
Inglês 中文
O Prompulate é uma estrutura de desenvolvimento de aplicativos de agentes de IA criada pelo Cogit Lab , que oferece aos desenvolvedores uma maneira extremamente concisa e eficiente de criar aplicativos de agentes por meio de um paradigma de desenvolvimento pitônico. A filosofia central do Promptulate é emprestar e integrar a sabedoria da comunidade de código aberto, incorporando os destaques de várias estruturas de desenvolvimento para diminuir a barreira à entrada e unificar o consenso entre os desenvolvedores. Com o Prompulate, você pode manipular componentes como LLM, agente, ferramenta, pano etc., com o código mais sucinto, pois a maioria das tarefas pode ser facilmente concluída com apenas algumas linhas de código.
pne.chat para encapsular todas as funcionalidades essenciais. O diagrama a seguir mostra a arquitetura principal do promptulate :

O conceito central de Promptulle é que esperamos fornecer uma maneira simples, pitônica e eficiente de criar aplicativos de IA, o que significa que você não precisa gastar muito tempo aprendendo a estrutura. Esperamos usar pne.chat() para fazer a maioria dos trabalhos e você pode criar facilmente qualquer aplicativo de IA com apenas algumas linhas de código.
Abaixo,
pnesignifica Promptulate, que é o apelido para o Promotate. Ospeerepresentam o início e o fim do Promptulate, respectivamente, enrepresenta 9, que é uma abreviação para as nove letras entrepee
O Promptulate integra as capacidades da Litellm, apoiando quase todos os tipos de modelos grandes no mercado, incluindo, entre outros, os seguintes modelos:
| Provedor | Conclusão | Transmissão | Conclusão assíncrona | Streaming assíncrono | Incorporação assíncrona | Geração de imagem assíncrona |
|---|---|---|---|---|---|---|
| Openai | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Azure | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| AWS - Sagemaker | ✅ | ✅ | ✅ | ✅ | ✅ | |
| AWS - Bedrock | ✅ | ✅ | ✅ | ✅ | ✅ | |
| Google - Vertex_ai [Gemini] | ✅ | ✅ | ✅ | ✅ | ||
| Google - Palm | ✅ | ✅ | ✅ | ✅ | ||
| Google AI Studio - Gêmeos | ✅ | ✅ | ||||
| API MISTRAL AI | ✅ | ✅ | ✅ | ✅ | ✅ | |
| CloudFlare AI Workers | ✅ | ✅ | ✅ | ✅ | ||
| coere | ✅ | ✅ | ✅ | ✅ | ✅ | |
| Antrópico | ✅ | ✅ | ✅ | ✅ | ||
| huggingface | ✅ | ✅ | ✅ | ✅ | ✅ | |
| replicar | ✅ | ✅ | ✅ | ✅ | ||
| juntos_ai | ✅ | ✅ | ✅ | ✅ | ||
| OpenRouter | ✅ | ✅ | ✅ | ✅ | ||
| AI21 | ✅ | ✅ | ✅ | ✅ | ||
| BASETEN | ✅ | ✅ | ✅ | ✅ | ||
| vllm | ✅ | ✅ | ✅ | ✅ | ||
| nlp_cloud | ✅ | ✅ | ✅ | ✅ | ||
| Aleph Alpha | ✅ | ✅ | ✅ | ✅ | ||
| Pétalas | ✅ | ✅ | ✅ | ✅ | ||
| Ollama | ✅ | ✅ | ✅ | ✅ | ||
| Deepinfra | ✅ | ✅ | ✅ | ✅ | ||
| perplexidade-ai | ✅ | ✅ | ✅ | ✅ | ||
| Groq ai | ✅ | ✅ | ✅ | ✅ | ||
| AnyScale | ✅ | ✅ | ✅ | ✅ | ||
| Viagem ai | ✅ | |||||
| Xinference [Xorbits Inference] | ✅ |
O poderoso suporte de modelo do PNE permite que você construa facilmente quaisquer chamadas de modelo de terceiros.
