What you send, to where you send it.
An API client knows the credentials for your systems. So this page says where they live, who sees them, what an AI agent may do and where RequestTailor opens connections – each with a source in the core's code.
The repo holds the name. The value stays on your device.
Secret values never appear in collection files.1 They enter the request only when sending and appear in the report as ***.
| What | Mac app and rt | Web app |
|---|---|---|
| Secret values, OAuth tokens | Operating system keychain | Encrypted in the browser's storage |
| Values from vaults | Through the provider's command line, in memory only | Through rt agent --allow-secrets, in memory only |
| AI provider key | Keychain | Encrypted in the browser |
| History | Name, method, masked URL, status, time – no bodies, no header values | |
Supported vaults: 1Password, HashiCorp Vault, AWS Secrets Manager, Azure Key Vault, Google Secret Manager. Of an OAuth token, the interface sees only information about the token, never its value.
The agent may read. Writing to production only with you.
The MCP server rt mcp2 gives agents your collections as tools: list_requests, get_request, list_environments, send_request, run_folder.
- Per environment: none, read, write. The default is read. An environment can be marked as production.
- Writing requests against production wait. You approve them in the app banner or with
rt approveand refuse withrt deny;rt approvalslists what is waiting. - Agents never see secret values. Every result, every error and every audit line passes a final mask that replaces every known secret value with
***. - The rules apply from the start. An agent that can edit files cannot give itself rights that way: later changes apply only where they are stricter.
- Local audit log. Every agent action is written to a file in the data folder on your machine.
No connection to us. Only the ones you cause.
No account, no licence key for Free, no telemetry, no crash reports, no update check.3 Telemetry would only ever come with your consent – Free has none.
- The requests you send.To the hosts in your collection, with your credentials.
- Git remotes, identity providers, vaults.When you pull or push, sign in with OAuth or have set up a vault.
- The AI provider you set up.Off until you configure it; all secret values are removed before every call.4 With your own key straight to your provider, also to a local model.
- Hosts a plugin may reach.You see them before you install the plugin.
- The companion
rt agent.Listens on 127.0.0.1 only, checks Host and Origin, pairs through a single-use code, keeps tokens in memory only.5 It logs method and host, never paths, headers or bodies.
Scripts and plugins see no files.
Neither has file or network access except through the request API.
- ScriptsQuickJS as WebAssembly, with limits for computation and 64 MiB of memory.6 Scripts only run in collections you trust on this device; otherwise the request is sent without scripts, with a note. Scripts see secret values; their output never appears unmasked.
- PluginsWebAssembly only, no native code; a fresh instance per call.7 Permissions one by one: network to named hosts, own storage, variables. Plugins never see secret values.
- HonestlyRequests that a script in a trusted collection sends itself (
rt.sendRequest) are not checked against the MCP server's rules. Do not trust a collection whose scripts you do not know.
No commit with plain-text credentials.
A commit in the app that contains plain-text credentials fails.8 For pipelines and Git hooks there is rt scan: exit code 0 when clean, 1 with findings.
- Before every commit. The app shows what it found and does not commit.
rt scanin pre-commit. The command line checks the changed YAML and JSON files.- Line by line only. With
rt-scan:alloworrt-scan:allow-next-line– visible in the diff.
Confidentially, to a person.
If you find a vulnerability in RequestTailor, please write to office@clouddna.at with the subject “Security”, not in a public issue. Describe how to reproduce it; do not send real credentials. The core's code is licensed under Apache 2.0 and becomes public with the first release.
This website sets no cookies, loads nothing from other servers and has no tracking. Details in the privacy policy.
Sources on this page
The superscript numbers in the text point here. Tap one to see the source right where it is used; all sources of this website are listed on one page.
- 1
Secrets in the keychain and from external vaults ↑Product · RequestTailor core repository, CLAUDE.md (Produktprinzip 3), crates/rt-cli/README.md (externe Secrets), website/docs/environments · checked on 2026-11-22
- 2
MCP server rt mcp with rules per environment ↑Product · RequestTailor core repository, crates/rt-cli/src/mcp, crates/rt-cli/README.md (MCP server), crates/rt-cli/tests/gate_w8.rs · checked on 2026-11-22
- 3
No account, offline, no connection to CloudDNA ↑Product · RequestTailor core repository, CLAUDE.md (Produktprinzipien 2 und 7), website/docs/privacy · checked on 2026-11-22
- 4
AI assistant, off until configured ↑Product · RequestTailor core repository, crates/rt-app (ai), website/docs/privacy · checked on 2026-11-22
- 5
The rt agent companion for the web app ↑Product · RequestTailor core repository, crates/rt-cli/src/agent, crates/rt-cli/README.md (Sicherheitsmodell) · checked on 2026-11-22
- 6
Script sandbox: QuickJS in WebAssembly ↑Product · RequestTailor core repository, crates/rt-script (api.d.ts, README) · checked on 2026-11-22
- 7
Plugins only as WebAssembly, with permissions ↑Product · RequestTailor core repository, crates/rt-plugin/API.md, plugins/README.md · checked on 2026-11-22
- 8
Secret scanner before every commit ↑Product · RequestTailor core repository, crates/rt-app (secret_scan), crates/rt-cli/README.md (rt scan) · checked on 2026-11-22
Questions about security.
- Does RequestTailor send data to CloudDNA?
- No. Free has no account, no telemetry, no crash reports and does not check for updates. Connections only come from you: the requests you send, Git remotes, your identity provider, the command lines of your secret managers, the AI provider you set up, and the hosts a plugin is allowed to reach.
- Where do passwords, API keys and OAuth tokens live?
- In the Mac app and with rt in the operating system's keychain, in the web app encrypted in the browser's storage. The collection files hold only the variable name. Values from 1Password, HashiCorp Vault, AWS Secrets Manager, Azure Key Vault or Google Secret Manager are fetched through their command lines and kept in memory only.
- Can an AI agent change my production data through the MCP server?
- Not without you. Every environment has a rule: none, read or write; the default is read. A writing request against an environment marked as production waits for your approval in the app banner or with rt approve. rt mcp reads the rules at start; loosening them needs a restart.
- Does an agent see my secrets?
- No. Agents work with variable names. The core adds the values when sending, and every output to the agent is masked at the end – also when a server echoes a token back in its body.
- Can scripts or plugins access my files?
- No. Scripts run in a sandbox (QuickJS as WebAssembly) with limits for computation and 64 MiB of memory. Plugins are WebAssembly without native code; they reach the network only for hosts named in their manifest, which you see before installing. Neither sees files.
- How do I report a vulnerability?
- Confidentially by e-mail to office@clouddna.at with the subject Security, not in a public issue. We confirm receipt and get back to you with the next steps.
Your question isn’t here? Talk to me: +43 664 1964199 office@clouddna.at Get in touch