Skip to content

MakerAI Suite v3.5.0 — Typed ModelConfig, Voice Suite & Provider Refresh

Latest
Latest

Choose a tag to compare

View all tags
@gustavoeenriquez gustavoeenriquez released this 01 Aug 22:39
· 55 commits to master since this release
v3.5.0
34d4ef3

MakerAI Suite v3.5.0 — Typed ModelConfig, Full-Duplex Voice & the August Provider Refresh

The biggest minor release since 3.3 — fully backward compatible: recompile and everything keeps working.

🎛️ Typed ModelConfig Channel

Capability configuration now lives in a single typed surface: ModelConfig.ModelCaps / SessionCaps / Tool_Active / ThinkingLevel moved out of the string-based Params/RTTI channel, with per-field user pins and transparent compatibility migration.

🎙️ Full-Duplex Voice Suite

  • TAiGrokRealtimeChat — xAI Grok Voice speech-to-speech: live transcription, streamed TTS, voice function calling (local functions + MCP), session resumption with turn replay, binary audio transport, ephemeral tokens
  • TAiOpenAiRealtimeTranslate — continuous streaming speech translation (one WebSocket per direction; see demo 071-VoiceBridgeTranslate)
  • gpt-transcribe / gpt-live-transcribe — OpenAI's Whisper successors integrated in TAiOpenAiAudio and TAiOpenAiRealtimeSTT (keywords, guided multi-language detection)
  • TAiRealtimeVoiceBase — shared full-duplex base; voice events flow through the universal TAiRealtimeConnection

🔄 August 2026 Provider Refresh — all 9 cloud providers, runtime-tested

Provider Highlights
Anthropic Claude 5 family, adaptive thinking + output_config.effort, FastMode, server-side compaction & fallbacks, web_search_20260209
OpenAI GPT-5.6 family, gpt-realtime-2.1, gpt-transcribe, DALL-E → gpt-image-1
Gemini 3.5/3.6 flash family, Nano Banana GA image models, gemini-embedding-2
Mistral Voxtral TTS (/v1/audio/speech) + OCR 4 (OcrIncludeBlocks)
Kimi K3 flagship (1M ctx, vision + reasoning); fix: new family rejects top_p
DeepSeek V4 flash/pro (1M ctx) with explicit thinking-mode control
Cohere Command A+ flagship (436K ctx, vision + always-on reasoning), north-mini-code, thinking capture
Groq qwen3.6-27b, orpheus TTS, retired-model cleanup
xAI Full catalog turnover: grok-4.3 (new default), grok-4.5, grok-build — with compatibility aliases

Retired models across all providers were removed or aliased to living equivalents, so existing code keeps running.

🎬 Grok Native Video & Image Generation

TAiGrokChat now generates video with grok-imagine (async job + polling + mp4 as TAiMediaFile, new VideoDurationSeconds property) and images with grok-imagine-image — activated by cmVideoGeneration/cmImageGeneration or the capability gaps.

🗄️ New: MSSQL Driver for RAG Vector

uMakerAi.RAG.Vector.Driver.MSSQL — SQL Server support for the vector RAG store via FireDAC.

🤖 Agents Hardening

Strict JSON graph validation, public RTTI mapper TAiToolParams, [TSecret] attribute, out_failure in conditional mode, and Compile no longer clears the Blackboard.

🔧 Framework Fixes

  • LastError now populated on every error path — synchronous callers can finally diagnose HTTP 4xx/5xx (previously: empty string, no exception)
  • Cohere synchronous tool-calling returned empty — fixed
  • ChatTools single surface with OnChange; ToolCall.ResMsg in streaming; FreeNotification audit

📦 Compatibility

  • Delphi 10.4 Sydney → 13.1 Florence (full support 11+)
  • MAKERAI_API_LEVEL = 35, no breaking changes
  • New demos: 037-MCPServerRAG, 071-VoiceBridgeTranslate

Full changelog: see the README changelog · Docs: https://makerai.cimamaker.com

Assets 2
Loading