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.
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
Claude Code v2.1.157 adds auto-loaded plugins, plugin scaffolding, richer telemetry, and smarter agents dispatch.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.157 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.157
└──▷ TRY IT
Bootstrap a new plugin without touching any marketplace — scaffold it locally and it loads automatically on next launch.
$ claude plugin init my-recon-tool
Capture detailed tool parameters (bash commands, MCP/skill names) in OTEL traces for audit or pipeline debugging.
$ OTEL_LOG_TOOL_DETAILS=1 claude --print 'Run nmap on 10.0.0.0/24 and summarize open ports'
›Adds automatic loading of plugins from .claude/skills directories — no marketplace required.
›New claude plugin init <name> command scaffolds a new plugin directly into .claude/skills.
›Adds autocomplete for /plugin arguments: subcommands, installed plugin names, and plugins from known marketplaces.
›Honors the agent field in settings.json for dispatched sessions, with --agent <name> to override per-run.
›EnterWorktree can now switch between Claude-managed worktrees mid-session.
+5 moreshow less
›tool_decision telemetry events now include tool_parameters (bash commands, MCP/skill names) when OTEL_LOG_TOOL_DETAILS=1 is set.
›Claude-managed worktrees are left unlocked after an agent finishes, enabling clean git worktree remove/prune workflows.
›Adds a 'Workflow keyword trigger' setting in /config to prevent the word 'workflow' in a prompt from triggering a dynamic workflow.
›Slash-command autocomplete in the claude agents dispatch input now matches substrings.
›Pressing backspace immediately after a workflow trigger keyword dismisses the workflow request instead of deleting a character.
Autonomous coding agent as an SDK, IDE extension, or CLI assistant.
Cline CLI v3.0.15 adds Cline Hub web monitoring, global AGENTS rules, Discord mute controls, and three new models.
└──▷ GET THIS VERSION
$ git clone --branch cli-v3.0.15 https://github.com/cline/cline.git
# already have the repo? check out this version:$ git checkout cli-v3.0.15
›Adds Cline Hub, a web app for monitoring connected clients, viewing and driving sessions, streaming assistant output, and restarting the local hub — with local, LAN, and tunnel access gated by a room secret.
›Supports global AGENTS rules so agent rules apply across all sessions, not just per-project.
›Lets plugins contribute static or dynamic rule content when installed in the sandbox.
›Binds Discord sessions to individual message authors so different Discord users no longer share chat state in a thread.
›Supports /mute and /unmute commands in Discord, resolved from user mentions or raw user IDs to mute a specific participant in a thread.
+3 moreshow less
›Makes OAuth URLs clickable in the TUI.
›Discovers SDK skill directories that are symlinked, including handling circular symlinks.
›Steers active connector sessions across turn keys by matching on session ID so replies continue the existing session instead of starting a duplicate.
$ git clone --branch v1.0.56 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:$ git checkout v1.0.56
└──▷ TRY IT
Enable or disable the rubber duck agent without editing config files manually.
$ copilot config set builtInAgents.rubberDuck true
›Adds builtInAgents.rubberDuck setting to enable or disable the rubber duck agent via copilot config.
›MCP tools that return both content text and a structuredContent payload now surface both to the agent; literal JSON serialization (per MCP spec §5.2.6) is deduplicated, otherwise the two are concatenated.
›When gh CLI is on PATH, the GitHub MCP server now omits redundant gh-replaceable tools by default, reducing token usage.
›Free and Student users can select models other than Auto in the model picker.
›Reasoning effort picker respects model capabilities, hiding options unsupported by the current model.
+3 moreshow less
›Diff view uses a continuous scroll layout with sticky file and hunk headers, full terminal width, and theme-aware colors.
›web_fetch tool prefers markdown content when available, using HTTP content negotiation for cleaner results from documentation sites.
›Context window tier selection now persists durably in session events and survives SDK-only resume paths, reapplying tier-derived limits to request, compaction, and truncation logic.
Qwen Code v0.17.0 adds a Feishu/Lark channel adapter, zero-config computer-use via MCP, and PermissionDenied hooks for AUTO classifier blocks.
└──▷ GET THIS VERSION
$ git clone --branch v0.17.0 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout v0.17.0
›Emits PermissionDenied hooks when the AUTO classifier blocks a tool, enabling downstream integrations to react to permission decisions.
