Anonmyz v0.2.0

Anonmyz v0.2.0

Anonmyz v0.2.0 is the first release under the Anonmyz name with the hardened local-first DLP pipeline, reproducible privacy demo, and one-command installation.

Install

Linux and macOS:

curl -fsSL https://raw.githubusercontent.com/3mre0s/ai-firewall/main/scripts/install.sh | sh

Windows PowerShell:

irm https://raw.githubusercontent.com/3mre0s/ai-firewall/main/scripts/install.ps1 | iex

Then run anonmyz demo --non-interactive. The demo needs no API key and contacts only dynamically selected loopback ports.

Highlights

• Shared DLP engine for explicit proxy and MITM traffic, including bounded stream inspection and placeholder restoration.
• Large-body performance hardening: a safe 32 MiB body now allocates about 21 KiB instead of roughly 1.07 GiB.
• Configuration validation through anonmyz doctor and anonmyz --check-config.
• Safer Codex sessions with protected provider routing and fail-closed behavior.
• Trusted binary discovery and lifecycle tests for the VS Code and JetBrains integrations.
• Checksum-verifying installers for Linux, macOS, and Windows.
• Release archives for five OS/architecture targets, compatibility binaries, VSIX, SPDX SBOMs, checksums, and GitHub/Sigstore provenance attestations.

Security notes

• Remote upstreams require HTTPS; plaintext HTTP is limited to loopback development services.
• Standard upstream responses are capped at 64 MiB.
• New encrypted MITM CA keys use PBKDF2-HMAC-SHA-256 with 210,000 iterations and retain legacy read compatibility.
• Workspace and project-root binaries are not automatically trusted by IDE extensions.

Anonmyz remains pre-1.0. Explicit proxy mode and the VS Code integration are the recommended paths; evaluate transparent MITM and JetBrains support in your own environment before broad rollout.

Anonmyz — OpenAI Build Week 2026

Anonmyz — OpenAI Build Week 2026

Final source commit: 4b995bbde26cf49690ff35fc5e3b25e8b7151243

This release contains the judge package for Anonmyz, a local-first AI security and DLP gateway for coding agents.

Highlights:

• deterministic, API-key-free security demo;
• protected Codex Safe Session using an existing ChatGPT subscription or optional API key;
• metadata-only local privacy audit trail;
• fail-closed proof: Codex stops with an error when the loopback proxy disappears, with zero direct model fallback attempts;
• Windows, Linux, and macOS binaries with SHA-256 checksums;
• Apache-2.0 license, judge guide, Devpost copy, video script, provenance, and sanitized evidence.

Verify anonmyz-build-week-jury.zip with the attached .sha256 file before extracting it.

Local AI Firewall v0.1.2

Local AI Firewall v0.1.2

First public binary release of Local AI Firewall, a local proxy that reduces accidental disclosure of recognized secrets and personal data in requests sent to supported AI providers.

Highlights

• Local request masking and response restoration with request-scoped in-memory vaults
• Explicit proxy mode for Claude Code and OpenAI-compatible clients
• Optional transparent MITM mode with a local CA
• Offline ai-firewall demo using synthetic credentials
• VS Code and Cursor extension packaged as a release asset
• Linux and macOS builds for amd64 and arm64, plus Windows amd64
• SHA-256 checksums for every platform archive and the VSIX

Install

Download the archive for your platform and checksums.txt below. Verify the checksum before extracting the executable. The release archives contain README.md, CHANGELOG.md, LICENSE, and the ai-firewall executable (ai-firewall.exe on Windows).

Important limitations

Detection is pattern-based and best-effort. Unsupported or encoded credentials may not be detected, and only traffic routed through the firewall is scanned. Transparent MITM mode changes the local trust model and is disabled by default. Review THREAT_MODEL.md before deployment.

Verification performed before publication

• GitHub Actions test, vet, format, lint, extension, and five-platform cross-compile jobs passed
• All downloaded release assets matched checksums.txt
• Windows binary reported ai-firewall 0.1.2
• Linux archive contents were verified
• Downloaded VSIX installed successfully as localai.local-ai-firewall@0.1.2

v0.1.0

Local AI Firewall v0.1.0

This release makes Local AI Firewall easier and safer to evaluate before configuring a real AI provider.

Added

• Fully offline ai-firewall demo command
• Synthetic credential masking demonstration
• Focused Claude Code setup guide
• Safe issue forms for missed secrets, false positives, bugs, and integration requests
• CI and release workflows with cross-platform binaries, VSIX packaging, and checksums

Improved

• Simplified first-time onboarding
• Explicit proxy mode is now the default documented quickstart
• Transparent MITM mode is documented as an advanced option
• Added clearer trust and limitations sections
• Fixed incorrect repository references
• Completed duplicate-label handling required by the production masker

Try it

ai-firewall demo

The demo uses production masking logic with synthetic values. It does not require an API key, start a proxy, install a certificate, initialize telemetry, or send a network request.

Detection remains pattern-based and is not guaranteed to identify every possible secret format. See THREAT_MODEL.md for documented limitations.

Full Changelog: https://github.com/3mre0s/ai-firewall/commits/v0.1.0