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.
›Renames sandbox setting allowDevToolCaches to allowDevToolAccess in settings.json (and managed/MDM policy) — the old key is silently ignored, so an existing false opt-out reverts to the default (on).
›Moves sandbox auth settings from sandbox.gitAuth/sandbox.ghAuth to sandbox.auth.git/sandbox.auth.gh in settings.json; old keys are rejected as invalid rather than ignored.
›Adds worktreeBaseRef setting to control whether /worktree, /worktree new, and --worktree start from HEAD or the remote default branch (all three now default to HEAD).
›Adds /worktree new subcommand to start a new session in a new worktree.
›Adds --plan flag combinable with --mode autopilot to plan first and then implement without waiting for approval.
+23 moreshow less
›Adds /sandbox command to show effective sandbox paths, denials, and network access.
›Adds /app command to open the current session in the GitHub Copilot desktop app (requires GitHub Copilot app 1.1.3 or later).
›Adds pinnedPrompts setting (default false) to enable prompt pinning; automatically stays off on terminals under 30 rows unless set explicitly.
›Adds autoUpdate: true option on extraKnownMarketplaces entries in user settings to auto-update plugins at session start.
›Makes /model session-scoped by default; use /config model to set defaults for future sessions.
›Replaces ripgrep with tgrep (trigram-indexed grep) for fast regex search in large monorepos.
›Adds support for the kimi-k3 model in the model picker.
›Adds enterprise allow-auto-only policy support so /allow-all auto works while full allow-all remains blocked.
›Allows enterprise-managed sandbox policy to enforce a proxy URL while keeping credentials user-controlled.
›Agent Plugins spec plugins can now ship extensions under a com.github.copilot/extensions/ directory.
›Adds multi-session management via a Sessions tab and sidebar.
›Model picker groups models into Recent, Recommended, New, and other sections, with Shift+Tab switching grouping views.
›Sandboxed git now authenticates to Azure DevOps, GitHub Enterprise Server, GitLab, and other non-GitHub remotes with stored HTTPS credentials.
›Sandboxed wrapper builds (make and friends) receive dev tool caches based on build manifests in the working directory.
›Ask user multi-select prompts now include an Other option for free-text answers.
›Improves teleported subagent /tasks navigation with nested tree browsing, current/all and finished-task filters, and a live timeline.
›Tool directories on PATH inside a workspace (.venv/bin, node_modules/.bin, an in-repo GOPATH) no longer turn that workspace area read-only in the sandbox.
›Sandboxed commands can use UNIX-domain sockets on macOS, unblocking tools like tsx, vite, esbuild, and jest workers.
›Sandboxed commands work when the working directory lives on a Windows Dev Drive.
›Sandboxed commands can reach the network on recent Windows builds where outbound connections were previously blocked.
›A sandbox that cannot start an MCP server now fails in seconds instead of stalling the session.
›Supports prompt and shell command queuing in local sessions to run in order after the current task finishes.
›Adds (disabled) tags for inactive sandbox settings with explanations in the /sandbox configuration dialog.
└──▷ BREAKING ON UPGRADE
!The sandbox setting allowDevToolCaches is renamed to allowDevToolAccess in settings.json and managed/MDM policy; the old key is silently ignored, so an existing allowDevToolCaches: false opt-out reverts to the default (enabled). Rename it in settings.json and any managed/MDM policy.
!Sandbox auth settings keys moved from sandbox.gitAuth/sandbox.ghAuth to sandbox.auth.git/sandbox.auth.gh in settings.json; old keys are rejected as invalid. No automatic migration is provided.
Lightweight coding agent that runs in your terminal
Codex CLI alpha.6.6 adds Amazon Bedrock web search, expanded MCP namespace descriptions, and Agent Plugin runtime isolation.
└──▷ GET THIS VERSION
$ git clone --branch rust-v0.147.0-alpha.6.6 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.147.0-alpha.6.6
›Raises the MCP namespace tool-spec description limit from 1,000 bytes to 512 KiB, truncating only at a UTF-8 character boundary, so complete server instructions are preserved in tool-search source metadata.
›Enables cached web search for Amazon Bedrock: advertises hosted text web search while marking external live and indexed web access as unsupported, and resolves those modes to cached search (or disables the tool when cached search is prohibited by managed requirements).
›Bounds model-visible skill instructions, plugin instructions, MCP descriptions, schemas, individual tools, and the aggregate Agent Plugin MCP tool set within plugin manifests.
›Gives externalAgentConfig/detect its own global serialization queue so configuration reads, writes, and external agent imports remain responsive while detection is running.
›Skips symbolic links (and other non-file, non-directory entries) when copying a plugin into the store, allowing plugin installation to proceed without failing on symlinked skill files or executables.
+1 moreshow less
›Bounds Cursor project path resolution to a set of 128 path candidates using common filename separators instead of walking the full directory tree, rejecting ambiguous matches and unsafe encoded components.
