Browser & OS Action

Chooses an action and its matching DOM element in one request; a text model generates input text.

↳ Separates UI choices, text generation, and browser execution so each step can be inspected.

Jev selects a target and action and estimates presence and risk; local policy decides whether to execute.

↳ Keeps the full UI tree out of the main Agent context while exposing the selected action.

Chooses DOM actions and judges goal completion or stalled progress; a separate model can supply typed text.

↳ Records proposed actions, execution and stop reasons for inspection.

Selects actions from current page controls and assesses risk and irreversibility.

↳ Combines automated page actions with user confirmation in one workflow.

Jev chooses actions, matches form fields or extracts structured content from page observations; Playwright executes.

↳ Supports persistent sessions and UI readback, which does not by itself prove database persistence.

The Agent supplies action candidates; Jev picks an ID, then the Agent executes and verifies it through Aside.

↳ Keeps choices within an application-owned action table while requiring separate outcome verification.

Uses window state to select an action, target, text-input need, and completion status.

↳ Grounds candidate controls in UI snapshots so choices can be inspected.

Chooses apps, controls and next actions from phone state; Mobilerun executes them.

↳ Records action traces and request latency for inspection.

Jev selects controls or groups from locally observed candidates; coordinates and input execution stay on the Mac.

↳ Makes candidate controls and execution checks inspectable, including Finder workflows.

Chooses controls and checks completion or blockage before the test runner operates the UI.

↳ Adds choices grounded in current UI state to mobile testing.

Selects only current snapshot candidates and supported actions; another model can supply typed text.

↳ Executes against current snapshot references and records step state.

Chooses intent, elements, URLs or verbatim spans and judges command completeness and sensitive actions.

↳ Shows probabilities, actions and request timing during voice interaction.

Batches candidate DOM elements for TypeSafe and uses Noul probabilities with thresholds to decide which to collapse.

↳ Adds text-based semantic judgments alongside local ad rules.

Reads DOM or supported visual-region descriptions and returns a supplied candidate action ID.

↳ Provides Python and TypeScript loops with separate offline and live checks.

Jev selects operations from a DOM target table; a separate text model supplies typed text when needed.

↳ Captures before-and-after screenshots and yields on sensitive actions; completion still needs readback verification.

Selects the next step from deterministically extracted controls and actions before desktop execution.

↳ Separates screen reading, action selection, and text generation.

Chooses targets and actions and assesses completion and risk; local policy controls execution or confirmation.

↳ Uses text candidates and starts with a dry run by default.

Batches element text, labels, and link information into Noul questions and applies a threshold.

↳ Shows how semantic judgments can be connected to concrete page elements.

Sends page state and permitted candidates to Jev; the existing browser connection executes the actions.

↳ Separates repetitive page choices into decision steps while reusing the browser connection.

Chooses elements, actions and values, and assesses completion, errors and irreversible steps.

↳ Separates browser-action loops from caller planning and returns status and traces.

Jev selects an operation and target in one request; a separate helper handles free-form text when needed.

↳ Offers observation, suggestions and execution, while uploads and dialogs still require other browser tools.

Jev classifies posts with Choice, rates signal, actionability and originality with Score, and assigns Noul labels; local thresholds and noise rules decide whether to fold each post.

↳ Adds scores, labels and adjustable filtering thresholds to the X feed, with reversible folding of posts and their attached media.

Chooses CLICK, TYPE_TEXT, SELECT, SCROLL_DOWN, SCROLL_UP, PRESS_ENTER, WAIT, DONE, or BLOCKED and the matching DOM target in one request; PRESS_ENTER is a separate key control. Separate yes/no checks cover goal completion and a stuck loop.

↳ Runs in the user’s own tabs without screenshots, and keeps the action, target, and typed string inspectable.

One System One request asks Nouls for worth-engaging plus optional tags such as spam, buzz, and AI-ish.

↳ Turns timeline filtering into inspectable probabilities instead of generated commentary. Post text is sent to TypeSafe.

POSTs to /v1/systemone as jev-latest: a goal Noul and a Choice over registered Flutter actions.

↳ Turns natural-language tests into choices over a closed action list instead of free-form tapping.

POSTs https://api.typesafe.ai/v1/systemone as jev-latest with a closed action Choice each step.

↳ Screen actions follow Jev's closed choices. Author demo numbers were not retested here. GitHub SPDX is empty; the LICENSE file is MIT.