summary |
shortlog | log |
commit |
commitdiff |
tree
first ⋅ prev ⋅ next
Stefan Gasser [Sun, 11 Jan 2026 18:42:09 +0000 (19:42 +0100)]
Fix PII detection to scan all message roles (#25)
Previously, PII detection only scanned the last user message initially,
then did a full scan only if PII was found. This caused PII in system
messages (e.g., RAG context from PDFs) to be missed entirely when the
user message contained no PII.
Changes:
- Consolidate analyzeMessages() to always scan all messages
- Scan system, developer, user, and assistant roles
- Remove analyzeAllMessages() as it's no longer needed
- Simplify decision.ts by removing the redundant full scan call
This ensures PII in system messages (common in RAG patterns) is properly
detected and masked before being sent to upstream LLMs.
Fixes #17
Max Wolf [Sun, 11 Jan 2026 17:55:34 +0000 (18:55 +0100)]
Refactor secrets detection into pattern registry (#18)
* Add PatternDetector and DetectionResult interfaces for secrets detection registry
* Move all interfaces to patterns/types.ts and use the existing SecretesDetectionResult interface instead of the new DetectionResult
* Move pattern detection utility to new patterns/utils.ts module
* Refactor secrets detection using a registry system
- Create privateKeysDetector, apiKeysDetector, tokensDetector modules
- Refactor detectSecrets() to use the pattern registry
- Re-export types from detect.ts for backwards compatibility
* Change default secrets_detection action to redaction
Hint: The example config still shows `action: block` explicitly, with a comment noting
that `redact` is the default action if not specified
* Fix README default action references and improve overall structure / formatting
- Update all references from 'block (default)' to 'redact (default)'
- Fix Bearer token documentation (20+ → 40+ chars)
- Reorganize Configuration section with consistent headers
- Improve table formatting and section descriptions
- Use references to reduce duplications and maintenance overhead
* Improve type safety in PatternDetector interface
Use SecretEntityType instead of string for enabledTypes Set parameter
* Update docs to reflect redact as new default action
Reorder actions to show default first
---------
Co-authored-by: Stefan Gasser <redacted>
Stefan Gasser [Sun, 11 Jan 2026 10:11:54 +0000 (11:11 +0100)]
Add Mintlify documentation and simplify README (#24)
* Add Mintlify documentation and simplify README
Documentation:
- Add complete Mintlify docs with introduction, quickstart, integrations
- Add concept guides: mask mode, route mode, PII detection, secrets detection
- Add API reference: chat completions, models, status, dashboard API
- Add configuration guides: overview, providers, PII, secrets, logging
- Include dashboard screenshot and branding assets
README:
- Simplify structure, move detailed docs to Mintlify
- Add centered badges and navigation links
- Add "What is PasteGuard?" section explaining problem/solution
- Update example to Dr. Sarah Chen (consistent across all docs)
- Reorder integrations (OpenAI SDK, LangChain, LlamaIndex first)
- Move Presidio attribution inline with PII section
- Add Tech Stack section
Code:
- Update description to "Privacy proxy for LLMs" in package.json,
startup banner, and /info endpoint
Closes #21
* docs: fix secrets default to block (matches current code)
Stefan Gasser [Sun, 11 Jan 2026 09:14:06 +0000 (10:14 +0100)]
fix: support developer and function roles for GPT-5.x compatibility (#23)
GPT-5.x models use the 'developer' role instead of 'system' for
instructions. The 'function' role is also added for legacy compatibility.
Fixes #22
Stefan Gasser [Sat, 10 Jan 2026 16:47:28 +0000 (17:47 +0100)]
feat: Update dashboard to match design system (#16)
* feat: Update dashboard to match design system
- Apply design system tokens (colors, typography, radius, shadows)
- Add branded "Redaction Bar Loader" animation for loading/empty states
- Add favicon route serving inline SVG
- Update README with wordmark SVG and optimized screenshot
- Move assets to assets/ folder (logo, wordmark, favicon, dashboard)
* fix: Formatting and browser compatibility fixes
- Fix biome formatting in favicon route (multi-line object)
- Add color-mix() fallback for card hover effect (older browsers)
Stefan Gasser [Sat, 10 Jan 2026 14:00:48 +0000 (15:00 +0100)]
Fix lint errors from multimodal PR (#11)
- Remove unused imports in multimodal.test.ts
- Remove unused secretsConfig variable
- Fix import ordering in masking.ts and content.test.ts
- Remove trailing whitespace in proxy.ts and masking.ts
Stefan Gasser [Sat, 10 Jan 2026 13:23:29 +0000 (14:23 +0100)]
Merge pull request #8 from mkroemer/fix/multimodal-content-support
Fix/multimodal content support
mkroemer [Sat, 10 Jan 2026 08:21:33 +0000 (09:21 +0100)]
You are right, and I (and claude) addressed these issues.
