deepagents==0.6.9

langchain-ai/deepagentsdeepagents==0.6.9Jun 12, 2026by github-actions[bot]

AI Summary

This release focuses on subagent configuration, performance optimizations for token counting and caching, and various bug fixes regarding error handling and state propagation.

Key Highlights

Configurable subagent response format
Performance improvements via token counting and caching
Fixed handling of Overwrite-wrapped messages

New Features

Support configurable subagent response format
Count tokens once per model call in summarization middleware
Cache filesystem system prompts
Cache grep glob matchers
Stream Python grep fallback

Full Release Notes

### Features

* Support configurable subagent response format ([#3882](https://github.com/langchain-ai/deepagents/issues/3882)) ([b0e4d7a](https://github.com/langchain-ai/deepagents/commit/b0e4d7aa8d0ad58c65bddef3835e539b3d5b2a99))

### Bug Fixes

* Don't swallow `TypeError` from custom `token_counter` ([#3927](https://github.com/langchain-ai/deepagents/issues/3927)) ([a6ec9d0](https://github.com/langchain-ai/deepagents/commit/a6ec9d0e797a8b8a7d53d7debb197bb25e80be41))
* Handle `Overwrite`-wrapped messages in tool result interception ([#3905](https://github.com/langchain-ai/deepagents/issues/3905)) ([a043c5a](https://github.com/langchain-ai/deepagents/commit/a043c5a6717374c405a3ed740cfdafdd196f5f21))
* Keep private state out of subagent propagation ([#3542](https://github.com/langchain-ai/deepagents/issues/3542)) ([7ff9553](https://github.com/langchain-ai/deepagents/commit/7ff9553fc057a682ba503ebfe9a870adb51ab848))
* Make sync glob timeout bound wall-clock time ([#3866](https://github.com/langchain-ai/deepagents/issues/3866)) ([cba6caf](https://github.com/langchain-ai/deepagents/commit/cba6caf8f708133381506ecc9e217e7e3ca1c7f2))
* Normalize read slices after windowing ([#3888](https://github.com/langchain-ai/deepagents/issues/3888)) ([33d900c](https://github.com/langchain-ai/deepagents/commit/33d900c98b686aa7e7782d0be3019010ced5f03a))
* Stream Python grep fallback ([#3886](https://github.com/langchain-ai/deepagents/issues/3886)) ([3673d95](https://github.com/langchain-ai/deepagents/commit/3673d95d41aa00f03fbfe2083e46afa485800f3a))

### Performance Improvements

* Count tokens once per model call in summarization middleware ([#3877](https://github.com/langchain-ai/deepagents/issues/3877)) ([6558c8c](https://github.com/langchain-ai/deepagents/commit/6558c8ca3d8e10e3309f95c807e740914af3175f))
* Cache filesystem system prompts ([#3889](https://github.com/langchain-ai/deepagents/issues/3889)) ([2f432ba](https://github.com/langchain-ai/deepagents/commit/2f432ba63687d7c8488d5d41d0551d02b4ae00b4))
* Cache grep glob matchers ([#3887](https://github.com/langchain-ai/deepagents/issues/3887)) ([eae3cf1](https://github.com/langchain-ai/deepagents/commit/eae3cf148de3f8c8ee9ffd8af6e5a5e06cce9d6f))

---

Thanks to our community contributors: @phamgialinhlx

Internal maintainers: @nick-hollon-lc, @hntrl, @mdrxy