The latest release of langchain-ai/deepagents is deepagents==0.7.29. It is a general-purpose agent harness (a ready-made execution frame with the needed components in place) for long-running tasks, bundling planning, filesystem, sub-agents, summarization, memory and approval into its default execution loop. The provided record does not contain what changed in 0.7.29. The source repository is https://github.com/langchain-ai/deepagents.
The harness is built on LangGraph and provides a filesystem, sub-agents (lower-level agents), context management, persistent memory and Skills. The README describes it as an opinionated (preferring fixed defaults over user configuration) open source harness that runs with the default setup alone, and explains that components can be extended or replaced without forking. It is stated to work with any LLM that supports tool calling. The language is Python, the license is MIT, and the GitHub star count is 30473.
Where the evidence stops is also clear. The change list and performance measurements for 0.7.29 are not in the provided record, and of the ten-point rating items only momentum scores 6.2, while novelty, tech and quality are empty. Real performance cannot be inferred from the feature list alone. Still, the MIT license together with a structure that allows components to be replaced without forking reads as a combination worth reviewing for teams that build on LangGraph.