Changelog | Gibs Docs

[0.5.7] - 2026-03-12

Added

GitHub Action — classify AI systems and check compliance directly in CI/CD. Blocks deploys on prohibited/high-risk classification. Two modes: classify (risk assessment) and check (compliance Q&A). Available at github.com/gibbrdev/gibs-action
Usage visibility — admin endpoint for monitoring API usage across organizations, aggregated by endpoint and regulation

Improved

Webhook reliability — idempotent event processing prevents duplicate handling
Data retention — automatic log cleanup enforced on startup

[0.5.6] - 2026-03-11

Added

MCP tool expansion — 3 new tools for AI agent integration: list_regulations (discover available regulations and topics), get_article (fetch raw article text by regulation and number), check_regulation_scope (determine which regulations apply to your company). Total: 6 MCP tools
Article lookup endpoint — GET /v1/articles/{regulation}/{article_number} returns full article text directly from the corpus. No RAG pipeline, single-digit ms latency. Accepts flexible input formats ("5", "Article 5", "Art. 5")

Improved

MCP tool descriptions — rewritten for better AI agent discovery. Each tool includes example inputs, detailed parameter docs, and usage hints

[0.5.5] - 2026-03-11

Added

Monthly usage quotas — each plan now enforces a monthly request limit (Free: 25, Developer: 500, Team: 5,000, Platform: 50,000). Returns 429 with X-Quota-Limit and X-Quota-Used headers when exceeded. Counter resets automatically each calendar month.

Improved

Accuracy — 93%+ on AI Act, 96%+ on DORA, 96%+ on GDPR, 97%+ on cross-regulation queries (200 evaluation questions). Improved abstention precision for out-of-scope questions.

Security

Removed unused configuration field that implied encryption-at-rest capability

[0.5.4] - 2026-03-02

Fixed

DORA query routing — resolved collection routing issue that caused DORA regulation queries to fail
GDPR corpus sync — production corpus updated to match full 644-chunk index (all 99 articles + 173 recitals)

Verified

All 3 regulations (AI Act, DORA, GDPR) confirmed operational with end-to-end query validation

[0.5.3] - 2026-02-23

Added

Structured JSON response mode — POST /v1/check now accepts response_format: "structured" to return machine-readable fields: summary, legal basis, requirements, timeline, and cited articles
Confidence score — POST /v1/check now returns a numeric confidence_score (0-1) alongside the text confidence level, computed from retrieval quality, citation density, and hedging signals
Public OpenAPI 3.1 spec — full API schema at api.gibs.dev/openapi.json with typed request/response models for SDK and tool integration

Improved

Faithfulness verification — verification now checks against all retrieved sources instead of a truncated subset
Completeness checking — single-aspect questions now run the full completeness pipeline instead of being skipped

[0.5.2] - 2026-02-22

Added

Knowledge graph — cross-reference retrieval powered by a graph database with typed edges across all three regulations
AI agent discoverability — llms.txt and AI-friendly robots.txt at gibs.dev for LLM and agent crawlers
MCP server authentication — connections now require API key authorization

Improved

GDPR corpus expanded — full coverage of all 99 articles + 173 recitals
Retrieval precision — graph-enhanced pipeline with multi-level enrichment and reranking reduces noise in single-regulation queries
Accuracy — 93%+ accuracy across all three regulations

[0.5.1] - 2026-02-21

Improved

Upgraded synthesis engine — 93%+ accuracy across all three regulations (200+ evaluation questions)
Verification pipeline optimization — reduced latency by skipping redundant checks on well-cited answers
DORA corpus fully validated with expanded evaluation benchmark

Security

MCP server authentication — API key required on all connections
Security headers on all API responses (HSTS, X-Content-Type-Options, X-Frame-Options, Referrer-Policy)
Improved API key validation
Request body size limits on optional fields
Rate limiting resilience improvements

Fixed

MCP server disconnect handling — clean shutdown on client disconnection
Demo streaming layout stability

[0.5.0] - 2026-02-16

Added

DORA regulation support — Digital Operational Resilience Act: 64 articles + 12 delegated/implementing acts. Covers ICT risk management, incident reporting, TLPT, third-party oversight
Streaming responses — POST /v1/check/stream for real-time synthesis via Server-Sent Events
MCP server listed on MCP directories for AI agent discovery

Improved

Question routing — expanded keyword detection for DORA-specific terms (ICT risk, TLPT, financial entities)
Evaluation infrastructure — retry on rate limits, configurable delay between requests

Corpus

AI Act: 113 articles, 13 annexes, 180 recitals
GDPR: 99 articles, 173 recitals
DORA: 64 articles + 12 delegated/implementing acts
Three fully indexed EU regulations with cross-reference resolution

[0.4.0] - 2026-02-15

Added

Python SDK (pip install gibs) — sync + async clients, full type coverage
JavaScript SDK (npm install @gibs-dev/sdk) — TypeScript, native fetch, ESM + CJS
Multi-regulation routing — questions automatically routed to the correct corpus

Improved

Source citations now include relevance scores from reranker
Synthesis engine upgraded with stronger scope discipline

[0.3.3] - 2026-02-13

Added

API key management — create, list, and revoke named keys from the dashboard
Multiple keys per organization (up to 5) with per-key usage tracking
Self-service key lifecycle: create for different environments, revoke individually

Fixed

Dashboard stability improvements
Infrastructure health check reliability

[0.3.2] - 2026-02-12

Added

Corpus isolation — each regulation now has its own dedicated index for better precision
Complete legal metadata on every source: document type, legal status, application date, binding status
MCP integration section on landing page

Fixed

GDPR application date corrected
Improved multi-corpus query routing

Corpus

AI Act: 113 articles, 13 annexes, 180 recitals
GDPR: 99 articles, 173 recitals
Full metadata enrichment on all sources

[0.3.1] - 2026-02-12

Improved

Inference caching — faster responses on repeated queries
Scoring accuracy — better boundary matching and negation handling

Added

Full purchase lifecycle: signup, checkout, upgrade, API key management, downgrade
Upgraded synthesis engine for higher accuracy
Abstention guard — smarter handling when no relevant sources exist

Fixed

Subscription cancellation properly clears payment state
Webhook handlers resilient to edge cases

[0.3.0] - 2026-02-10

Added

90-question evaluation benchmark
Cross-reference resolution across regulations
Multi-pass verification pipeline with completeness checks
API key authentication with per-organization rate limiting
Stripe integration for subscription management

Performance

88% accuracy on regulatory benchmark
100% abstention accuracy on out-of-scope questions
Article-level citation in every response

[0.1.0] - 2026-02-10

Added

Initial API release
POST /v1/classify — AI system risk classification
POST /v1/check — Compliance Q&A with citations
GET /health — Health check endpoint
EU AI Act corpus (113 articles, 13 annexes)
GDPR selected articles
Commission guidelines integration
MCP server for AI agent integration

Version Format

We use Semantic Versioning:

MAJOR: Breaking API changes
MINOR: New features, backward compatible
PATCH: Bug fixes, corpus updates

Corpus versions are tracked separately and included in API responses.