Backfill issue -135 — The AI Toolchain

EDITIONS tail grep head diff uniq

The daily firehose — everything the toolchain shipped today, already filtered.

// HOW THIS ISSUE IS MADE

We read every release from the 174 tools on our watchlist at the source — GitHub and GitLab release notes, vendor release pages and changelogs, project blogs and feeds, vendor press releases, and the source code behind the tag. Bug-fix-only releases and non-product newsroom noise are dropped; what's left is summarized down to the new capability, how to try it, and any screenshots or videos the release itself published. Every entry links to the sources it was built from.

VIEW

ISSUE VIEW full issue

Do you prefer this view?

$ tct list   # 7 tools matched

AI & LLM Tooling

◆ AI Coding Agents

GitHub Copilot CLI

v1.0.19 NOTES

Copilot CLI v1.0.19 persists MCP state across sessions and adds OpenTelemetry streaming latency tracking.

└──▷ GET THIS VERSION

$ git clone --branch v1.0.19 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:
$ git checkout v1.0.19

›Adds github.copilot.time_to_first_chunk attribute to chat spans in OpenTelemetry monitoring for streaming latency visibility.
›/mcp enable and /mcp disable now persist across sessions, surviving restarts without re-configuration.
›Subagent spans now use INTERNAL span kind in OpenTelemetry, enabling more accurate trace categorization.
›Slash command timeline entries now include the command name (e.g., 'Review', 'Plan') for better audit context.

Was this useful?

SST OpenCode

v1.3.16 NOTES

The open source coding agent.

OpenCode v1.3.16 adds Azure model option support, ACP session config exposure, mouse capture disabling, and org-switching from the provider UI.

└──▷ GET THIS VERSION

$ git clone --branch v1.3.16 https://github.com/sst/opencode.git
# already have the repo? check out this version:
$ git checkout v1.3.16

└──▷ TRY IT

Disable mouse capture when running opencode inside a multiplexer like tmux where mouse events conflict.

$ OPENCODE_DISABLE_MOUSE=1 opencode

›Supports Azure model options on both chat and responses paths for @ai-sdk/azure users.
›Exposes session model and mode config options through ACP (Agent Control Protocol).
›Enables disabling TUI mouse capture via config or the OPENCODE_DISABLE_MOUSE environment variable.
›Labels Console-managed providers and adds org switching directly from the provider UI.
›Changes default Ctrl+Z binding on Windows to undo instead of terminal suspend.

Was this useful?

◆ Local LLM Runtimes

LocalAI

v4.1.2 NOTES

LocalAI v4.1.2 wires speculative decoding settings into the llama.cpp backend.

└──▷ GET THIS VERSION

$ git clone --branch v4.1.2 https://github.com/mudler/LocalAI.git
# already have the repo? check out this version:
$ git checkout v4.1.2

›Enables speculative decoding settings in the llama.cpp backend, unlocking faster inference via draft-model speculation.
›Adds Qwen3.5 model files to the model index.

Was this useful?

SGLang

v0.5.10 NOTES

SGLang v0.5.10 adds elastic failure tolerance, LoRA-on-MoE, native Apple Silicon, and a raft of new model and kernel capabilities.

└──▷ GET THIS VERSION

$ git clone --branch v0.5.10 https://github.com/sgl-project/sglang.git
# already have the repo? check out this version:
$ git checkout v0.5.10

└──▷ TRY IT

Control sparse MLA attention kernel activation threshold during DeepSeek V3.2 prefill to tune accuracy vs. speed trade-off.

$ SGLANG_NSA_DENSE_ATTN_KV_LEN_THRESHOLD=4096 python -m sglang.launch_server --model deepseek-ai/DeepSeek-V3 --tp 4

Include token usage stats in every streaming response for billing or monitoring without per-request opt-in.

$ python -m sglang.launch_server --model meta-llama/Llama-3-8B-Instruct --stream-response-default-include-usage

›Adds SGLANG_NSA_DENSE_ATTN_KV_LEN_THRESHOLD environment variable to control the KV length threshold at which the sparse MLA attention kernel is applied during prefill for DeepSeek V3.2.
›Adds --stream-response-default-include-usage server flag to control whether usage stats are included in streaming responses by default.
›Adds --strict-ports option for predictable, deterministic port assignment.
›Adds NetworkAddress abstraction for IPv6-safe address handling, enabling IPv6 dual-stack deployments.
›Enables piecewise CUDA graph capture by default, reducing memory overhead and improving throughput for models with complex control flow.