›Adds a Feishu (Lark) channel adapter for routing Qwen Code interactions through Feishu/Lark.
›Enables zero-config computer-use capability built in via the open-computer-use MCP — no manual setup required.
›Lays telemetry foundation for skill-based real-time optimization (P0+P1 priority tiers).
└──▷ BREAKING ON UPGRADE
!Context compaction strategy replaced: tail-preservation compaction is removed and replaced with summary + restoration attachments, which may change how long conversation contexts are truncated on upgrade.
Hermes v2026.5.29 adds explicit Docker insecure opt-in, full 19,932-entry skills catalog, and memory providers now receive completed-turn context.
└──▷ GET THIS VERSION
$ git clone --branch v2026.5.29 https://github.com/NousResearch/hermes-agent.git
# already have the repo? check out this version:$ git checkout v2026.5.29
└──▷ TRY IT
Run Hermes in Docker with LAN access while explicitly opting into insecure dashboard mode — required after the bind-host inference removal.
$ docker run -e HERMES_DASHBOARD_INSECURE=1 -p 0.0.0.0:8080:8080 nous-research/hermes-agent
›Adds HERMES_DASHBOARD_INSECURE=1 env var as an explicit opt-in to disable the dashboard's loopback auth guard in Docker, replacing silent bind-host inference.
›Expands the skills.sh catalog from 858 to 19,932 entries by walking the full sitemap instead of a single paginated page.
›Memory providers now receive completed-turn message context, enabling richer recall across sessions.
›Narrows hindsight recall_types default to observation-only, giving more targeted hindsight behavior out of the box.
└──▷ BREAKING ON UPGRADE
!Docker setups that previously relied on bind-host inference to enable --insecure must now explicitly set HERMES_DASHBOARD_INSECURE=1 or the dashboard's same-origin guard will remain active.
llama.cpp b9415 adds a skip_download option to prevent automatic model downloads.
└──▷ GET THIS VERSION
$ git clone --branch b9415 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b9415
›Adds skip_download flag to the download subsystem, allowing operators to prevent automatic model fetching — including when the target file does not exist locally.
llama.cpp b9411 adds DeepSeek V3.2 model support with sparse attention and NVFP4 quantization.
└──▷ GET THIS VERSION
$ git clone --branch b9411 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b9411
›Adds support for the DeepseekV32ForCausalLM architecture (LLM_ARCH_DEEPSEEK32) with a generic DeepSeek Sparse Attention (DSA) implementation including a lightning indexer cache.
›Adds convert tooling support for the DeepseekV32ForCausalLM architecture to enable GGUF conversion of DeepSeek V3.2 model weights.
›Adds NVFP4 quantization support for DeepSeek V3.2 models.
›Adds llama_kv_cache_dsa memory type combining KV cache with a lightning indexer cache to accelerate DSA inference.
›Adds f16 support for the GGML_OP_FILL operation in the ggml backend.
›Adds LLAMA_ARG_API_KEY_FILE environment variable as an alternative to passing --api-key-file directly, letting operators supply API keys via file path through the environment.
llama.cpp b9389 automatically applies the iGPU flag for integrated CUDA/HIP devices
└──▷ GET THIS VERSION
$ git clone --branch b9389 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b9389
›Automatically applies the integrated GPU (iGPU) flag for CUDA and HIP backends when an integrated device is detected, removing the need for manual configuration.
llama.cpp b9387 adds per-quantization MMQ dispatch thresholds on AMD MFMA hardware, yielding up to +76% prompt throughput.
└──▷ GET THIS VERSION
$ git clone --branch b9387 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b9387
›Introduces ggml_cuda_should_use_mmvq(type, cc, ne11) to apply per-quantization MMVQ/MMQ batch thresholds on AMD CDNA (MFMA) hardware, replacing the single global MMVQ_MAX_BATCH_SIZE = 8 cutoff: K-quants Q3_K/Q4_K/Q5_K switch to MMQ at batch≥4, Q2_K/Q6_K at batch≥6, legacy and IQ quants retain the original threshold — delivering up to +76% pp512 throughput on MI250X.
›Non-AMD-MFMA paths (NVIDIA, RDNA, CDNA1 without MFMA) are unchanged by this dispatch logic update.
