Security that fits how your business runs.
Start quickly on a dedicated server we host for your business, or keep your files and saved conversations on a small device in your own office. Here is what protects your work on each path.
How your work is protected.
Two ways to run Marling, with a clear boundary on each.
Our hosted option gives your business its own dedicated server with no hardware to set up in your office. The office device keeps your files, search index, and saved conversations on hardware in your building, with its sign-in on your own network. Both use commercial AI services whose terms exclude training models on your requests.
Reduce the identifying details sent to AI.
On the office device, with the Mental Health or Legal plan. The device checks requests for customer names, dates of birth, phone numbers, email and street addresses, Social Security numbers, and common record or matter numbers. It replaces the details it recognizes with stand-ins before the request reaches an AI model over the internet, then restores them in the reply. In our tests on deliberately awkward text, it hides about 96% of names as standard and about 99% once your customer list is loaded, though a surname used on its own can still be missed even when the full name is on your list. The substance of the request still reaches the AI model. If the masking step on the device is unavailable, the message is refused and stays on the device. Hosted Marling and the general small-business plan do not apply this masking step.
Sign-in controls who reaches your workspace.
On the office device, your team signs in through your business’s network, or through your own VPN when working remotely; we set up the device with a private network address so its sign-in page is not publicly accessible on the internet. Hosted Marling puts a sign-in gate in front of your workspace, with passkeys or emailed sign-in codes and access limited to active team members in Manage users.
Saved logins and keys are encrypted at rest.
When Saved logins and keys is enabled on a system with a security chip. An Admin adds passwords, API keys, and six-digit code setup details through the Saved logins and keys page. They are encrypted with a key sealed to the system’s security chip, and saved values are not shown to the AI model, included in chat history, or written to the activity log. This protection covers values saved through that page; passwords or keys pasted into chat are not covered.
Choose who can use Marling and what they can access.
Your Admins add and deactivate accounts, reset passwords, choose each person’s role, and can switch Dictation or Shared Memory off for the office at any time. In mental health practices, Admins also choose which assistants each User can open. Settings and Manage users require a separate sign-in from chat, and Admin changes are recorded in an activity log.
Dictation turns speech into text without keeping an audio archive.
On the Dictation page, recording starts only after you press the microphone button and confirm. Your Marling device or hosted server turns the recording into text and discards the audio after transcription; with Hosted, the audio travels to your hosted server to be transcribed. The microphone in chat uses the same transcription service and can hold audio in memory for up to half an hour. You review the transcript before sending it to Marling.
You control what Marling remembers.
Shared Memory suggests lasting preferences and ways of working from your team’s conversations, rather than individual customer details. Suggestions wait for an Admin or Marling Support to approve them before they guide future replies, and approved memories can be reviewed, changed, or forgotten at any time. On an office device with identifier removal enabled, conversations sent to an AI model for this review pass through that check first.
A daily cap stops runaway AI bills.
Your team shares a prepaid AI balance with a daily cap enforced where requests go out. When the day’s cap is reached, new requests pause until the next day rather than running up an unexpected invoice, and Marling Support can set a lower daily cap at your request.
Remote support runs over a private support connection.
Your system connects outbound to our private support network so we can maintain it and help when you ask, without opening incoming support ports on the public internet. While our team is connected for maintenance or support, we can see what is on the system, the same as an IT administrator working on a server, and our customer agreements govern how we handle that access.
Encrypted storage, with offsite backups for recovery.
On the office device; backups begin once Marling Support enrolls your device. Your files, saved conversations, and search index are stored on the device’s encrypted disk. Its security chip unlocks the disk at startup; a drive removed from the device remains encrypted. Backups are scheduled every four hours and encrypted on the device before being sent to our backup server, so we can restore your records and settings after hardware fails. Saved logins and keys remain locked to the original device’s security chip and must be entered again on a replacement.