+34 moreshow less

›Integrates Elastic NIXL-EP (Elastic EP) for partial failure tolerance in DeepSeek MoE deployments — when a GPU fails, expert weights are redistributed from DRAM and serving continues without a full restart.
›Adds GPU staging buffer with dynamic ring allocator for PD disaggregation, gathering scattered GQA head slices into contiguous memory for bulk RDMA transfer (~1000x fewer RDMA requests, ~5x TPS/GPU at large concurrency on Qwen3.5 with Prefill TP4+Decode DEP4).
›Integrates HiSparse sparse attention backend for efficient long-context inference with sparsity-aware attention, also enabling HiSparse direct cache transfer from Prefill to Decode DRAM.
›Integrates FlashInfer MXFP8 kernels for GEMM and MoE operations, enabling mixed-precision FP8 inference with microscaling for RL and general workloads.
›Adds LoRA fine-tuning support for Mixture-of-Experts layers, including JIT alignment kernels, fused Triton kernels, TP support, CUDA graph support, and auto-detection of LoRA target modules.
›Upgrades from transformers 4.57.1 to 5.3.0, adding support for latest HuggingFace model architectures including GLM-5 on the main branch.
›Upgrades to the official Flash Attention 4 package, bringing Blackwell GPU support and the latest attention optimizations.
›Enables speculative decoding for the FA4 attention backend.
›Multi-modal attention now uses FA4 by default on SM100 (Blackwell) hardware for improved VLM performance.
›Enables context parallelism during prefill for multi-head attention models (e.g. Qwen3 MoE), distributing long sequences across GPUs.
›Adds Score API via EngineScoreMixin implementation.
›Adds subprocess liveness monitor to detect scheduler crashes.
›Adds direct model loading from object storage with RunAI Model Streamer.
›Adds MFU (Model FLOPs Utilization) metrics exposed via Prometheus.
›Enables IndexCache for DeepSeek V3.2, improving throughput by more than 10% on high-concurrency workloads.
›Adds TRT-LLM prefill/decode DSA kernels as the default for Blackwell (SM100/SM103) hardware.
›Enables NCCL/RCCL pre-warming to reduce P99 TTFT cold-start latency.
›Enables multi-thread weight loading by default.
›Adds CuTeDSL KDA decode kernel support and fused GDN projection Triton kernel for improved Qwen3.5 performance.
›Adds enhanced transformers modeling backend with full TP, PP, MoE, VLM support, and torch.compile compatibility.
›Renames the kernel package from sgl-kernel to sglang-kernel (v0.4.1), consolidating kernels and removing deprecated ops.
›Replaces decord with torchcodec for video decoding in VLM pipelines.
›Replaces soundfile+torchaudio with torchcodec AudioDecoder in load_audio for VLM audio ingestion.
›Adds chunk-aware ViT encoding with per-image cache and lazy device transfer for VLM workloads.
›Adds macOS platform support for diffusion models.
›Integrates Cache-DiT optimizations into the diffusers backend for improved diffusion performance.
›Adds NVFP4 support for Flux.2 diffusion model.
›Adds LTX-2 two-stage pipeline support for diffusion.
›Adds EPLB rebalance support for Kimi K2.5 in Elastic EP.
›Adds reasoning tokens usage reporting.
›Adds new model support: Nemotron-3-Super (bf16/fp8/nvfp4), Mistral Small 4 (Pixtral), LFM2-VL, Voxtral (speech-to-text), Helios, Hunyuan3D-2, LTX-2, MOVA, FireRed-Image-Edit.
›Adds AMD: FP8 prefill integration with radix cache path for DeepSeek models, MHA FP8-KV support, AMD MXFP4 support for Qwen3.5-397B-A17B, and FP8 KV cache + FP8 attention kernel for NSA on MI300/MI355 with TileLang.
›Adds CPU backend kernel apply_rotary_pos_emb_cpu for Qwen3-VL and Qwen3-Omni, and MXFP4 GEMM kernels for Intel AMX to support GPT-OSS series.
›Adds NPU/Ascend support for Kimi-K2.5-w4a8 and diffusion models (with enable_torch_compile), plus GLM-4.7-Flash on NPU.

