AI.info
AI Tools Updates — Page 37
What changed in the AI tools this week — dated, from the makers’ own changelogs.
In inglese
- Opik — 2.2.133-7333
- Opik — 2.2.133
Online-eval rules select OpenAI and OpenRouter reasoning effort; annotator views were simplified; LLM clients fail over without replaying requests.
- Opik — 2.2.134-7336: [NA] [DOCS] Add Pixeltable integration page (#9205)
Added documentation for exporting Pixeltable OpenTelemetry traces to Opik, including the OTLP endpoint, headers, and an init() example.
- Opik — 2.2.133-7334
- Keploy — v3.6.174
Replay stays off the real network by default (userland part).
- Keploy — v3.6.175
Mappings.yaml keeps hand edits; refusals and miss destinations are fixed.
- Keploy — v3.6.176
Reads each app container’s time zone through Docker in Docker and Compose modes.
- Snowflake Cortex Analyst — Snowflake Data Clean Rooms updates
- llama.cpp — b11980
Adds a SYCL DPAS flash attention kernel for quantized KV caches.
- llama.cpp — b11985
Speeds up Metal MXFP4 mat-vec, unary and binary ops, and GLU.
- llama.cpp — b11986: opencl: initial support for iq2 (#31267)
- llama.cpp — b11983
Fixes unary kernels compiling on older Metal toolchains.
- llama.cpp — b11979
Vectorizes SYCL reordered-layout dequantization stores for Q4_0, K-quants, IQ3 and MXFP4.
- llama.cpp — b11982
Adds Hexagon support for the UPSCALE op.
- llama.cpp — b11981
Improves DDR bandwidth utilization for transposed CONCAT on Hexagon.
- llama.cpp — b11977
OpenVINO SCALE accepts strided view inputs and rejects in-place SCALE on non-contiguous views.
- Everlaw Ediscovery Platform — Everlaw Omni Brings Agentic AI to Complex, High-Stakes Legal Matters
Everlaw Omni launches as an agentic AI experience that orchestrates tools to find evidence.
- Everlaw Ediscovery Platform — Everlaw Brings Frontier AI Closer to the Evidence with ChatGPT Plugin
Everlaw launches a ChatGPT Enterprise plugin connecting its governed evidence layer with ChatGPT Enterprise.
- Everlaw Ediscovery Platform — Everlaw for Word Brings Your Evidence Directly to Where You Draft
Everlaw for Word brings evidence into Word and lets users ask questions across transcripts and manage designations.
- Zilliz Cloud — 🐦 Milvus v3.0.x Public Preview is now available on Dedicated clusters — upgrade from the console with no downtime and no index rebuild:
ORDER BY, query aggregation, search aggregation, regular expression filtering, MinHash, TEXT, schema changes, and entity-level TTL.
- n8n — n8n@2.46.2
Added managed OAuth app selection by version store; fixed queue jobs, migrations, execution access, and AI-builder chat routing.
- n8n — n8n@2.45.6
Recorded dynamic time saved for queue-mode executions and matched compact sidebar icon sizes.
- n8n — stable
Recorded dynamic time saved for queue-mode executions and matched compact sidebar icon sizes.
- n8n — n8n@1.123.90
Required signed tokens to accept invitations; Google Drive Trigger now reads watched items instead of listing changes.
- Hugging Face — Granite Switch 4.1.1 — Update models like you update software
Granite Switch 4.1.1 lets users update models like software.
- Hugging Face — Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platforms
AnySpeechRecognizer and AnyDecisionModel launch as open models for Apple platforms.
- Ollama — v0.40.13
LFM3 models, including lfm3-audio and lfm3-omni, now run on Linux and Windows as well as Apple silicon.
- Magnific — Meet Magnific Move: motion graphics you can talk to
Motion graphics you can talk to
- Hyperscience Platform — Copying user-installed Python packages to a newer Python version
Upgrades no longer copy user-installed packages built for a different Python version and list packages not copied in a warning.
- Hyperscience Platform — Brotli decompression (CVE-2026-111497)
The system uses Brotli 1.2.1, which rejects malformed compressed data.
- Deepdub GO — Deepdub and Spikes Partner to Turn One Piece of Content Into a Global Revenue Stream
Deepdub and Spikes partner to turn one piece of content into a global revenue stream.
- Zed — Zed v1.26.2
Added Pyrefly as an optional built-in Python language server, nested variable inspection in debugger hovers, and DWARF debugging for WebAssembly sources.
- AssemblyAI — Dictation API: Send Your Own LLM Request or Saved Prompt With llm
Dictation API accepts llm requests with your messages or prompt, sampling settings and output limits.
- Rox — Rox Dialer 2.0
Choose your outbound number and start power dialing from a selected task in your call list. Review contact details, take notes, and log outcomes.
- Resemble AI — Custom Investigators via API with structured output
Run your team's custom Investigators through the API and have them return results in a catalog agent's layout.
- Greptile — Computer use in TREX
TREX starts the services your change touches and uses computer use to test them end to end.
- LiteLLM — LiteLLM v1.106.6 and v1.105.7 fix the gzip performance regression
The proxy gzips only dashboard analytics responses, so Accept-Encoding: gzip no longer slows down other requests.
- Speakeasy SDK Generation — MCP gateways open up to every organization, with a tool picker and Code Mode
MCP gateways are available to every organization and add a consent-screen tool picker, Code Mode, and server hiding.
- Speakeasy SDK Generation — Tool Logs narrows as you filter, and admins get a Security & risk dashboard
Tool Logs filters narrow to available results; admins get a risk dashboard, and personal connections and server credential settings are added.
- Speakeasy SDK Generation — Platform MCP returns exact install steps for whichever client you use