Keep the decisions that should survive the chat.
Requirements, preferences, outcomes, and corrections retain their source—a file, a meeting, a browser session, or a direct instruction—and stay within the relevant project.
MokenLogic Harness / macOS
Preview available · Product direction shownChange the model. Keep your context.
MokenLogic Harness carries project decisions, preferences, and unfinished work into later sessions. It connects that context with apps, browser sessions, files, and tools, while keeping answers and actions open to inspection.
PlatformmacOS
ArchitectureModel-flexible
StatePreview
Personal continuity / 01
Say it once. Bring it back when it matters.
Episodic Memory is off by default. When enabled, project decisions, preferences, constraints, and their sources can inform later tasks. Review remembered information at any time and update it when your requirements change.
Use earlier context to answer a new question.
A later task can connect what MokenLogic remembers about your situation with current, checked information and show the basis for the conclusion.
Move from the answer to the next authorized step.
The same context can guide a file, browser check, scheduled task, or app action. The result stays reviewable and can inform later work after verification.
System loop / 02
The work continues beyond one model response.
A harness coordinates the surrounding system: what should be remembered, which tools may act, how progress survives, and what evidence supports the final result.
Intent
Define the goal, boundaries, and expected artifact.
Context
Recover only the project state relevant to this decision.
Plan
Turn relations and constraints into inspectable work.
Execute
Work through authorized apps, browser sessions, files, and tools.
Verify
Check the artifact, sources, actions, and unresolved questions.
Persist
Store evidence-backed state so the work can continue later.
In the app / 03
What the work looks like, not another chat column.
Each window is a different job: a research digest, a calendar hold, a live browser pass, a Koe call, meeting notes, schedules, and files. With Episodic Memory on, they share the same memory.
01 · Agent loop
A turn is a system, not a reply.
Intent, recovered context, tool actions, and the artifact stay in one inspectable loop. If the work pauses, the next step still has the evidence.
02 · Computer use
The desktop is a tool, not a screenshot.
Authorized native apps can be driven directly — Calendar, Finder, Mail. Consequential actions stay reviewable, and unattended control requires an explicit grant.
03 · Playwright
A real browser session, then a real file.
Playwright works in a live Chrome window: open pages, extract tables, fill forms, and write the result to disk instead of pasting a summary into chat.
04 · Koe
Speak the task. The same agent runs it.
Koe is the voice model for Harness: a call, not dictation. You talk; the agent uses the same tools, files, and computer access as a typed task, and talks back when there is something to report.
05 · Meeting notes
A meeting becomes a notes file.
Recording and transcription are aimed at a deliverable: decisions, action items, open questions, and a document on disk that the next session can recover.
06 · Schedules
Recurring work is a system job.
Timed and repeating routines are first-class. A digest, a monitor, or a weekly brief can run without re-prompting, and only ask for attention at an exception.
07 · Artifacts
The output is a file, not a reply.
Briefs, decks, and tables land in the artifact library as real documents. Skills package the method; the folder keeps the result.
Reviewable work / 04
Know what carried forward and what happened next.
MokenLogic is designed to keep the important parts visible: where relevant context came from, what action ran, what changed, and what still needs confirmation.
Remembered context
See which earlier decision, preference, or correction shaped the result.
Completed work
Review tool actions and the artifacts they produced.
Your control
Review consequential steps before they take effect.
Model flexibility
Changing a supported model should not erase your context.
Supported models, exact permissions, data handling, and pricing are documented with the preview and will continue to be updated.
MokenLogic Harness / macOS
The native product surface for AI that does not start over.
macOS preview is available to download. Capabilities will continue to expand.