amajaha.

Shreyas Subramannian Ungusosayensi wedatha oyinhloko futhi usiza amakhasimende ngokusebenzisa umshini wokufunda ukuxazulula izinselelo zebhizinisi labo usebenzisa ipulatifomu ye-AWS. UShreyas unesizinda sokwenza kahle kwesilinganiso esikhulu kanye nokufunda komshini, nokusetshenziswa kokufunda komshini nokuqinisa ukuqinisa ukusheshisa imisebenzi yokwenza kahle.

UMark Roy Ingabe umdwebi womshini oyinhloko wokufunda umshini wama-AWS, usiza amakhasimende aklame futhi wakhe izixazululo ze-AI akhiqizayo. Ukugxila kwakhe kusukela ekuqaleni kuka-2023 bekulokhu kuholela emizamweni yokwakha isixazululo sokwethulwa kwe-Amazon Bedrock, umnikelo we-AI Ovelayo we-AI kusuka kuma-AWS. Umsebenzi kaMark uhlanganisa uhla olubanzi lokusebenzisa amacala, ngentshiseko eyinhloko ekwenzeni i-AI, ama-ejenti, kanye nokulinganisa i-ML ngaphesheya kwebhizinisi. Uye wasiza ezinkampanini zemithi yomshuwalense, izinsizakalo zezezimali, kwabezindaba kanye nokuzijabulisa, ukunakekelwa kwezempilo, ezinsizakalweni nasekukhiqizweni. Ngaphambi kokujoyina ama-AWS, uMark wayengumdwebi wezakhiwo, unjiniyela, kanye nobuchwepheshe bezobuchwepheshe iminyaka engaphezu kwengu-25, kubandakanya neminyaka eyi-19 ezinsizakalweni zezezimali. UMark ubamba izitifiketi ze-AWS eziyisithupha, kufaka phakathi isitifiketi esikhethekile se-ML.