Abstract
Conventional management of session context in conversational artificial intelligence, such as automatic history accumulation, can be opaque and may contribute to token limit overruns. A system for human-in-the-loop, artifact-based context management can reframe a session into a user-controlled workspace. This method can replace implicit chronological history with a curated collection of discrete data objects, or artifacts. An agent's default access to session history may be disabled, prompting it to use a curated context. A user can select relevant artifacts through an interface, and the agent may retrieve their content on-demand using a retrieval-augmented generation pattern. This approach can provide granular and auditable control over the agent's context and enable workflows that can be less straightforward with some conventional techniques.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Kochera, Michael; Cavalli, Cristian; Lee, John; and Khidhir, Besam, "Human-in-the-Loop Context Management for AI Agents via a Curated Artifact Workspace", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/9784