OpenAI open sourced a new Customer Service Agent framework — learn more about its growing enterprise strategy

Publisher’s note: Carl will run an editorial round table on this subject to VB transform next week. Register today.

OPENAI has published a new open source demo which gives developers an overview of how to build intelligent and workflow -conscious AI agents using agents’ SDK.

As Noticed by the influencer and engineer of AI Tibor Blaho (from the Third -party AIPRM pussy browser extension), The new Openai Customer Service Agent was published earlier in the day on the Code of the AI ​​sharing community sharing the embraced face According to a permissive license, which means that any third -party developer or user can take the code, modify it and deploy it for free for their own commercial or experimental crimes.

This example of an agent shows how to send requests related to airlines between specialized agents – such as seats booking, flight status, cancellation and FAQ – while applying security and relevance guards.

The version is designed to help teams go beyond theoretical use and start operational agents with confidence.

This practical demonstration arrives right in front The upcoming presentation of Openai to Venturebeat Transform 2025 Next week in San Francisco, from June 24 to 25, where the chief of the OpenAi platform Olivier Godement Inera deeper into the architecture of a business quality agent fueling use cases in companies like Stripe and Box.

A routing plan, railings and specialized agents

Today’s version includes both a Backend Python and a Fronend Next.js. The Backend operates the SDK of OPENAI agents to orchestrate the interactions between specialized agents, while the frontend visualizes these interactions in a cat interface, showing how decisions and transfer take place in real time.

In a flow, a customer asks to change the seat. The sorting agent determines the request and transports it to the headquarters’ booking agent, which confirms the change of reservation interactively. In another scenario, a flight cancellation request is processed via the cancellation agent, which validates the customer’s confirmation number before completing the task.

Above all, demo also shows how the railings work in production: a Relevance blocks out of score queries as asking for poetry, while a Jailbreak Prevents attempts at rapid injection, such as requests to expose the system instructions.

The architecture reflects the support flows of real world airlines, showing how organizations can create assistants focused on the domain that are reactive, compliant and aligned with user expectations. Openai published the code under the MIT license and encouraged teams to personalize it and adapt it to their own needs.

From open source to the use of real -world companies: Read OPENAI foundations for the construction of practical AI agents

This open source version is based on the wider OPENAI initiative to help teams design and deploy systems based on large -scale agents.

Earlier this year, the company published “A practical guide to build agents», A 32 -page manual for product and engineering teams seeking to implement intelligent automation.

The guide presents fundamental components – LLM model, external tools and behavioral instructions – and covers strategies to build both single agent systems and complex multi -agent architectures. It offers design models for orchestration, the implementation of the goalkeeper and observability, drawing from Openai’s experience by supporting large -scale deployments.

The main points to remember of the guide include:

• Model selection: Use higher level models to establish basic performance lines, then experience smaller models for profitability.
• Tool integration: Equip API agents or external functions to recover data or perform actions.
• Manufacturing of instructions: Use clear and oriented prompts and conditionals oriented to guide the decisions of the agents.
• Railing: Lay on safety, relevance and compliance constraints to ensure safe and predictable behavior.
• Human intervention: Configure the thresholds and climbing paths for cases that require human surveillance.

The guide emphasizes the start of the complexity of small and evolving agents over time – an approach reproduced in the newly released demo, which shows how modular tools for tools can be orchestrated properly.

Find out more about Openai at VB Transform 2025

The teams seeking to go from prototype to production will have a more in -depth overview of the approach ready for the company of Openai during Transform 2025Hosted by Venturebeat.

Currently scheduled For Wednesday June 25 at 3:10 p.m. PTThe session – The year of agents: How Openai offers the next vague intelligent automation wave—The features Olivier Godenment, product manager for the OPENAI API platformin conversation with me, Carl Franzen,, Editor -in -chief at Venturebeat.

The 20 -minute speech will cover:

• Agent architectural models: when to use unique loops, sub-agents or orchestrated transfer.
• Integrated guard-rails for regulated environments, including political refusals, SOC-2 journalization and data residence support.
• Costs / King Costs and Stripe and Box benchmarks, including 35% faster invoice resolution and zero-touch support.
• Roadmap insights: which then happens for multimodal actions, agent memory and cross orchestration.

Whether you are experimenting with open-source tools such as the demo of the customer service agent or scaling agents in critical workflows, this session promises a founded overview of what works, what to avoid and the next step.

Why this counts for companies and developers

Between the newly published demo and the principles described in A practical guide to build agentsOPENAI doubles its strategy: allowing developers to go beyond LLM applications in turn and towards autonomous systems that can include the context, browse the tasks intelligently and operate safely.

By offering transparent tools and clear implementation examples, Openai pushes agent systems outside the laboratory and in daily use, whether in customer service, operations or internal governance. For organizations exploring intelligent automation, these resources provide not only inspiration, but a working manual.