Codex CLI adds paginated TUI history, dual-WebSocket code-mode transport, cached web search for Amazon Bedrock, and tool registry collision policies.
└──▷ GET THIS VERSION
$ git clone --branch rust-v0.148.0-alpha.6 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.148.0-alpha.6
›Adds --psp flag to app-server for process-scoped PSP routing of first-party ChatGPT requests.
›Adds tool registry collision policy configuration (#36954) and enforces strict tool name collision errors.
›Adds includeTurns reads for paginated threads to support incremental transcript access.
›Adds a dual-WebSocket transport for code mode, enabling a new host-communication path.
›Enables cached web search for Amazon Bedrock.
+16 moreshow less
›Enables remote compaction for Amazon Bedrock.
›Adds paginated transcript history in the TUI, allowing large session histories to load incrementally.
›Supports deferred loading for freeform tools and deferred custom tools in tool search.
Qwen Code v0.21.9 adds Qoder plugin installs, Local Control QR pairing, batch Skills API, and model.openAILogRetentionDays log cleanup.
└──▷ GET THIS VERSION
$ git clone --branch v0.21.9 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout v0.21.9
└──▷ USE IT
Set OpenAI API log retention to 30 days so logs older than a month are automatically cleaned up in the background.
json
{
"model.openAILogRetentionDays": 30
}
Get the review URL immediately after submitting a code review, suitable for piping into CI notifications.
$ qwen review submit | jq '.url'
›Adds model.openAILogRetentionDays setting (default: 7 days) to automatically purge expired OpenAI API logs in the background.
›Adds a daemon API endpoint and SDK helpers to enable or disable up to 100 Skills in a single batch request with per-target error reporting.
›Adds native support for installing Qoder plugins from directories, archives, Git repos, URLs, and npm packages, automatically loading system-prompt.md as extension context.
›The qwen review submit command now outputs the direct URL to posted reviews in both stderr and JSON stdout.
›Adds Local Control pairing to the CLI and Desktop app, enabling secure LAN access via QR code and a new Control menu option.
+7 moreshow less
›Web Shell now supports drag-and-drop of PNG, JPEG, GIF, WebP, and BMP images into the composer with attachment management and concurrency limits.
›Web Shell displays context window usage as a mini progress pill in the status bar with tooltips and accessible labels.
›Adds a fullscreen toggle to the Web Shell right panel for easier viewing of artifacts and subagent details.
›Desktop automatically creates a default workspace at ~/Documents/Qwen and starts the runtime on first launch without blocking the main thread.
›Adds a guard to pause tool execution after repeated failures within a single interactive ACP session.
›Stable Qwen Live Host packages are now mirrored to Aliyun OSS with automatic fallback to GitHub and support for up to one-hour archive downloads.
›Automatically assigns issues to area owners based on labels using a round-robin strategy that prefers owners with the fewest open assigned issues.
Get up and running with Kimi-K2.6, GLM-5.2, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
Ollama v0.32.8 adds Muse Glimmer model support across NVIDIA, AMD, and additional platforms.
└──▷ GET THIS VERSION
$ git clone --branch v0.32.8 https://github.com/ollama/ollama.git
# already have the repo? check out this version:$ git checkout v0.32.8
└──▷ TRY IT
Pull and run Muse Glimmer locally for interactive coding-agent use.
$ ollama run muse-glimmer
Connect Muse Glimmer to Claude Code as its local model backend.
$ ollama launch claude --model muse-glimmer
›Adds muse-glimmer model support on NVIDIA, AMD, and additional platforms (Apple Silicon was supported since v0.32.7), enabling local coding-agent and long-running personal-assistant workloads.
›Supports running muse-glimmer with ollama launch claude --model muse-glimmer to power Claude Code locally.
›Supports running muse-glimmer with ollama launch pi --model muse-glimmer for a lighter-weight coding-agent workflow.
›Supports running muse-glimmer with ollama launch openclaw --model muse-glimmer and ollama launch hermes --model muse-glimmer for personal-assistant frameworks.
›Adds muse-glimmer:30b-mlx model tag, enabling ollama run muse-glimmer:30b-mlx to download and run the 30B multimodal Muse Glimmer model locally on Apple Silicon via Ollama's MLX engine.
›Adds ollama launch subcommand to start coding agent and personal assistant frameworks (e.g. claude, pi, openclaw, hermes) directly against a local model via --model.
›MLX engine on Apple Silicon gains DFlash support and image input for Muse Glimmer as of this release.
An MCP that lets AI tools securely connect to your infrastructure, write IaaS code, debug issues, and assist during incidents - without risking production stability. Built for security teams to approve and infrastructure teams to experience like magic.
emisar v0.39.0 overhauls runbook versioning into a publish/release model with diff-on-confirm, MCP slug-keyed drafts, and broader secret redaction.
└──▷ GET THIS VERSION
$ git clone --branch v0.39.0 https://github.com/AndrewDryga/emisar.git
# already have the repo? check out this version:$ git checkout v0.39.0
›Adds get_runbook and update_runbook_draft MCP surface keyed by slug, so an agent replaces the single unpublished draft under the hash it read; publication remains human-only.