Agora vamos ver como administrar modelos locais de Ollama com PNE.
import promptulate as pne
resp : str = pne . chat ( model = "ollama/llama2" , messages = [{ "content" : "Hello, how are you?" , "role" : "user" }])? 2024.5.14 O OpenAI lançou seu mais novo modelo "Omni", oferecendo uma velocidade e preços aprimorados em comparação com o Turbo.
Você pode usar os recursos multimodais disponíveis em qualquer um dos seus aplicativos PROMPRULADOS!
import promptulate as pne
messages = [
{
"role" : "user" ,
"content" : [
{ "type" : "text" , "text" : "What's in this image?" },
{
"type" : "image_url" ,
"image_url" : "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" ,
},
],
}
]
resp = pne . chat ( model = "gpt-4o" , messages = messages )
print ( resp ) Use provider/model_name para chamar o modelo e você pode criar facilmente quaisquer chamadas de modelo de terceiros.
Para mais modelos, visite a documentação da Litellm.
Você também pode ver como usar pne.chat() na documentação inicial/oficial.
Construa Chatbot Streamlit por pne
Construa graduado chatbot por pne
Construir aplicativo de matemática com agente
Groq, llama3, simplit para criar um aplicativo
Construa mapa de conhecimento com streamlit e pne
Construa um chatbot usando pne+simplit para conversar com o repo do github
Crie um aplicativo de matemática com agente [SteamLit, ToolAgent, Hooks].
Uma estrutura de agente de robôs Mulitmodal do ROS2 e Promotate [agente]
Use StreamLit e PNE para comparar diferentes modelos A Playground. [Streamlit]
GCOP: seu copiloto Git AI, baseado em Promotate
pip install -U pneNOTA: Sua versão Python deve ser 3,8 ou superior.
Embora o PNE forneça muitos módulos, em 90% dos cenários de desenvolvimento de aplicativos LLM, você só precisa usar a função pne.chat (), para que você só precisa começar com o bate -papo para entender o uso de PNE e quando precisar usar módulos adicionais, você pode aprender mais sobre os recursos e o uso de outros módulos.
Você pode usar pne.chat() para conversar como o OpenAI. Documento da API de bate-papo OpenAI: https://platform.openai.com/docs/api-reference/chat
import promptulate as pne
messages = [
{ "role" : "system" , "content" : "You are a helpful assistant." },
{ "role" : "user" , "content" : "Who are you?" },
]
response : str = pne . chat ( messages = messages , model = "gpt-4-turbo" )
print ( response ) Muitas bibliotecas de terceiros podem usar o OpenAI SDK chama seus modelos, como o DeepSeek. No PNE, você pode usar diretamente a função pne.chat() para chamar esses modelos, ele não precisa usar o SDK OpenAI e fornece recursos aprimorados para simplificar a dificuldade de desenvolvimento. Use o prefixo do provedor openai/xxx no modelo e você pode usar o modelo OpenAI para fazer chamadas.
import os
import promptulate as pne
os . environ [ "DEEPSEEK_API_KEY" ] = "your api key"
messages = [
{ "role" : "system" , "content" : "You are a helpful assistant." },
{ "role" : "user" , "content" : "How are you?" },
]
response = pne . chat (
messages = messages ,
model = "openai/deepseek-chat" ,
)
print ( response )A formatação robusta de saída é uma base fundamental para o desenvolvimento de aplicativos LLM. Esperamos que o LLMS possa retornar dados estáveis. Com o PNE, você pode executar facilmente a saída formatada. No exemplo a seguir, usamos o modelo Basemodel da Pydantic para encapsular uma estrutura de dados que precisa ser retornada.