1. Secrets redaction (proxy.ts):
- Add per-part offset tracking to prevent partial secret leaks
- Filter and adjust redaction positions for each text part
- Fixes issue where 'sk-proj-' prefix could remain visible
2. PII masking (masking.ts):
- Actually mask array content instead of returning original
- Add per-part offset tracking for accurate entity positions
- Properly handle multimodal arrays with text and images
3. Tests:
- Add content.test.ts for text extraction utilities
- Add multimodal.test.ts with integration tests
mkroemer [Fri, 9 Jan 2026 20:06:49 +0000 (21:06 +0100)]
fix: Update type system for multimodal content support
- Update ChatMessage interface to accept MessageContent type
- Fix masking.ts to handle multimodal content
- Fix redact.ts to extract text before processing
- Fix import order and formatting per biome linter
mkroemer [Fri, 9 Jan 2026 19:58:56 +0000 (20:58 +0100)]
fix: Add support for multimodal content (text + images)
Fixes crashes when processing OpenAI Chat Completion requests with
multimodal content (array format). Previously, the code assumed message
content is always a string, causing Presidio errors and 502/503 responses
when LibreChat Agents sent vision requests.
Changes:
- Add extractTextContent() utility to safely extract text from both
string and array content formats
- Update PII detection to handle multimodal messages
- Update secrets detection to extract text from array content
- Update message redaction to preserve images while redacting text
- Fix dashboard display to show readable content instead of [object Object]
This enables full support for OpenAI's multimodal API format while
maintaining PII/secrets protection on text portions.
Resolves issues with LibreChat Agents feature returning 502 errors.
Stefan Gasser [Fri, 9 Jan 2026 17:26:03 +0000 (18:26 +0100)]
Merge pull request #7 from maximiliancw/main
Add secrets detection to prevent credential leaks
maximiliancw [Fri, 9 Jan 2026 16:10:15 +0000 (17:10 +0100)]
Refine bearer token and JWT token detection patterns to enforce minimum character requirements;
Updated tests to reflect changes in bearer token format
maximiliancw [Fri, 9 Jan 2026 16:03:55 +0000 (17:03 +0100)]
Update all references to outdated versions
maximiliancw [Fri, 9 Jan 2026 16:00:19 +0000 (17:00 +0100)]
Enhance request logging to conditionally include detected secret types based on configuration; ensuring sensitive information is only logged when explicitly allowed, improving security and compliance
maximiliancw [Fri, 9 Jan 2026 15:57:49 +0000 (16:57 +0100)]
Clarify max_scan_chars behavior for secrets detection; Added notes on detection limits and performance implications
maximiliancw [Fri, 9 Jan 2026 15:56:09 +0000 (16:56 +0100)]
Fix: removed space in joined header string for consistency
maximiliancw [Fri, 9 Jan 2026 15:52:46 +0000 (16:52 +0100)]
Use OpenAI-compatible error format for secrets blocking:
- Use 400 status code instead of 422
- Use standard error format {message, type, param, code}
- Remove non-standard 'details' field (secret types already in headers)
- Update tests to match new format
maximiliancw [Fri, 9 Jan 2026 15:23:47 +0000 (16:23 +0100)]
Add database migration for secrets detection columns
Existing installations will fail with 'table request_logs has no column
named secrets_detected' since the new columns don't exist in their SQLite
database. This adds a migration check that adds the missing columns if they
don't exist.
maximiliancw [Fri, 9 Jan 2026 15:04:32 +0000 (16:04 +0100)]
Implement redact and route_local actions:
- Integrate redact action: redacts secrets before PII detection, unredacts in responses
- Implement route_local action: routes requests with secrets to local provider
- Update stream transformer to handle both PII and secrets contexts
- Add comprehensive tests for secrets routing logic
- Update config.example.yaml with new entity types and action documentation
- Update README.md with complete secrets detection features
New secret entity types (opt-in):
- API_KEY_OPENAI, API_KEY_AWS, API_KEY_GITHUB
- JWT_TOKEN, BEARER_TOKEN
Response headers:
- X-PasteGuard-Secrets-Redacted: true (when action=redact)
maximiliancw [Fri, 9 Jan 2026 15:02:22 +0000 (16:02 +0100)]
Add reversible redaction module for secret masking:
- Create redact.ts with RedactionContext for tracking secret mappings
- Implement redactSecrets() with configurable placeholder format
- Implement unredactSecrets() for restoring original secrets in responses
- Add streaming helpers for unredacting SSE responses
- Add comprehensive tests covering roundtrip, multiple messages, and streaming
maximiliancw [Fri, 9 Jan 2026 14:52:56 +0000 (15:52 +0100)]
feat(secrets): add detection for API keys, JWT tokens, and Bearer tokens
- Add new secret entity types: API_KEY_OPENAI, API_KEY_AWS, API_KEY_GITHUB, JWT_TOKEN, BEARER_TOKEN
- Extract pattern detection into reusable helper function
- Add comprehensive tests for all new secret types with false positive checks
- Update config schema with typed entity enum
maximiliancw [Fri, 9 Jan 2026 14:38:08 +0000 (15:38 +0100)]
Improve error messaging for secrets detection configuration validation on startup
maximiliancw [Fri, 9 Jan 2026 14:24:41 +0000 (15:24 +0100)]
Merge remote-tracking branch 'origin/main'
# Conflicts:
# README.md
# src/routes/proxy.ts
maximiliancw [Fri, 9 Jan 2026 14:21:23 +0000 (15:21 +0100)]
Prepare for merge with breaking changes from remote origin
maximiliancw [Fri, 9 Jan 2026 13:59:18 +0000 (14:59 +0100)]
Add secrets shield feature documentation to README
maximiliancw [Fri, 9 Jan 2026 13:55:10 +0000 (14:55 +0100)]
Add simple integration tests for secrets detection blocking behavior in proxy.ts
maximiliancw [Fri, 9 Jan 2026 13:52:51 +0000 (14:52 +0100)]
Add comprehensive unit tests for secrets detection
maximiliancw [Fri, 9 Jan 2026 13:49:27 +0000 (14:49 +0100)]
Add validation for secrets detection config on startup and update banner
maximiliancw [Fri, 9 Jan 2026 13:44:59 +0000 (14:44 +0100)]
Enhance secrets detection functionality in proxy routes:
- Integrate secrets detection logic into the proxy request handling.
