langchain-ai/deepagents deepagents==0.7.0a3

on GitHub

Released from alpha/deepagents-0-7-0a3 at commit 7d5e7c0

This alpha catches the 0.7 prerelease line up to current main and adds the latest SDK behavior changes on top of the recursive filesystem work from 0.7.0a2.

Features

• Allow custom middleware passed to create_deep_agent(..., middleware=[...]) to replace a default middleware instance when the .name matches, so users can override defaults such as SummarizationMiddleware without also excluding the built-in instance (#4251) (90c8472).
• Reduce round trips when sandbox.execute offloads large tool results by capturing oversized command output directly inside the sandbox artifact path (#4230) (02f5bd7).
• Lift arbitrary SDK max_iterations limit on RubricMiddleware (#4405) (d6692a7).
• Include the deepagents[aws] extra for Bedrock prompt caching support from the latest stable line (#4108) (a398382).

Bug Fixes

• Emit max_iterations_reached as the terminal RubricMiddleware status when the iteration cap is exhausted, instead of emitting a final needs_revision event that will not loop (#4406) (a51c8d2).

Internal maintainers: @imnishitha, @ccurme, @mdrxy

Released by: @mdrxy