1. From input to execution
LabMind sends your instructions and necessary context through the selected AI processing route. The model replies or proposes tool operations, which run on your selected computer or an available cloud executor. The mobile app shows results, questions and approvals. Model processing and command execution locations are separate choices.
- Input: text, recognized speech, selected photos and files, and related instructions.
- Context: necessary message history, enabled memory and file excerpts read for the task.
- Execution feedback: tool calls, command output, errors and task artifacts may become further model context.
- Controls: model, execution location, approval mode and conversation permissions define available processing scope.
2. Model providers and processing locations
Available models are processed by their respective providers. Providers and locations vary by model and service route. Review the recipients, information involved and conditions in model selection and feature notices. Choosing “local” or your own computer does not automatically make the model offline; task information may still go to third-party AI or associated routing services.
Processing abroad may involve international transfers of personal information and requires applicable notices, safeguards and authorization. Before accepting a route, confirm that you and your institution permit the content to be sent through it. If you decline, use an available alternative or stop the feature. Contact privacy@sagaway.ai about information or consent.
Providers have different retention, training and improvement conditions. We do not make a blanket promise that every third party never retains information or never trains on it. Check specific conditions before using sensitive, restricted or confidential files.
3. Memory, private values and sync
Memory supplies enabled longer-term or work context to AI. Do not put passwords, verification codes or keys in memory; memory is not a confidential vault.
Private Values manages actual values separately from ordinary messages. Models can use names, purposes and placeholders, while authorized execution devices use actual values when running commands. Grant only individual values needed by a conversation. A cloud executor can use values granted to that conversation; authorization does not mean the cloud can never see them.
Cloud sync affects cross-device storage and access to conversations, memory and private values. Cloud private values are stored encrypted but can be decrypted by the server in authorized workflows; this is not end-to-end encryption. Sync is separate from model-sharing authorization. Turning it off does not delete all existing cloud data; follow the feature’s instructions for existing data.
4. File and tool permissions
Executors may read, write or delete working-directory content, run programs and access networks. Files, web pages and third-party tool results may contain malicious instructions. Check the execution location, directory and approval mode, reject unclear commands, back up important files and avoid unnecessary permissions.
Approvals and access controls reduce risk but do not guarantee safe execution. Completed file changes, requests and resource consumption may not be reversed by stopping a task. Malicious commands can transmit authorized private values externally. Revoke permissions when they are no longer needed.
5. Verify and report
AI can be wrong. Independently check references, calculations, code and important conclusions. Do not use unverified output directly for medical, legal, financial, safety or other consequential decisions. Follow research ethics and academic integrity requirements.
Report inappropriate content, task failures or billing issues to business@sagaway.ai with the task identifier, time and necessary details. Redact screenshots and logs; do not send passwords, verification codes, tokens or keys. Contact privacy@sagaway.ai about information, consent withdrawal or deletion.