Flick
Local MCP computer-use agent that lets Jev choose browser and macOS actions from observed controls.
Added to Jevfast
How it uses Jev
Flick sends Jev a compact state of the user goal, supplied inputs, success conditions, optional task plan and artifacts, current interface observation, available apps, memory, and recent action effects. Jev chooses a typed operation from the currently available candidates; for most operations it also chooses an observed target. Filling requires separate target and supplied-value choices, so Jev selects text supplied or remembered by the user rather than inventing field contents.
What Jev decides
Choose the next available computer-use operationExample answers · not a recorded Jev response · Source ↗
Question 1 · operation
YOUR APP
INSTRUCTION
Choose among action groups available for this observation; the set changes with the visible controls and candidate actions.
STATE
The state includes goal, supplied_inputs, success_conditions, optional task_state {active_objective, plan, artifacts}, optional app_guides, now {iso, local, timezone}, interface {title, url, target, OCR status, optional tabs/loading/scroll/open_layer, text, observed elements, more_controls_not_listed}, optional clipboard, available_apps, optional memory, and recent_actions_and_effects (last 8). A separate completion verifier later asks a Noul question; this chart is only the action-selection request.
JEV · CHOICE
- click
- right_click
- double_click
- fill
- compose
- draft
- copy_text
- switch_tab
- fill_submit
- select
- press
- scroll
- wait
- scroll_top
- scroll_bottom
- wait_for_load
- wait_for_change
- wait_for_images
- scan_screen
- inspect_more
- request_vision
- complete_milestone
- modify_plan
- back
- forward
- refresh
- switch
- remember
- emergency_stop
- done
- blocked
App workflow
- Describe the task and constraints
The host gives Flick the goal, supplied input values, and completion conditions. Flick observes the foreground interface and maintains task state, milestones, and retained artifacts.
- Select and execute an action
The TypeSafe decider offers only actions available from the current observation. Jev chooses the operation and a matching target; fill decisions also pair an observed field with an exact supplied or remembered value. Flick executes the choice and observes the result.
- Repeat and verify
The controller repeats while the task progresses. Jev separately judges evidence for milestone completion and final coverage of the user goal; the host returns the task status and collected evidence.
Why it is interesting
The integration factors computer use into auditable decisions over live, observed choices. It couples operation selection to a matching target question, keeps fill values constrained to user-provided or remembered text, and uses separate Jev judgments to assess milestone evidence and whole-goal completion.