Voice-Controlled Computer Use Is Now Part of DictatorFlow
DictatorFlow started as fast speech-to-text. The next step is using the same voice loop to control the computer: inspect files, edit project workspaces, search folders, run bounded commands, and hand coding tasks to agents without leaving chat.
- > Chat can now use local workspace tools when explicitly enabled.
- > File reads, writes, search, and exact replacements stay inside a configured root.
- > Command execution and Codex delegation are separate opt-in switches.
- > The chat UI shows which tools were used so the session remains inspectable.
From text entry to action
Dictation is useful because it removes friction from writing. Computer use removes the next layer of friction: the gap between describing work and doing it. A developer should be able to say what needs changing, then let the assistant inspect the repo, make the edit, and report what happened.
The first implementation is deliberately practical. DictatorFlow does not need broad desktop automation for every workflow on day one. It needs a clear tool boundary, a workspace root, and enough primitives to make real changes in a project.
Designed for local control
The local tool layer is off by default. When enabled, it is scoped to a root directory and rejects path traversal or symlink escapes. That lets chat act on the files you choose without turning every prompt into unrestricted access to the machine.
This is the direction we want: voice as a control surface for useful computer work, with explicit configuration for the parts that can modify state.