Releases: rohansx/cloakpipe
Release list
v0.10.0
What's Changed
- docs: align license to MIT + fix CLI/install docs to match the binary by @rohansx in #4
- feat: Dockerfile + green CI + compliance policies + SDK fixes by @rohansx in #5
- ci: fix clippy unnecessary_sort_by so main CI is green by @rohansx in #6
- chore(release): v0.10.0 + GHCR release workflow by @rohansx in #7
New Contributors
Full Changelog: v0.9.0...v0.10.0
Contributors
Assets 2
v0.9.0 — DistilBERT-PII NER, Real-World Benchmarks
v0.9 — DistilBERT-PII NER (63MB, 33 entity types, any CPU), nvidia/gliner-PII sidecar, expanded regex patterns, 91.7% real-world PII protection. See README for full benchmarks.
v0.8.0
v0.8.0 — Format-Preserving Pseudonymization, Response Scanning, Python Integrations
New Features
- Format-preserving pseudonymization: phone, email, Aadhaar, PAN replaced with format-identical fakes
- Response output scanning: LLM responses scanned for PII leakage, auto-redacted
- RAG scan CLI (
cloakpipe scan): scan files/dirs for PII before vector DB indexing - LangChain & LlamaIndex packages: drop-in
ChatCloakPipe,CloakPipeLLMwrappers - Batch detect API (Cloud):
POST /api/detect/batchfor up to 100 texts - Compliance framing: DPDP 2023, GDPR, HIPAA, PCI-DSS policy files
Stats
- 122 tests passing (12 new)
- 8 crates published to crates.io
- Clippy clean, zero warnings
v0.7.0 — Context-Aware Pseudonymization + GLiNER + Benchmarks
What's New
Context-Aware Pseudonymization
- Session tracking with coreference resolution (pronouns, abbreviations, possessives)
- Sensitivity escalation across sessions
- CLI session commands + MCP session_context tool
GLiNER Zero-Shot NER Backend
- New backend = gliner option alongside BERT NER
- Define any entity type in plain English, no training needed
- Pure ONNX Runtime, no Python dependency
Detection Engine (F1 = 0.99)
- Added: SSN, Aadhaar, PAN, IPv4, general URLs, natural dates
- Added: sk-proj/sk-live/github_pat secret patterns, INR/USD/EUR amounts
- Fixed: Phone regex no longer matches IPs, years, port numbers
PII Benchmark Harness
- 25 samples, 10 categories, 44 ground-truth spans + adversarial edge cases
- Run: cargo run -p cloakpipe-core --example pii_benchmark
Dashboard
- Admin dashboard with privacy chat, detection feed, compliance views
- PowerSync local-first SQLite
Full Changelog: v0.6.0...v0.7.0
v0.6.0 — Fuzzy Entity Resolution
Fuzzy entity resolution using Jaro-Winkler similarity + prefix matching + user-defined alias groups. Category-gated matching (won't merge person 'Rishikesh' with location 'Rishikesh'). Integrated into vault.get_or_create() — zero changes needed in detection pipeline. Sub-millisecond overhead. 66 tests passing (9 new resolver tests).
v0.5.0 — Industry Profiles + MCP Server
Industry profiles (general, legal, healthcare, fintech) with pre-tuned detection configs. MCP server with 5 tools (pseudonymize, rehydrate, detect, vault_stats, configure) via rmcp 1.1.0. 57 tests passing.
v0.4.0 — ADCPE Vector Encryption
ADCPE: Approximately Distance-Preserving Cryptographic Encryption for embedding vectors. Orthogonal matrix transforms preserve cosine similarity exactly. Includes encrypt/decrypt CLI commands and 10 tests.
v0.3.0 — SQLite + NER + Multi-user
SQLite vault and audit backends with per-value AES-256-GCM encryption. ONNX BERT-based NER behind --features ner flag. Multi-user token scoping.
v0.2.0 — CloakTree
CloakTree: vectorless, LLM-driven document retrieval for structured documents (PDF + Markdown). Includes parser, tree indexer, search, and CLI subcommands.