ucbepic/actor-gatherer is a Python framework that manages the context an AI agent needs from outside the agent itself. The latest release is v0.1.0. The provided material contains no changelog for this version, so newly added or changed features cannot be confirmed. The project separates the role of the agent that performs the work from the role of investigating and delivering context.
The Gatherer selects the information needed for the task before the model call and passes it to the agent. Agents that use the same Gatherer can share what they are working on even across different harnesses, and the material also covers selecting the parts of past conversations that relate to the current task in order to reduce duplication. It also describes a feature that notifies of changes when a file the agent trusts is modified. The README explains how to use the Python command-line tool, which attaches a runner in front of Claude Code, Codex, OpenCode, and Pi Agent. The repository language is Python and the license is MIT.
The point of this design is that instead of the agent managing all context by itself, that role is handed to a separate collector. However, the provided material contains no specific changes for v0.1.0 and no performance comparison results. So the description of the context sharing and selection features is confirmed, but how effective they really are in actual work is hard to judge from this material alone.