from typing import List
import promptulate as pne
from pydantic import BaseModel , Field
class LLMResponse ( BaseModel ):
provinces : List [ str ] = Field ( description = "List of provinces' names" )
resp : LLMResponse = pne . chat ( "Please tell me all provinces in China." , output_schema = LLMResponse )
print ( resp )Saída:
provinces=['Anhui', 'Fujian', 'Gansu', 'Guangdong', 'Guizhou', 'Hainan', 'Hebei', 'Heilongjiang', 'Henan', 'Hubei', 'Hunan', 'Jiangsu', 'Jiangxi', 'Jilin', 'Liaoning', 'Qinghai', 'Shaanxi', 'Shandong', 'Shanxi', 'Sichuan', 'Yunnan', 'Zhejiang', 'Taiwan', 'Guangxi', 'Nei Mongol', 'Ningxia', 'Xinjiang', 'Xizang', 'Beijing', 'Chongqing', 'Shanghai', 'Tianjin', 'Hong Kong', 'Macao']
Além disso, influenciado pelo artigo de plano e solução, a PNE também permite que os desenvolvedores construam agentes capazes de lidar com problemas complexos por meio de planejamento, raciocínio e ação. As habilidades de planejamento do agente podem ser ativadas usando o parâmetro enable_plan .

Neste exemplo, usamos Tavily como mecanismo de pesquisa, que é uma ferramenta poderosa para pesquisar informações na web. Para usar a Tavily, você precisa obter uma chave da API da Tavily.
import os
os . environ [ "TAVILY_API_KEY" ] = "your_tavily_api_key"
os . environ [ "OPENAI_API_KEY" ] = "your_openai_api_key"Nesse caso, estamos usando a ferramenta TavilySearchResults embrulhada por Langchain.
from langchain_community . tools . tavily_search import TavilySearchResults
tools = [ TavilySearchResults ( max_results = 5 )] import promptulate as pne
pne . chat ( "what is the hometown of the 2024 Australia open winner?" , model = "gpt-4-1106-preview" , enable_plan = True )Saída:
[Agent] Assistant Agent start...
[User instruction] what is the hometown of the 2024 Australia open winner?
[Plan] {"goals": ["Find the hometown of the 2024 Australian Open winner"], "tasks": [{"task_id": 1, "description": "Identify the winner of the 2024 Australian Open."}, {"task_id": 2, "description": "Research the identified winner to find their place of birth or hometown."}, {"task_id": 3, "description": "Record the hometown of the 2024 Australian Open winner."}], "next_task_id": 1}
[Agent] Tool Agent start...
[User instruction] Identify the winner of the 2024 Australian Open.
[Thought] Since the current date is March 26, 2024, and the Australian Open typically takes place in January, the event has likely concluded for the year. To identify the winner, I should use the Tavily search tool to find the most recent information on the 2024 Australian Open winner.
[Action] tavily_search_results_json args: {'query': '2024 Australian Open winner'}
[Observation] [{'url': 'https://ausopen.com/articles/news/sinner-winner-italian-takes-first-major-ao-2024', 'content': 'The agile right-hander, who had claimed victory from a two-set deficit only once previously in his young career, is the second Italian man to achieve singles glory at a major, following Adriano Panatta in1976.With victories over Andrey Rublev, 10-time AO champion Novak Djokovic, and Medvedev, the Italian is the youngest player to defeat top 5 opponents in the final three matches of a major since Michael Stich did it at Wimbledon in 1991 – just weeks before Sinner was born.n He saved the only break he faced with an ace down the tee, and helped by scoreboard pressure, broke Medvedev by slamming a huge forehand to force an error from his more experienced rival, sealing the fourth set to take the final to a decider.n Sensing a shift in momentum as Medvedev served to close out the second at 5-3, Sinner set the RLA crowd alight with a pair of brilliant passing shots en route to creating a break point opportunity, which Medvedev snuffed out with trademark patience, drawing a forehand error from his opponent. “We are trying to get better every day, even during the tournament we try to get stronger, trying to understand every situation a little bit better, and I’m so glad to have you there supporting me, understanding me, which sometimes it’s not easy because I am a little bit young sometimes,” he said with a smile.n Medvedev, who held to love in his first three