- Add configuration checks for enabling/disabling secrets detection.
- Implement logging for detected secrets and their types.
- Update the logger to accommodate new fields for secrets detection.
- Refactor related functions for improved clarity and maintainability
Stefan Gasser [Fri, 9 Jan 2026 13:37:25 +0000 (14:37 +0100)]
Rename setShieldHeaders to setPasteGuardHeaders
Stefan Gasser [Fri, 9 Jan 2026 13:27:51 +0000 (14:27 +0100)]
Update dashboard screenshot with PasteGuard branding (#6)
maximiliancw [Fri, 9 Jan 2026 13:02:05 +0000 (14:02 +0100)]
Extend .gitignore
Stefan Gasser [Fri, 9 Jan 2026 12:57:21 +0000 (13:57 +0100)]
Rename project from LLM-Shield to PasteGuard (#5)
After positive community response (100+ GitHub stars), committing fully
to this project. The name PasteGuard better describes the core function:
"guard what you paste" before sending to LLMs.
Changes:
- Update all branding: package.json, README, CONTRIBUTING
- Rename response headers: X-LLM-Shield-* → X-PasteGuard-*
- Update dashboard UI and page title
- Update /info endpoint metadata
- Update startup banner
- Update config files and defaults
- Update CI workflow docker tag
- Regenerate bun.lock with new package name
Domain pasteguard.com secured. Old GitHub links redirect automatically.
maximiliancw [Fri, 9 Jan 2026 12:55:01 +0000 (13:55 +0100)]
Add text extraction utility function
maximiliancw [Fri, 9 Jan 2026 12:49:45 +0000 (13:49 +0100)]
Implement secrets detection module for OpenSSH and PEM private keys
maximiliancw [Fri, 9 Jan 2026 12:33:47 +0000 (13:33 +0100)]
Add secrets_detection section to config.example.yaml
maximiliancw [Fri, 9 Jan 2026 12:31:10 +0000 (13:31 +0100)]
Add secrets detection config schema with validation
Stefan Gasser [Fri, 9 Jan 2026 08:07:56 +0000 (09:07 +0100)]
Rename chat routes to proxy for clarity (#4)
- Rename chat.ts → proxy.ts to better reflect purpose (LLM proxy routing)
- Update CLAUDE.md architecture to match actual file structure
Stefan Gasser [Fri, 9 Jan 2026 07:28:55 +0000 (08:28 +0100)]
Fix EISDIR error when config.yaml is a directory (#3)
Add isFile() check before reading config to give a clear error message
when Docker creates a directory instead of mounting a missing file.
Fixes #2
Stefan Gasser [Thu, 8 Jan 2026 16:15:59 +0000 (17:15 +0100)]
Add PII accuracy benchmark with multi-language phone context (#1)
- Add benchmark framework with precision/recall/F1 metrics
- Add 30 test cases across 5 languages (DE, EN, ES, FR, IT)
- Add phone_context words for all 24 supported languages
- Each language has 5-7 native words for: phone, number, mobile, call
Test with: bun run benchmark:accuracy
Stefan Gasser [Thu, 8 Jan 2026 11:15:42 +0000 (12:15 +0100)]
fix: support environment variables in numeric config values
Use z.coerce.number() to handle string values from environment variable
substitution in YAML config. Add validation constraints:
- port: integer 1-65535
- retention_days: integer >= 0
- score_threshold: float 0-1
Stefan Gasser [Thu, 8 Jan 2026 10:14:12 +0000 (11:14 +0100)]
Initial release
OpenAI-compatible privacy proxy with two modes:
- Mask: Replace PII with placeholders before upstream, unmask in response
- Route: Send PII-containing requests to local LLM
Features:
- 24 language support for PII detection
- Real-time streaming with unmasking
- Dashboard for monitoring
- Microsoft Presidio integration