└──▷ BREAKING ON UPGRADE

!The kernel package is renamed from sgl-kernel to sglang-kernel (v0.4.1); any install scripts, requirements.txt, or import paths referencing sgl-kernel or sgl_kernel must be updated.
!Transformers upgraded from 4.57.1 to 5.3.0; model configs or custom modeling code that relied on transformers 4.x APIs may require updates.
!decord is replaced by torchcodec for video decoding; environments with decord installed for VLM video pipelines must install torchcodec instead.
!soundfile+torchaudio replaced by torchcodec AudioDecoder in load_audio; environments relying on those libraries for audio ingestion in VLM pipelines must install torchcodec.

Was this useful?

llama.cpp

b8683 4 RELEASES · 2026-04-06 NOTES STABLE

llama.cpp b8683 adds MUL_MAT_ID support to the WebGPU backend.

└──▷ GET THIS VERSION

$ git clone --branch b8683 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8683

›Adds MUL_MAT_ID operation support to the WebGPU (ggml-webgpu) backend, enabling mixture-of-experts model inference via WebGPU.

3 more releases in this issue · 2026-04-06

b8682 NOTES STABLE

llama.cpp b8682 adds Q1_0 1-bit quantization support on CPU, enabling extreme model compression.

└──▷ GET THIS VERSION

$ git clone --branch b8682 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8682

›Adds Q1_0 1-bit quantization format (group size 128) with CPU support, including a generic fallback for x86 and other backends.

b8679 NOTES STABLE

llama-bench gains -fitc and -fitt flags for finer benchmark filtering control.

└──▷ GET THIS VERSION

$ git clone --branch b8679 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8679

└──▷ TRY IT

Filter llama-bench runs to specific test cases and test types to isolate prompt-ingestion or token-generation performance.

$ llama-bench -fitc <case-pattern> -fitt <type-pattern>

›Adds -fitc and -fitt arguments to llama-bench for filtering benchmark test cases and test types.

b8670 NOTES STABLE

llama.cpp b8670 adds HunyuanOCR vision model support with a perceiver-based projector and dedicated chat template.

└──▷ GET THIS VERSION

$ git clone --branch b8670 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b8670

›Adds HunyuanOCR vision projector (perceiver-based, with Conv2d merge) and tensor mappings for mm.before_rms, mm.after_rms, and related perceiver projector tensors.
›Adds a dedicated HUNYUAN_OCR chat template (content-before-role format) for HunyuanOCR models.
›Registers HunYuanVLForConditionalGeneration in convert_hf_to_gguf.py for both text and mmproj conversion.
›Supports the xdrope RoPE scaling type introduced by HunyuanOCR.

Was this useful?

◆ AI Agent Frameworks

OpenAI Agents SDK

v0.13.5 NOTES

OpenAI Agents SDK v0.13.5 adds callable approval policies for local MCP servers and a public flush_traces API.

└──▷ GET THIS VERSION

$ git clone --branch v0.13.5 https://github.com/openai/openai-agents-python.git
# already have the repo? check out this version:
$ git checkout v0.13.5

└──▷ USE IT

Flush all buffered traces immediately — useful in short-lived scripts or tests where the process may exit before traces are sent.

python

from agents import flush_traces

await flush_traces()

›Adds flush_traces as a public API to programmatically flush buffered trace data on demand.
›Supports callable approval policies for local MCP servers, enabling dynamic, code-driven control over tool-call approvals.

Was this useful?

OpenClaw

v2026.4.5 NOTES

Your own personal AI assistant. Any OS.

OpenClaw v2026.4.5 adds built-in video/music generation tools, ComfyUI integration, 12 new providers, Matrix exec approvals, multilingual Control UI, and experimental memory dreaming.

└──▷ GET THIS VERSION

$ git clone --branch v2026.4.5 https://github.com/openclaw/openclaw.git
# already have the repo? check out this version:
$ git checkout v2026.4.5

└──▷ TRY IT

Force-reinstall an existing plugin (e.g. after an update) without triggering the dangerous-code override prompt.

$ openclaw plugins install --force <plugin-or-hookpack-target>

Inspect prompt-cache efficiency and see exactly where cache breaks occur across a running agent session.

$ openclaw status --verbose

Manually trigger a dreaming REM preview run to see which memories would be promoted before committing them.

