Privacy Policy
SecretHarbor is engineered under an uncompromising cryptographic and architectural rule: your code, environment variables, credentials, and agent sessions never leave your local workstation. All boundary enforcement executes 100% on-device at the OS kernel layer. We run zero remote telemetry servers, collect zero analytics, and transmit zero runtime data over the network.
Zero phone-home calls, zero cloud telemetry, and zero remote analytics. Operates completely air-gapped.
Secrets stored strictly in ~/.secretharbor/vault.enc with authenticated hardware/passphrase keys.
Audit trail in ~/.secretharbor/audit.log with automated entropy and regex credential masking.
01. Core Architectural Commitment
This Privacy Policy applies to the SecretHarbor CLI application (shb), the kernel-level runtime security boundary, associated transparent shims, and the marketing website at secretharbor.dev.
Unlike developer tooling that routes command telemetry, usage metrics, or error dumps to remote cloud endpoints, SecretHarbor is deliberately designed to provide host-level defense-in-depth without external observation. We believe that tools built to safeguard developer secrets must not themselves become an exfiltration channel.
02. Zero Telemetry & On-Device Processing
SecretHarbor enforces a total ban on background data collection:
- No Telemetry or Analytics: The
shbbinary contains no telemetry libraries, tracking beacons, crash uploaders, or pingback routines. - No Code or Context Uploads: Workspace files, diffs, repository names, Git history, file names, and shell commands are never transmitted to any external server.
- No Agent Session Interception: Prompts, LLM completions, and context windows exchanged between your local AI agent (Claude Code, Cursor, Codex, Google Antigravity, OpenCode, Kiro, etc.) and model providers are not inspected, copied, or stored by SecretHarbor.
- Air-Gapped Operation: SecretHarbor functions identically with full fidelity on systems with zero active internet access.
03. Local Encrypted Credential Vault
When you store credentials using shb secret set <KEY>, your data is protected using state-of-the-art cryptography:
- Storage Location: All managed secrets reside exclusively on your local filesystem in
~/.secretharbor/vault.enc. - Cryptographic Standard: The vault is encrypted using AES-256-GCM (Galois/Counter Mode) authenticated encryption, providing both confidentiality and cryptographic integrity against tampering.
- Key Derivation: Encryption keys are derived using memory-hard key derivation functions (Argon2id/scrypt) or bound to your OS hardware-backed secure enclave/keychain where available.
- Zero Escrow: No recovery keys, master passphrases, or cryptographic backdoors exist. Only the human operator with local workstation access can unlock the vault.
04. Local Redacted Audit Logging
To maintain transparency into how AI agents interact with your system, SecretHarbor records operational security decisions locally:
- Log Location: Audit events are appended exclusively to
~/.secretharbor/audit.logon the local machine. - Automated Credential Redaction: Before any log record is written to disk, an automated redaction engine scans command lines, environment assignments, and argument buffers. Any value matching high-entropy secret patterns, API key prefixes, or vault keys is replaced with format-safe tokens (e.g.
[REDACTED_SECRET]). - Local Retention & Deletion: You retain complete control over audit records. Logs are never forwarded off-host. You can inspect your log via
shb audit viewor completely purge records usingshb audit clearat any time.
05. Operating System Boundary Mechanics
SecretHarbor intercepts capability requests at the host operating system level beneath the AI agent process:
- macOS (Apple Seatbelt): Employs kernel-level Seatbelt profiles (
sandbox-exec/libsandbox) to isolate filesystem access, sanitize environment variables, and virtualize sensitive paths without modifying system-wide files. - Linux (Namespaces & Mount Isolation): Leverages unprivileged user namespaces (
CLONE_NEWUSER,CLONE_NEWNS,CLONE_NEWIPC) and temporary mount overlays to restrict agent visibility. Memory dumping is suppressed viaPR_SET_DUMPABLE=0. - Process Inheritance: Child processes, subprocesses, MCP servers, and background build scripts spawned by the agent inherit the restrictive boundary down the process tree.
06. Synthetic Secret Virtualization (Two-World Projection)
SecretHarbor solves the agent credential dilemma by establishing two isolated projections of the same workspace:
- Human Developer View: Retains full access to real credentials, database connection strings, SSH agent sockets, and cloud tokens.
- AI Agent View: Receives format-preserving synthetic values (e.g.
sk_test_sh_fake_••••orpostgresql://fake@localhost:5432/db). The agent can execute tests, validate syntax, and refactor code without ever touching authentic secrets. - In-Memory Isolation: Real secrets never enter the memory address space of the untrusted agent process.
07. Network Egress & Socket Interception
SecretHarbor provides granular controls over network and socket exposure:
- Local Policy Enforcement: Network filtering occurs strictly via host-level routing rules and local proxies. No external VPNs, third-party inspection relays, or cloud middleboxes are used.
- Docker & Daemon Sockets: Direct access to
/var/run/docker.sockis blocked. Capabilities are granted exclusively via explicit human approval through short-lived leases. - No Egress Analytics: Metadata regarding outgoing connection requests is logged strictly to your local
audit.logfile.
08. Website Properties (secretharbor.dev)
Our web presence at secretharbor.dev adheres to the same uncompromising privacy standard:
- Zero Cookies: We set no first-party or third-party cookies. No cookie consent banners are necessary.
- Zero Tracking Scripts: We do not load Google Analytics, Meta Pixel, ad-network tracking scripts, behavioral profiling tools, or third-party fonts.
- Privacy-First Aggregate Metrics: To keep the project healthy, we measure aggregate visit counts via Cloudflare Web Analytics and aggregate installer download counts via a Cloudflare edge counter. These mechanisms use no cookies, no browser storage, no cross-site identifiers, and no personal data.
- No Personal Data Collection: This website collects no name or contact details beyond what you explicitly volunteer through the contact form. The only personal data processed via the site is the email address and message you choose to submit in the form, which is used solely to reply to you. It is never used for analytics, profiling, or marketing, and it is not shared with third parties beyond the form delivery service.
- Server Logs: Edge hosting infrastructure (such as Cloudflare Pages or GitHub Pages) records standard ephemeral HTTP request logs (IP address, user-agent, requested path) strictly for routing, security defense (DDoS mitigation), and protocol compliance. These logs are automatically rotated and discarded according to standard provider lifecycles.
09. Complete Data Deletion & Portability
Because all SecretHarbor state is stored locally, data erasure is instant and irreversible:
# Completely remove all SecretHarbor state, keys, and logs:
rm -rf ~/.secretharbor
# Remove project-level boundary configuration:
rm -f secretharbor.yaml
Executing the commands above immediately destroys the encrypted vault (vault.enc), local configuration (config.yaml), transparent shims (shims/), and audit logs (audit.log). No residual data remains on any cloud server because no cloud server was ever used.
10. Open Source Auditing & Verification
SecretHarbor is open-source software licensed under the Apache License, Version 2.0. You do not have to take our privacy claims on faith:
- Every line of source code is publicly auditable on our repository.
- Build artifacts and release binaries are published with cryptographic checksums (SHA-256) and reproducible build recipes.
- You can compile SecretHarbor directly from source using Go 1.22+ to independently verify every behavior described in this policy.
11. Policy Updates
Any updates to this Privacy Policy will be reflected on this page with a revised "Effective Date" and published in our source repository change log. Because we collect no email addresses, we encourage users to review our documentation and repository release notes periodically.
12. Security Disclosure & Contact
If you have questions regarding this Privacy Policy or discover a potential security vulnerability in SecretHarbor, please contact the security team via our contact page, our GitHub security advisory portal, or email:
mail@secretharbor.dev