service games of the second set, piled pressure on the Italian, forcing the right-hander to produce his best tennis to save four break points in a nearly 12-minute second game.n'}, {'url': 'https://www.cbssports.com/tennis/news/australian-open-2024-jannik-sinner-claims-first-grand-slam-title-in-epic-comeback-win-over-daniil-medvedev/', 'content': '"nOur Latest Tennis StoriesnSinner makes epic comeback to win Australian OpennSinner, Sabalenka win Australian Open singles titlesn2024 Australian Open odds, Sinner vs. Medvedev picksnSabalenka defeats Zheng to win 2024 Australian Openn2024 Australian Open odds, Sabalenka vs. Zheng picksn2024 Australian Open odds, Medvedev vs. Zverev picksnAustralian Open odds: Djokovic vs. Sinner picks, betsnAustralian Open odds: Gauff vs. Sabalenka picks, betsnAustralian Open odds: Zheng vs. Yastremska picks, betsnNick Kyrgios reveals he's contemplating retirementn© 2004-2024 CBS Interactive. Jannik Sinner claims first Grand Slam title in epic comeback win over Daniil MedvedevnSinner, 22, rallied back from a two-set deficit to become the third ever Italian Grand Slam men's singles championnAfter almost four hours, Jannik Sinner climbed back from a two-set deficit to win his first ever Grand Slam title with an epic 3-6, 3-6, 6-4, 6-4, 6-3 comeback victory against Daniil Medvedev. Sinner became the first Italian man to win the Australian Open since 1976, and just the eighth man to successfully come back from two sets down in a major final.n He did not drop a single set until his meeting with Djokovic, and that win in itself was an accomplishment as Djokovic was riding a 33-match winning streak at the Australian Open and had never lost a semifinal in Melbourne.n @janniksin • @wwos • @espn • @eurosport • @wowowtennis pic.twitter.com/DTCIqWoUoRn"We are trying to get better everyday, and even during the tournament, trying to get stronger and understand the situation a little bit better," Sinner said.'}, {'url': 'https://www.bbc.com/sport/tennis/68120937', 'content': 'Live scores, results and order of playnAlerts: Get tennis news sent to your phonenRelated TopicsnTop StoriesnFA Cup: Blackburn Rovers v Wrexham - live text commentarynRussian skater Valieva given four-year ban for dopingnLinks to Barcelona are 'totally untrue' - ArtetanElsewhere on the BBCnThe truth behind the fake grooming scandalnFeaturing unseen police footage and interviews with the officers at the heart of the casenDid their father and uncle kill Nazi war criminals?n A real-life murder mystery following three brothers in their quest for the truthnWhat was it like to travel on the fastest plane?nTake a behind-the-scenes look at the supersonic story of the ConcordenToxic love, ruthless ambition and shocking betrayalnTell Me Lies follows a passionate college relationship with unimaginable consequences...n "nMarathon man Medvedev runs out of steamnMedvedev is the first player to lose two Grand Slam finals after winning the opening two setsnSo many players with the experience of a Grand Slam final have talked about how different the occasion can be, particularly if it is the first time, and potentially overwhelming.n Jannik Sinner beats Daniil Medvedev in Melbourne finalnJannik Sinner is the youngest player to win the Australian Open men's title since Novak Djokovic in 2008nJannik Sinner landed the Grand Slam title he has long promised with an extraordinary fightback to beat Daniil Medvedev in the Australian Open final.n "nSinner starts 2024 in inspired formnSinner won the first Australian Open men's final since 2005 which did not feature Roger Federer, Rafael Nadal or Novak DjokovicnSinner was brought to the forefront of conversation when discussing Grand Slam champions in 2024 following a stunning end to last season.n'}]
[Execute Result] {'thought': "The search results have provided consistent information about the winner of the 2024 Australian Open. Jannik Sinner is mentioned as the winner in multiple sources, which confirms the answer to the user's question.", 'action_name': 'finish', 'action_parameters': {'content': 'Jannik Sinner won the 2024 Australian Open.'}}
[Execute] Execute End.
