Acti and OpenCues share a thesis: AI belongs where you type, not in a chat window. Acti delivers it as an agentic keyboard for iOS and Android. OpenCues delivers it as an open-source inline layer for the desktop: Claude Code, OpenCode, Gemini CLI, Chrome and your shell.
What Acti does well
Acti replaces your phone keyboard. You type an intent in any text field, hold the Acti Bar, and its agents act, with Skill Keys mapping single keys to multistep routines and a large catalogue of app and API connections. Free at launch, with a no-code Skill Builder and a community skill hub.
Where Acti struggles
Acti is a replacement keyboard, and everything follows from that. On iOS it needs Full Access, so every keystroke on your phone flows through a third-party keyboard talking to a cloud model, a large trust surface. Its agents act through OAuth grants into your accounts, which is powerful and correspondingly broad. Inference is Gemini-only in the cloud: no local mode, no key of your own, and nothing works offline.
How OpenCues differs
OpenCues keeps your existing input method and uses a typed _ instead of a special bar. It is Apache-2.0 open source where Acti is proprietary, and model agnostic where Acti is Gemini-only: ten providers, your existing Claude or ChatGPT subscription, or a fully local model via Ollama. Design stance differs too: OpenCues LLM output only ever lands as draft text you review, never wired directly to actions.
The honest split: Acti is mobile and OpenCues is not, there is no iOS or Android host today. In return, OpenCues edits land revertable per word, and each feature can route to a different model.
Both products are extensible, in different ways: Acti through its no-code Skill Builder and skill hub, OpenCues through cues and blanks you write as plain markdown and shell scripts, under an open standard anyone can implement. See write your own blank.
Related: Full comparison · AI in any text field · Acti