Gen Ai Architect at a tech vendor with 10,001+ employees
Real User
Top 20
Jul 29, 2026
Portkey does not have an agent registry, which is a feature they should implement in a future release. We need to capture metadata of agents, and this capability is currently missing. End-to-end traceability is not fully fulfilled. They are only fulfilling about 50 to 60 percent of our end-to-end agent traceability requirements, and the remaining functionality needs to be implemented. The communication between MCP and AI Gateway is lacking. Additionally, pricing could be improved.
Portkey is a very good software for AI applications. It is a single place solution that you can use if you want your AI application to be in production. However, one thing I would like to say is that it could have better documentation. As of now, the documentation is very concise and short. If possible, they could provide more beginner-friendly tutorials with step-by-step instructions on how to set up and how to see the things in the dashboard. It would also be great if the documentation included sample projects or example projects so that users can easily see and learn from those. Another thing I would like to add is that it has a steep learning curve. It takes time to learn Portkey because the dashboard is very complex. There are many things that a user can see in Portkey. If they can create good documentation, it could definitely increase their value in the market.
There are definitely some places where Portkey can be improved. Overall, the experience has been positive, but the first area is analytics and reporting. While the observability feature is excellent, we would like to have richer historical analytics and customizable dashboards. For example, it would be useful to see trends by applications and teams or features over long periods without exporting data to an external BI tool. Another area is governance for large organizations. As AI adoption grows, enterprises need more granular, role-based access control. We would like to see more advanced AI evaluation capability built into the platform itself, including features such as prompt versioning and automated quality scoring and regression testing. Finally, while the platform is supported with multiple providers, we would welcome even more intelligent routing capabilities, such as automatically selecting the best model based on latency, cost, and task complexity using configurable policies. These are not major pain points, but they are enhancements that would make an already strong platform even more valuable for an organization that scales their AI workloads. Documentation and onboarding could be enhanced. Portkey is developer-friendly, but we need more end-to-end references, architecture, and implementation guides for common AI patterns. This would help teams adopt it even faster. I did not give Portkey a perfect score because while it has become a foundational component in our AI stack and solves several operational challenges, I would still like to see deeper analytics, stronger enterprise governance features, and more built-in AI evaluation capabilities, especially for prompt testing, regression analysis, and model benchmarking. I would be comfortable recommending Portkey to any organization that is building multiple AI applications or wants to manage a scalable way to handle LLM providers and produce AI traffic.
One major problem I see with Portkey is that when I had not yet started any trial, it started to denote that I had exceeded the prompt limit. Users are usually expecting a trial stage with more tokens to at least work through a few days to understand the products inside out. That might be something that could be improved. The onboarding process of Portkey is very good. However, it requests that there are no free tokens or trial tokens. That was the limitation.
Portkey offers a centralized AI gateway that simplifies architecture and reduces maintenance by providing a unified interface for model interactions, with features ensuring detailed observability and flexibility in model operations.Portkey serves as a centralized hub for AI management, allowing organizations to streamline AI operations through a singular interface that accommodates multiple language models. It ensures efficient resource management with robust observability features for...
Portkey does not have an agent registry, which is a feature they should implement in a future release. We need to capture metadata of agents, and this capability is currently missing. End-to-end traceability is not fully fulfilled. They are only fulfilling about 50 to 60 percent of our end-to-end agent traceability requirements, and the remaining functionality needs to be implemented. The communication between MCP and AI Gateway is lacking. Additionally, pricing could be improved.
Portkey is a very good software for AI applications. It is a single place solution that you can use if you want your AI application to be in production. However, one thing I would like to say is that it could have better documentation. As of now, the documentation is very concise and short. If possible, they could provide more beginner-friendly tutorials with step-by-step instructions on how to set up and how to see the things in the dashboard. It would also be great if the documentation included sample projects or example projects so that users can easily see and learn from those. Another thing I would like to add is that it has a steep learning curve. It takes time to learn Portkey because the dashboard is very complex. There are many things that a user can see in Portkey. If they can create good documentation, it could definitely increase their value in the market.
There are definitely some places where Portkey can be improved. Overall, the experience has been positive, but the first area is analytics and reporting. While the observability feature is excellent, we would like to have richer historical analytics and customizable dashboards. For example, it would be useful to see trends by applications and teams or features over long periods without exporting data to an external BI tool. Another area is governance for large organizations. As AI adoption grows, enterprises need more granular, role-based access control. We would like to see more advanced AI evaluation capability built into the platform itself, including features such as prompt versioning and automated quality scoring and regression testing. Finally, while the platform is supported with multiple providers, we would welcome even more intelligent routing capabilities, such as automatically selecting the best model based on latency, cost, and task complexity using configurable policies. These are not major pain points, but they are enhancements that would make an already strong platform even more valuable for an organization that scales their AI workloads. Documentation and onboarding could be enhanced. Portkey is developer-friendly, but we need more end-to-end references, architecture, and implementation guides for common AI patterns. This would help teams adopt it even faster. I did not give Portkey a perfect score because while it has become a foundational component in our AI stack and solves several operational challenges, I would still like to see deeper analytics, stronger enterprise governance features, and more built-in AI evaluation capabilities, especially for prompt testing, regression analysis, and model benchmarking. I would be comfortable recommending Portkey to any organization that is building multiple AI applications or wants to manage a scalable way to handle LLM providers and produce AI traffic.
One major problem I see with Portkey is that when I had not yet started any trial, it started to denote that I had exceeded the prompt limit. Users are usually expecting a trial stage with more tokens to at least work through a few days to understand the products inside out. That might be something that could be improved. The onboarding process of Portkey is very good. However, it requests that there are no free tokens or trial tokens. That was the limitation.