[Revised Plan] {"goals": ["Find the hometown of the 2024 Australian Open winner"], "tasks": [{"task_id": 2, "description": "Research Jannik Sinner to find his place of birth or hometown."}, {"task_id": 3, "description": "Record the hometown of Jannik Sinner, the 2024 Australian Open winner."}], "next_task_id": 2}
[Agent] Tool Agent start...
[User instruction] Research Jannik Sinner to find his place of birth or hometown.
[Thought] To find Jannik Sinner's place of birth or hometown, I should use the search tool to find the most recent and accurate information.
[Action] tavily_search_results_json args: {'query': 'Jannik Sinner place of birth hometown'}
[Observation] [{'url': 'https://www.sportskeeda.com/tennis/jannik-sinner-nationality', 'content': "During the semifinal of the Cup, Sinner faced Djokovic for the third time in a row and became the first player to defeat him in a singles match. Jannik Sinner NationalitynJannik Sinner is an Italian national and was born in Innichen, a town located in the mainly German-speaking area of South Tyrol in northern Italy. A. Jannik Sinner won his maiden Masters 1000 title at the 2023 Canadian Open defeating Alex de Minaur in the straight sets of the final.n Apart from his glorious triumph at Melbourne Park in 2024, Jannik Sinner's best Grand Slam performance came at the 2023 Wimbledon, where he reached the semifinals. In 2020, Sinner became the youngest player since Novak Djokovic in 2006 to reach the quarter-finals of the French Open."}, {'url': 'https://en.wikipedia.org/wiki/Jannik_Sinner', 'content': "At the 2023 Australian Open, Sinner lost in the 4th round to eventual runner-up Stefanos Tsitsipas in 5 sets.[87]nSinner then won his seventh title at the Open Sud de France in Montpellier, becoming the first player to win a tour-level title in the season without having dropped a single set and the first since countryman Lorenzo Musetti won the title in Naples in October 2022.[88]nAt the ABN AMRO Open he defeated top seed and world No. 3 Stefanos Tsitsipas taking his revenge for the Australian Open loss, for his biggest win ever.[89] At the Cincinnati Masters, he lost in the third round to Félix Auger-Aliassime after being up a set, a break, and 2 match points.[76]nSeeded 11th at the US Open, he reached the fourth round after defeating Brandon Nakashima in four sets.[77] Next, he defeated Ilya Ivashka in a five set match lasting close to four hours to reach the quarterfinals for the first time at this Major.[78] At five hours and 26 minutes, it was the longest match of Sinner's career up until this point and the fifth-longest in the tournament history[100] as well as the second longest of the season after Andy Murray against Thanasi Kokkinakis at the Australian Open.[101]nHe reached back to back quarterfinals in Wimbledon after defeating Juan Manuel Cerundolo, Diego Schwartzman, Quentin Halys and Daniel Elahi Galan.[102] He then reached his first Major semifinal after defeating Roman Safiullin, before losing to Novak Djokovic in straight sets.[103] In the following round in the semifinals, he lost in straight sets to career rival and top seed Carlos Alcaraz who returned to world No. 1 following the tournament.[92] In Miami, he reached the quarterfinals of this tournament for a third straight year after defeating Grigor Dimitrov and Andrey Rublev, thus returning to the top 10 in the rankings at world No. In the final, he came from a two-set deficit to beat Daniil Medvedev to become the first Italian player, male or female, to win the Australian Open singles title, and the third man to win a Major (the second of which is in the Open Era), the first in 48 years.[8][122]"}, {'url': 'https://www.thesportreview.com/biography/jannik-sinner/', 'content': '• Date of birth: 16 August 2001n• Age: 22 years oldn• Place of birth: San Candido, Italyn• Nationality: Italiann• Height: 188cm / 6ft 2insn• Weight: 76kg / 167lbsn• Plays: Right-handedn• Turned Pro: 2018n• Career Prize Money: US$ 4,896,338n• Instagram: @janniksinnThe impressive 22-year-old turned professional back in 2018 and soon made an impact on the tour, breaking into the top 100 in the world rankings for the first time in 2019.n Jannik Sinner (Photo: Dubai Duty Free Tennis Championships)nSinner ended the season as number 78 in the world, becoming the youngest player since Rafael Nadal in 2003 to end the year in the top 80.n The Italian then ended the 2019 season in style, qualifying for the 2019 Next Gen ATP Finals and going on to win the tournament with a win over Alex de Minaur in the final.n Sinner then reached the main draw of a grand slam for the first time at the 2019 US Open, when he came through qualifying to reach the first round, where he lost to Stan Wawrinka.n Asked to acknowledge some of the key figures in his development, Sinner replied: “I think first of all, my family who always helped me and gave me the confidence to actually change my life when I was 13-and-a-half, 14 years old.n'}]
[Execute Result] {'thought': 'The search results have provided two different places of birth for Jannik Sinner: Innichen and San Candido. These are actually the same place, as San Candido is the Italian name and Innichen is the German name for the town. Since the user asked for the place of birth or hometown, I can now provide this information.', 'action_name': 'finish', 'action_parameters': {'content': 'Jannik Sinner was born in San Candido (Italian) / Innichen (German), Italy.'}}
[Execute] Execute End.