$ openclaw memory rem-harness

›Adds built-in video_generate agent tool so agents can create and return video media directly in replies.
›Adds built-in music_generate tool with bundled Google Lyria and MiniMax providers plus ComfyUI workflow-backed support, including async task tracking and follow-up audio delivery.
›Adds bundled comfy workflow media plugin for local ComfyUI and Comfy Cloud, covering image_generate, video_generate, and music_generate with prompt injection, reference-image upload, live tests, and output download.
›Adds bundled Qwen, Fireworks AI, StepFun, MiniMax TTS, Ollama Web Search, and MiniMax Search providers for chat, speech, and search workflows.
›Adds Amazon Bedrock Mantle support plus inference-profile discovery and automatic request-region injection for Bedrock-hosted Claude, GPT-OSS, Qwen, Kimi, GLM, and similar routes.

+19 moreshow less

›Adds Amazon Bedrock embeddings for Titan, Cohere, Nova, and TwelveLabs models with AWS credential-chain auto-detection for memory/search.
›Adds localized Control UI for Simplified Chinese, Traditional Chinese, Brazilian Portuguese, German, Spanish, Japanese, Korean, French, Turkish, Indonesian, Polish, and Ukrainian.
›Adds openclaw plugins install --force to replace existing plugin and hook-pack targets without using the dangerous-code override flag.
›Adds ClawHub search, detail, and install flows directly in the Skills panel.
›Adds iOS APNs exec approval notifications that open an in-app approval modal and clear stale state on resolution.
›Adds Matrix-native exec approval prompts with account-scoped approvers, channel-or-DM delivery, and room-thread aware resolution.
›Adds configurable contextVisibility per channel (all, allowlist, allowlist_quote) to filter supplemental context by sender allowlists.
›Adds shared model and media request transport overrides (headers, auth, proxy, TLS) across OpenAI-, Anthropic-, Google-, and compatible provider paths.
›Exposes OpenClaw tools to background Claude CLI runs via a loopback MCP bridge with stdin + stream-json streaming for live progress on long replies.
›Adds experimental structured plan updates and structured execution-item events for step-by-step agent progress in compatible UIs.
›Adds bundled xAI (grok-imagine-video), Alibaba Model Studio Wan, and Runway video providers with live-test and default model wiring.
›Adds experimental memory dreaming with weighted short-term recall promotion, /dreaming command, Dreams UI, multilingual conceptual tagging, and three cooperative phases (light, deep, REM).
›Adds configurable dreaming aging controls (recencyHalfLifeDays, maxAgeDays) and optional verbose logging for recall decay tuning.
›Adds REM preview tooling (openclaw memory rem-harness, promote-explain) and makes deep promotion replay-safe.
›Adds Dream Diary surface in Dreams UI with simplified user-facing config (enabled plus optional frequency).
›Adds prompt-cache break diagnostics and explicit cache reuse reporting in openclaw status --verbose.
›Enriches openclaw config schema JSON Schema output with field titles and descriptions for editor and agent consumers.
›Embeds ACP runtime directly in the bundled acpx plugin with a generic reply_dispatch hook, removing the external ACP CLI hop.
›Adds plugin-config TUI prompts to guided onboarding/setup flows.

└──▷ BREAKING ON UPGRADE

!The legacy config aliases talk.voiceId, talk.apiKey, agents.*.sandbox.perSession, browser.ssrfPolicy.allowPrivateNetwork, hooks.internal.handlers, and channel/group/room allow toggles are removed; use the canonical public paths and enabled instead (openclaw doctor --fix can migrate existing configs).
!Bundled CLI text-provider backends and the agents.defaults.cliBackends config surface are removed; ACP harness sessions and Gemini media understanding remain on native bundled providers.
!The Claude CLI backend and setup-token are removed from new Anthropic onboarding; existing anthropic:claude-cli profiles remain runnable but openclaw doctor will repair or remove stale state.

Was this useful?

// END OF TRANSMISSION

The AI Toolchain · curated daily · April 6, 2026
Releases & features from the open-source and commercial AI tools that practitioners actually run.

The AI Toolchain — issue -135, April 6, 2026

GitHub Copilot CLI

SST OpenCode

LocalAI

SGLang

llama.cpp

OpenAI Agents SDK

OpenClaw