Skip to content

AI & Machine Learning

LLMs, vector databases, ML frameworks, and AI infrastructure tooling.

Subscribe
All Breaking Security Features
Important Recent
← Releases
semantic-kernel python-1.43.0 Breaking risk
Breaking changes
  • Update OpenAPI document parsing options (breaking change)
No immediate action
Glq v0.5.1 Breaking risk

Default inline-dequant E8 KV

Open
Upgrade now
ArcadeData/arcadedb 26.6.1 Breaking risk
Auth RBAC RCE / SSRF +2 more

TLS HA, durability hardening, security fixes

Open
Review required
Transformers v5.10.1 Breaking risk
Breaking upgrade

Gemma4 Unified Sapiens2 OCR-2 Mellum fixes

Open
No immediate action
lancedb v0.30.1-beta.1 Breaking risk

Remote tables + metadata updates

Open
Review required
haystack v2.30.0 Breaking risk
Breaking upgrade

PythonCodeSplitter + string ChatGenerators + async retrievers

Open
No immediate action
Forge v0.7.4 Breaking risk

Tool-error correction + max‑errors flag

Open
Config change
LightRAG v1.5.0 Breaking risk
Breaking upgrade

Multimodal support + native DOCX parsing

Open
Config change
graphify v0.8.29 Breaking risk
Auth

providers.json opt‑in

Open
No immediate action
Aura v1.4.4 Breaking risk

Persisted Worker logs + UI cleanup

Open
Review required
brunoamaral/gregory-ai v24 Breaking risk
Auth RBAC Breaking upgrade

Multi-tenant GregoryAI with search & source upgrades

Open
No immediate action
weaviate v1.36.17 Breaking risk

Fixes across db components

Open
v1.35.22 (1d) Compaction + delete fixes
No immediate action
Nexa-gauge v0.1.18 Breaking risk

refalign node + refmatch rename

Open
No immediate action
weaviate v1.37.7 Breaking risk

Compaction abort + async-delete fix + test improvements

Open
No immediate action
open-webui v0.9.6 Breaking risk

Routine maintenance and dependency updates.

Open
Upgrade now
photoprism 260601-a7d098548 Breaking risk
Auth Dependencies

Security hardening + disk threshold + video fixes

Open
Ktx v0.8.0 Breaking risk
⚠ Upgrade required
  • Update CI configurations to handle normalized trailing newline in `star-history.svg` and refresh the chart twice daily via RELEASE_PAT on protected main branch.
Breaking changes
  • Removed `fast` mode from `ktx ingest`; ingestion always builds enriched context (KLO-721).
Notable features
  • Added shell completion for commands, flags, and entity names.
  • Enabled PostHog GeoIP enrichment in telemetry.
No immediate action
opik 2.0.54 Breaking risk

Dominant project prompt assignment

Open
LLMKube v0.8.1 Breaking risk
⚠ Upgrade required
  • After upgrading to v0.8.1, re‑apply all Agent CRs so existing Agents pick up explicit values for the new requestTimeoutSeconds and requestTurnTimeoutSeconds fields.
Breaking changes
  • Agent.spec.requestTimeoutSeconds now represents a loop-wide wall-clock budget (default 3600) instead of per-request HTTP timeout; the former behavior is moved to Agent.spec.requestTurnTimeoutSeconds (default 120). Re‑apply Agent CRs after upgrade.
Notable features
  • **inferenceservice:** adds typed spec.ropeScaling for RoPE/YaRN context extension
Forge v0.7.3 Breaking risk
⚠ Upgrade required
  • Update any scripts or configs using `--mode` to use `--backend-capability native` or remove the flag for default behavior
  • Ensure backend capability selection matches backend support (prompt capability now rejected loudly for ollama, vllm, anthropic)
Breaking changes
  • Flag `--mode {native,prompt}` renamed to `--backend-capability {native,prompt}`; no deprecation alias – migration required (drop flag for native default or explicitly set `--backend-capability`)
  • Runtime `auto` function‑calling mode removed from LlamafileClient
Notable features
  • Added OpenAICompatClient for arbitrary OpenAI‑compatible endpoints
  • Added configurable `--backend-timeout` proxy option (default 300s)
  • Native function calling now transparent passthrough, forwarding client tools/messages verbatim

Beta — feedback welcome: [email protected]