[Revised Plan] {"goals": ["Find the hometown of the 2024 Australian Open winner"], "tasks": [], "next_task_id": null}
[Agent Result] Jannik Sinner was born in San Candido (Italian) / Innichen (German), Italy.
[Agent] Agent End.
No cenário do desenvolvimento do agente, geralmente precisamos dividir muitos componentes atômicos dos agentes para obter uma melhor personalização. O PNE fornece componentes de agentes atomizados, como o planejador. O exemplo a seguir mostra o uso de um componente planejador separado para o planejamento de tarefas.
import promptulate as pne
model = pne . LLMFactory . build ( "gpt-4-turbo" )
planner = pne . Planner ( model , system_prompt = "You are a planner" )
plans = planner . run ( "Plan a trip to Paris" )
print ( plans )Saída:
('goals', ['Plan a trip to Paris'])
('tasks', [Task(task_id=1, description='Check passport validity', status=<TaskStatus.TODO: 'todo'>), Task(task_id=2, description='Determine travel dates', status=<TaskStatus.TODO: 'todo'>), Task(task_id=3, description='Research and book flights', status=<TaskStatus.TODO: 'todo'>), Task(task_id=4, description='Book accommodations', status=<TaskStatus.TODO: 'todo'>), Task(task_id=5, description='Plan itinerary for the trip', status=<TaskStatus.TODO: 'todo'>), Task(task_id=6, description='Investigate and purchase travel insurance', status=<TaskStatus.TODO: 'todo'>), Task(task_id=7, description='Set a budget for the trip', status=<TaskStatus.TODO: 'todo'>), Task(task_id=8, description='Pack luggage', status=<TaskStatus.TODO: 'todo'>), Task(task_id=9, description='Notify bank of international travel', status=<TaskStatus.TODO: 'todo'>), Task(task_id=10, description='Check weather forecast and pack accordingly', status=<TaskStatus.TODO: 'todo'>)])
('next_task_id', 1)
Para obter informações mais detalhadas, consulte a documentação inicial/oficial.
Os princípios de design da estrutura PNE incluem modularidade, extensibilidade, interoperabilidade, robustez, manutenção, segurança, eficiência e usabilidade.
Seguindo esses princípios e a aplicação das mais recentes tecnologias de inteligência artificial, pne visa fornecer uma estrutura poderosa e flexível para a criação de agentes automatizados.
Para mais informações, entre em contato com: [email protected]
Veja qualquer coisa Changelog, descreva o canal do Telegram
Agradecemos o seu interesse em contribuir para a nossa iniciativa de código aberto. Fornecemos um guia do desenvolvedor descrevendo as etapas para contribuir para solicitar. Consulte este guia para garantir colaboração suave e contribuições bem -sucedidas. Além disso, você pode ver o plano de desenvolvimento atual para ver o mais recente progresso do desenvolvimento?