›Adds --cpu-distributed-timeout-seconds CLI flag for CPU-distributed deployments.
›Adds thinking_token_budget parameter support to the Completions API.
›Adds reasoning_effort mapping to enable_thinking in the Completions API.
›Adds chat_template_kwargs support to the Responses API.
›Adds reset_cache() method to the KV offloading framework.
+53 moreshow less
›Adds update_config support to Model Runner V2.
›Adds logprob_token_ids field support in Model Runner V2.
›Adds X-data-parallel-rank header forwarding for data-parallel serving.
›Adds API-key authorization for /v2 endpoints.
›Adds AuthenticationMiddleware path extraction simplification and utility call-ID newtype to the Rust frontend.
›Adds PoolingOfflineMixin split for the pooling offline API.
›Adds QuantKey with activation override, reworking quantization_config in the quantization framework.
›Adds ExpertMapManager as part of the MoE refactor, with experts moved to the experts/ directory and a RoutedExperts alias for FusedMoE.
›Adds truncation-side control for OpenAI-compatible endpoints.
›Adds reasoning_content → reasoning normalization in the frontend.
›Adds consolidated Speech-to-Text entrypoints.
›Adds beam-search consolidation via BeamSearchMixin.
›Adds score/rerank chat-template instructions to the frontend.
›Introduces experimental Rust frontend integration with in-tree code, a DP Supervisor for data-parallel serving.
›Introduces multi-tier KV cache offloading framework with a Python filesystem secondary tier, DSv4 support, and Mooncake disk offloading via MooncakeStoreConnector.
›Adds DeepSeek V4 model package reorganized into dedicated vllm/models/deepseek_v4/ package with NVFP4 fused MoE, full and piecewise CUDA graph, and MTP speculative decoding.
›Model Runner V2 (MRv2) is now default for Qwen3 dense models, with automatic fallback to MRv1 for unsupported features.
›Adds batch-invariant inference with Cutlass FP8 support for a 28.9% end-to-end latency improvement and an NVFP4 Cutlass linear path.
›Adds CutlassFP8 padding pre-processing yielding a 13.5% TTFT improvement.
ONNX Runtime ships the WebGPU Execution Provider as a standalone plugin (onnxruntime_providers_webgpu) compatible with ORT 1.24.4+.
└──▷ GET THIS VERSION
$ git clone --branch plugin-ep-webgpu/v0.1.0 https://github.com/microsoft/onnxruntime.git
# already have the repo? check out this version:$ git checkout plugin-ep-webgpu/v0.1.0
›Introduces the onnxruntime_providers_webgpu plugin EP, registered into an existing ONNX Runtime 1.24.4+ installation at runtime via onnxruntime.register_execution_provider_library(...) (Python) or OrtEnv.RegisterExecutionProviderLibrary(...) (.NET) — no core ORT package swap required.
›Adds the onnxruntime-ep-webgpu Python wheel (PyPI: onnxruntime-ep-webgpu) for installing the WebGPU EP alongside the onnxruntime package.
›Adds the Microsoft.ML.OnnxRuntime.EP.WebGpu NuGet package for .NET, referenced alongside Microsoft.ML.OnnxRuntime.
›Ships cross-platform native binaries for Windows x64/arm64 (bundled with dxil.dll and dxcompiler.dll), Linux x64, and macOS arm64.
›Provides WebGPU kernels for transformer and LLM contrib ops: Attention, MultiHeadAttention, GroupQueryAttention, RotaryEmbedding, SkipLayerNormalization, SkipSimplifiedLayerNormalization, SimplifiedLayerNormalization, BiasAdd, BiasGelu, BiasSplitGelu, FastGelu, Gelu, QuickGelu, CausalConvWithState, and LinearAttention.
+4 moreshow less
›Supports quantized inference via MatMulNBits (with DP4A and subgroup-matrix acceleration paths), DequantizeLinear, GatherBlockQuantized, and QMoE (quantized Mixture-of-Experts) on WebGPU.
›Includes a FlashAttention kernel and Intel-optimized MatMul/Gemm code paths for improved throughput on Intel GPUs.
›Adds program caching to amortize WebGPU shader compilation costs across inference runs.
›Adds optional PIX frame capture and WebGPU profiler integration for GPU performance investigation.
THE AI TOOLCHAIN