Privacy and Local Data

Cola is a desktop Agent that can read files, run tasks, remember preferences, and call models. Data boundaries matter: what stays on your Mac, what may be sent to a model or cloud service because of a task you started, and what can appear in logs or feedback bundles.

Where Local Data Is Stored

Cola's local data is split across two default locations:

~/.cola/
~/cola/outputs/

~/.cola/ is the runtime state directory. It may contain:

• Settings: model choice, shortcuts, appearance, device preferences, and some account state.
• Sessions: conversations, task context, and tool-call records.
• Memory: long-term preferences, facts, names, and collaboration style.
• Task state: running tasks, background task state, and related metadata.
• Logs: startup, connection, model-call, error, and troubleshooting information.
• Plugin or channel state: local state for external channels or plugins, if enabled.

~/cola/outputs/ is the user-visible output directory. It stores files Cola generated, downloads, task outputs, and other results you may want to open or back up. Older outputs created before the new directory may still be under ~/.cola/outputs/; Cola tries to keep old records compatible while moving toward the new output directory.

Do not directly edit or delete the whole ~/.cola/ or ~/cola/ directory unless you have confirmed that you do not need historical sessions, memory, task state, or output files.

What May Leave Your Mac

When you ask Cola to do a task, task-relevant content may be sent to a model or cloud service, such as:

• The prompt you entered.
• File content you asked Cola to read, summarize, or modify.
• Text produced from voice recognition.
• Images, screenshots, or attachment content relevant to the task.
• Intermediate summaries, tool results, or error information needed to finish the task.
• Request information needed for account, model, quota, payment, and authorization state.

This does not mean Cola automatically uploads all local data. It means an Agent needs task context to send to models or services in order to complete the work. You can reduce exposure by limiting scope:

Read only the title and table of contents of this file. Do not read the body.

First list the files you plan to read, then wait for my confirmation.

Summarize the error types in this log without repeating full paths or user information.

Voice and Microphone

Microphone permission is used only while you actively record. Voice input is transcribed into text, then sent to Cola like a typed task.

If you do not want to use voice, do not grant microphone permission, or turn off Cola's microphone access in System Settings. You can still use Cola by typing.

What Memory Is

Cola organizes long-term useful information into memory so future collaboration is more continuous, for example:

• How you want to be addressed.
• Your preferred writing style or communication style.
• Long-term projects, common directories, or repeated workflows.
• Facts you explicitly asked Cola to remember.

Memory is not a word-for-word copy of normal chat history. It is more like long-term collaboration notes, and it can affect how Cola understands you and future tasks.

If you do not want something used long term, tell Cola directly:

Do not remember that.

Delete the long-term memory about this project.

What Logs May Contain

Logs help diagnose startup, connection, model, task, and network issues. They may include:

• Error messages and stack traces.
• Local file paths.
• Tool names, command summaries, or task snippets.
• Model, account-state, or network diagnostic information.
• Recent runtime state.

Logs are usually stored in:

~/.cola/logs/

Before sharing logs, quickly review and remove content you do not want exposed, such as private paths, account details, filenames, customer names, task text, or sensitive information visible in screenshots.

What Feedback Bundles May Include

When you submit feedback from the app, Cola may package recent logs, Network Diagnostics, environment information, and screenshots or images you attached.

Feedback bundles help diagnose problems, but they can still contain sensitive context. Before submitting, confirm:

• The issue description does not contain passwords, keys, or private content.
• Screenshots do not show windows, chats, filenames, or account details you do not want to share.
• Logs do not include business content you should remove.

Do not send the full ~/.cola/ or ~/cola/ directory to anyone. Usually, only specific logs or a feedback bundle requested by support are needed.

Product Analytics

Cola may use product analytics events to understand whether features work, where users get blocked, and whether payment flows complete.

Product analytics should not record full conversations, file contents, or screen contents. You can review and adjust this preference in Settings, under Privacy or analytics-related options.

File Access and Safety Confirmation

Cola can process files it can access and content you direct it to read. To reduce risk in sensitive directories, use instructions like:

First tell me which files you plan to read. Do not open their contents yet.

Only list the edit plan. Do not write files.

Do not delete any files. Move files only to a location I confirm.

For deletion, overwrites, commits, external messages, or sensitive content, ask Cola to explain the plan and wait for confirmation.

Sign Out and Clean Local Data

Signing out stops using the current account state, but it does not automatically delete all local history. Sessions, memory, logs, and output files may remain on your Mac.

If you are handing off a computer, using a shared computer, or fully cleaning data:

1. Sign out in Cola.
2. Export or back up output files you still need.
3. Delete sessions, tasks, and outputs you no longer need.
4. Decide whether to clean runtime state under ~/.cola/ and output files under ~/cola/outputs/.

Be careful before deleting local data. Cola may lose historical sessions, long-term memory, task state, local configuration, and generated outputs.

Usage Suggestions

• Do not send passwords, private keys, API keys, or card information directly to Cola.
• For sensitive files, ask Cola to list the plan and read scope first.
• Review logs, JSON output, and feedback bundles before sharing.
• In company or customer projects, confirm your organization's rules for AI tools and data handling first.
• When unsure, split the task and ask Cola to produce only a summary or plan first.

If you encounter startup, login, model, voice, or log issues, see Troubleshooting.