Skip to main content
Imagine that you ask a coding agent on a remote server to deploy your application. The agent needs your Cloudflare API token, and after deployment it creates a database password that you need to save. Pasting either value into the agent conversation would leave plaintext in its transcript. With Wipe.me, you send the API token as a one-time private link. On the remote server, Wipe.me places it directly into the deployment environment without returning it to the model. When the server creates the database password, Wipe.me encrypts it there and the agent sends a private link back to you. The tools and people receive the values they need; the agent conversation carries only the temporary links. Wipe.me turns that transfer into a private, encrypted, one-time handoff. Encryption and decryption happen locally, and the service stores only an opaque encrypted envelope.
The arrows below describe who initiates and coordinates the workflow—not a plaintext path through an AI model. When an agent participates, it should handle only protected link references, file paths, and commands. Plaintext is delivered by the local CLI to a human, a protected file, an environment variable, or a trusted tool.

Human → Human

Share a secret or private file without placing its plaintext in chat, email, or a ticket.

Human → Remote agent

Send a credential from your notebook to a tool running in a remote agent environment.

Remote agent → Human

Receive a database password, recovery artifact, or private configuration created on a remote machine.

Agent → Agent

Pass a kubeconfig, generated database password, or private artifact between agents on different machines.

​
Choose the narrowest exposure

Prefer a workflow where the fewest components ever receive plaintext:
  1. A private link is better than raw plaintext in a durable conversation.
  2. A protected link file is better than a link in command history or agent output.
  3. Direct environment injection into a trusted tool is better than exposing the secret to an agent and asking it to relay the value.
In agent-coordinated workflows, do not use plaintext stdout as the handoff. Run Wipe.me beside the sensitive value: use structured MCP tools on compatible agent hosts, or use CLI file and process workflows on remote shells and runners.
An Auto-mode private link is a complete bearer capability. Anyone who obtains it may be able to open or delete the message. A Custom-mode link also requires its separately shared passphrase. Keep links in protected files or similarly narrow channels where possible. Wipe.me reduces incidental retention; it does not make an untrusted human, model provider, plugin, or child process trustworthy.

​
Availability

Browser-based creation and opening are available today. The published v0.3.0-alpha.3 CLI preview supports creation, deletion, local retrieval to protected files, direct exec injection, compact links, manual passphrases, attachments, generated-password workflows, and a local agent-safe MCP server. These allow an agent to coordinate an operation without receiving the plaintext in model context.

Command-line interface

Review exact availability, installation options, protected link sources, and CLI security behavior.

MCP server

Coordinate generated secrets, private files, environment injection, and trusted commands through MCP.