Improved Latency: From Minutes to Seconds

One of the most noticeable changes in this release is a huge improvement in latency of context curation and query operations: from minutes to seconds.

Previously, both query and curate relied heavily on sub-agent execution. While this approach was efficient in terms of reasoning and relevance, its overuse introduced significant end-to-end latency. In some cases, operations could take up to a few minutes to complete.

In this update, we reworked the execution path and removed unnecessary sub-agent hops in the critical path. As a result, operations that previously took minutes now complete in seconds.

When curation and retrieval are fast, they become a natural part of the workflow rather than something developers hesitate to run.

Enhanced Stability with the “Bring Your Own Key” Option

ByteRover continues to support Bring Your Own Key for LLM usage, and this release improves stability when running with user-provided models. In practice, using your own provider often means access to more isolated resources and a more mature infrastructure, which helps reduce variability and improves overall reliability.

We’ve observed a clear pattern from users: many prefer self-hosted or externally managed models. Some want tighter cost control, while others want greater flexibility in model choice. That’s why ByteRover doesn’t aim to lock users into a single model provider or pricing model.

Our goal is to maintain model-agnostic and focus ByteRover on what it does best: memory, context, and workflow reliability.

Reference Folder in Curation

With this release, ByteRover now supports Reference Folder in Curation. Users can curate an entire folder directly using:

This makes it easier to capture structured knowledge from directories that already reflect logical groupings, such as modules, components, or documentation folders, without curating files one by one.

Better File Format Support

This update also improves how ByteRover handles common document formats, including PDF, DOCX, PPT, and others.

File parsing and normalization have been refined so that agents can extract and reason over content from these formats more reliably. This matters in real projects, where important context often lives outside plain text files.

Why These Upgrades Matter

As agents become more deeply embedded in development workflows, infrastructure quality matters more than feature count.

Low latency determines whether memory feels usable. Stable model integration determines whether teams trust the system in production. Comprehensive curation sources determine whether knowledge actually gets captured.

This release is about strengthening those fundamentals.

Try It Now

We hope you enjoy building with what we’ve shipped and would really love to hear your feedback once you try the updated CLI.

We have much more planned on our product roadmap, and we’re excited to share what’s coming in the weeks ahead.

If you’re already using ByteRover CLI, you’ll see a prompt in your terminal to upgrade to the latest version. If you’re new to ByteRover, you can get started at app.byterover.dev.