Veai 5.7 for JetBrains IDE adds directories to chat and auto-retry for agent tasks

Veai released version 5.7 of its AI agent for JetBrains IDE and focused not on flashy demonstrations, but on everyday work scenarios. The update removes unnecessary manual actions in the chat, reduces delays between steps, and makes long tasks more resilient to random failures.

Context with entire directories

The main practical change is that you can now pass entire directories to the chat, not just individual files. Previously, developers had to manually assemble context: attach needed files one by one, describe the module structure in words, or hope the agent would correctly find neighboring parts of the project itself. In version 5.7, you can simply drag a directory into the chat window or add it through the Add to Chat action in Project View. After that, the agent gets the full path and starts working immediately with a group of related files.

On paper, this looks like a minor UX improvement, but in real development, the effect is more noticeable. If you need to analyze a subpackage, prepare a refactoring, understand module architecture, or go through configs, the completeness of context directly affects the quality of the answer. The less manual assembly before starting a task, the lower the chance that the agent will miss an important dependency, a neighboring class, or a utility file. For large codebases, this is not cosmetics, but a time savings on every cycle of communication with AI.

Models and resilience

Another important change is that model selection is now exposed directly in the chat. Previously, access to models from different providers was hidden in settings, and this slowed down work: to switch models for a specific task, you had to make unnecessary navigational jumps. Now the list is in a familiar place, where it's easier to perceive it as part of the workflow. One menu brings together models from different providers, including options with different reasoning modes, such as Think, if available for a specific model. This makes the product more flexible in everyday use.

The developer can switch faster between a cheaper model for routine fixes, a more powerful one for complex refactoring, and reasoning mode for tasks that require depth of thought. At the same time, Veai added automatic retry on agent errors. Previously, a background task could stop after the first failure and require manual restart, but now the system makes retry attempts on its own. For long autonomous scenarios, this is no longer a minor detail, but basic reliability.

• Adding directories to chat via drag-and-drop or Project View
• Switching between OpenAI and Anthropic models directly in the chat interface
• Support for different reasoning modes for suitable models
• Automatic retry attempts after agent failures
• Fewer manual actions in long work chains

Interface without noise

Veai separately worked on those areas of the interface that didn't completely break the scenario, but annoyed every day. One example is the behavior of the terminal when running commands. Previously, it could unexpectedly open and capture focus, pulling the developer out of the editor at the moment when the agent was simply executing a background command. In 5.7, the terminal no longer appears unnecessarily and doesn't draw attention to itself. The user can continue writing code or reading results while the tool does utility work in the background.

In parallel, the team reworked the feedback buttons in the chat. Instead of the previous like/dislike interface, the interface received a more understandable division between reaction to the answer and an actual error report. Sending a bug report was moved to the top, next to chat saving, so the path to reporting an issue is shorter. This is an important signal of product maturity: good AI tools win not only on the quality of the model, but also on how quickly the user can report a failure without dropping out of work context.

What it means

Veai 5.7 shows a clear trend in AI tools for development: the battle is no longer only about generation quality, but also about friction around it. Products that reduce unnecessary clicks, don't interrupt background tasks, and let you quickly change models for a scenario will win. For JetBrains IDE users, this update looks like a set of small fixes that in total make the agent noticeably more convenient in daily work.