Implement a Virtual File System (VFS) that provides each agent execution with an isolated, persistent file system, bound to a context ID (e.g., run or conversation). This will allow file changes made by agents or tools during execution to be tracked and persisted across runs.
Currently, the agent environment is stateless — any files written to disk during an execution are lost once the process ends. This limits functionality for agents and tools that rely on file I/O for intermediate state management, project generation, or incremental development.
By introducing a VFS that binds to a given context (run, conversation, or project), we can persist file changes and allow agents to seamlessly resume work where they left off. This will enable more advanced use-cases like multi-step code generation, debugging, or document editing workflows.
Note: we already implemented something similar in https://github.com/i-am-bee/bee-code-interpreter
Related: #1974
Implement a Virtual File System (VFS) that provides each agent execution with an isolated, persistent file system, bound to a context ID (e.g., run or conversation). This will allow file changes made by agents or tools during execution to be tracked and persisted across runs.
Currently, the agent environment is stateless — any files written to disk during an execution are lost once the process ends. This limits functionality for agents and tools that rely on file I/O for intermediate state management, project generation, or incremental development.
By introducing a VFS that binds to a given context (run, conversation, or project), we can persist file changes and allow agents to seamlessly resume work where they left off. This will enable more advanced use-cases like multi-step code generation, debugging, or document editing workflows.
Note: we already implemented something similar in https://github.com/i-am-bee/bee-code-interpreter
Related: #1974