EVO CAPABILITIES / Local and cloud AI cooperation
Choose where the work happens.
Local and cloud AI serve different needs. EVO is developing a shared workspace where your task, available device and data choices guide the processing route.
Mac development · See current scopeThe real problem
Not every task needs the same model or the same data boundary. A private note may be suitable for local work, while a more demanding request may benefit from an available cloud service. A larger model can also be a poor experience on a machine that cannot run it comfortably. The right choice depends on the task, your hardware and what you are willing to share.
What chat history leaves you to do
A single chat box can hide important differences between running an application on your Mac and running its AI on your Mac. It can also make a change of model look like a cosmetic preference when it changes where data is processed. Users need an intelligible choice, and a failure should not silently make that choice for them.
How EVO approaches it
Match the route to the actual task
EVO's development workflow distinguishes local model availability from cloud service readiness. Evo Swift and Evo Sage are local model directions with different resource and capability trade-offs. A local answer, an image review and a tool-supported task do not automatically have the same requirements; the chosen route must support the kind of work requested.
Keep the workspace consistent
The goal, selected material and saved results stay attached to the Work even when you choose a different supported route for a later request. This gives a project continuity without implying that all previous content is sent again. Changing a route should keep the task understandable while respecting the new request's selected data and permissions.
Make cloud use a real choice
Selecting local work does not silently grant cloud access. In supported selected-file flows, a cloud request requires a separate sharing choice for that request. If the local model cannot perform the requested kind of analysis, the useful response is to identify the limitation or offer an explicit alternative, rather than pretend it used a capability it does not have.
Evaluate usefulness as well as model size
A small model can be useful when a task is clear and supported tools provide reliable intermediate results. That does not make it equal to a stronger model for every problem. EVO's development evaluations keep latency, incomplete responses and tool outcomes visible, so a larger model is not automatically presented as the better experience on every Mac.
A concrete example
ILLUSTRATIVE WORKFLOW · NOT A CUSTOMER RESULT
Illustrative workflow: start with a private local task
You select a local model to discuss a chosen text file. Later, you decide that a more demanding follow-up is worth considering through an available cloud route.
Start with the selected local material and check whether the local model is ready for the task.
Review the answer and any limitations before deciding whether another route is useful.
If you choose cloud processing, review the exact material to include and the new request's sharing choice.
What you take away
The intended outcome is a deliberate trade-off between privacy, capability and performance, with a continuous project record. A local difficulty becomes a clear decision point rather than an invisible transfer of the file.
Your choices
- Choose a supported route and inspect whether the required model or service is actually ready.
- Decline cloud sharing without granting it through a saved preference, a model error or a request to continue.
- Keep local files separate from selected cloud context and review new requests before sending.
Current scope
Implemented foundation
- Swift has run natively on Apple Silicon in development tests. Sage remains Preview; cloud work requires a connection and an eligible account.
- Local answers still have factual and completeness failures. Continuous local/cloud handoff has not passed a complete current-version journey.
Next milestones
- Public model downloads, final supported-device requirements and customer-ready cloud service remain separate release work.
- Local AI does not make web access, account features, every tool or every document workflow available offline. Model speed and quality depend on hardware, model configuration and the task; choose against published requirements when a release becomes available.
These are development capabilities, not a public release. See platform availability before requesting a trial.
Platforms and availabilityRelated questions
Can I keep AI work on my Mac?
Selected local requests can run with a compatible installed model. This does not make search, account services or every feature offline.
Swift has real Apple Silicon tests, with unresolved accuracy and incomplete-answer failures. Sage remains Preview.
Final hardware requirements and public model downloads are pending. Proposed plans are on the pricing page; checkout is closed.
How should I compare the proposed plans?
Compare the proposed Free, Core, Pro · 5x and Studio · 20x plans by working capacity and usage, rather than assuming a higher plan makes every answer better.
The multipliers describe relative cloud allowance against Core. Local usage and the number of simultaneously active Works are separate plan dimensions.
Someone who runs frequent research may need a different allowance from someone mainly using local text tools.
Review the current plan comparison and future purchase terms before choosing; existing EVO memberships are not silently converted.
Pricing is proposed and checkout is closed. Final allowances, billing periods, taxes and renewal terms must be confirmed before purchase becomes available.
How do local and cloud AI fit into one workspace?
They are separate processing choices. Cloud use sends selected request content to the service and needs a connection; local use needs an installed compatible model.
Their capabilities and answer quality differ. Joint continuity after disconnection is still being qualified.
Will a weak local model quietly switch my task to the cloud?
No. Choosing local processing does not authorize a hidden cloud fallback when the local route is unavailable or insufficient.
The current workflow keeps the selected processing route and its data permission separate. A different route needs a new explicit choice.
If a local model cannot review images, the honest result is a capability limit or supported text-only work, not an unannounced screenshot upload.
Stay local, narrow the task or choose an available cloud route after reviewing the sharing scope.
Local selection does not make unrelated network tools offline, and it cannot guarantee that a local model can complete every requested task.
What can I do when I am offline?
An installed compatible local model can answer selected local text. Search, cloud models and new downloads need a connection.
Local responses can be incomplete or inaccurate; reconnecting does not guarantee automatic cloud handoff.
The App and public model downloads are not yet released.
Try the workflow
Bring a task you want to move forward.
Explore the Mac release, or tell us which workflow you would like to evaluate. An application does not guarantee an invitation.