›Runbook publishing now renders a line-level diff of what changed from the live release before confirming, using the canonical content hash as the definition's identity.
›Introduces a single-draft model per runbook: only one unpublished change can exist at a time; the history list labels the live release on the Run button itself (e.g. Run v3) and marks a waiting draft with a dot.
›Running an older release now returns not_live instead of silently dispatching current content; running an unpublished draft requires explicit consent plus the hash of exactly what was read.
›Each execution snapshots the definition and release number it dispatched, making audit history immutable to later edits, publishes, or deletes.
+3 moreshow less
›Runner v0.19.0 expands secret redaction to cover connection strings, database URLs, key-derivation inputs (salt, pepper), cookie and session signing keys, and passphrase spellings — acting as a safety net beneath action-level redaction declarations.
›pfSense pack gains resolver, NTP, and WireGuard peer reads (secrets excluded), plus a DHCP reservation write staged for operator approval.
›Behavior harness in Packs now exercises a real service manager booted as PID 1, a per-case Docker daemon, iptables inside its own namespace, and a real dpkg database for install, remove, and autoremove scenarios.
└──▷ BREAKING ON UPGRADE
!The portal migration collapses all per-save version rows into a single runbook record, renumbers published versions into releases, and repoints execution history — this runs automatically before the instance serves traffic on upgrade.
llama.cpp b10355 enables multi-output backend sampling with token speculation support across CPU and GPU backends.
└──▷ GET THIS VERSION
$ git clone --branch b10355 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b10355
›Adds multi-output backend sampling, enabling GPU-accelerated sampling across multiple output sequences simultaneously, including support for token speculation workflows.
›Introduces a numeric context parameter declaring the maximum outputs per sequence, allowing fine-grained control over multi-output sampling capacity.
llama.cpp b10344 adds Multi-Token Prediction (MTP) support for the Nemotron Nano model.
└──▷ GET THIS VERSION
$ git clone --branch b10344 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b10344
›Adds mtp_flags configuration and Multi-Token Prediction (MTP) support for the Nemotron Nano model, enabling speculative decoding workflows with this architecture.
llama.cpp b10342 adds the Granite-Switch architecture: a dense Granite-4.1 model with per-token LoRA adapter switching via in-graph router attention.
└──▷ GET THIS VERSION
$ git clone --branch b10342 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b10342
›Adds graniteswitch arch registration (in llama-arch.h/cpp, llama-model.h/cpp) and src/models/granite_switch.cpp implementing per-token switched-LoRA inference via ggml_mul_mat_id over stacked lora_a/lora_b tensors.
›Adds conversion/granite.pyGraniteSwitchModel converter that stacks N LoRA adapters plus a zero base slot into per-projection A/B tensors and emits switch metadata into GGUF.
›Introduces llm_graph_input_switch in granite_switch.cpp providing stateless per-token adapter-index resolution via a single-head causal router attention layer (hparams.router_layer) stored in the KV cache, isolating concurrent requests automatically.
›Adds hparams.has_rope(il) predicate to skip k-shift RoPE on the router layer, preventing corruption of the literal-magnitude router K signal.
›Granite-Switch conversion honors --outtype for LoRA tensor precision instead of forcing F16.
+1 moreshow less
›Validates substitute (adapter) token IDs against n_vocab and bounds adapter count and LoRA rank from GGUF at load time, throwing on bad metadata.
vLLM v0.27.0 adds Kimi K3, Qwen3.5, FlashAttention 4 FP8/headdim-256 on SM100, fault-tolerance for DP+EP, gRPC control plane, and sm_107 Rubin support.
└──▷ GET THIS VERSION
$ git clone --branch v0.27.0 https://github.com/vllm-project/vllm.git
# already have the repo? check out this version:$ git checkout v0.27.0
└──▷ TRY IT
Run a Triton tensor-descriptor fused MoE kernel path for MoE models to improve kernel performance.
$ VLLM_TRITON_USE_TD=1 vllm serve <model>
›Adds --linear-backend flag honored for ModelOpt W4A16 quantization.
›Adds VLLM_TRITON_USE_TD environment variable to enable the Triton tensor-descriptor path for fused MoE kernels.
›Adds sample_from_anchor loaded from speculators config for speculative decoding.
›Adds has_pending_push_work to encoder-cache connectors for distributed disaggregated serving.
›Adds vllm-bench integrated into the vllm CLI for benchmarking.
+43 moreshow less
›New model support for Kimi K3 with full stack: core model, Python and Rust frontends, AttnRes kernels, DeepGEMM, DSpark AR fusion, and optional shared-expert sharding.
›New model support for Qwen3.5 text-only dense and MoE models with EVS video token pruning.
›New model support for K-EXAONE-2.0-750B-A37B.
›New model support for VaultGemma via the Transformers modeling backend.
›New model support for jina-embeddings-v5-text-nano with EuroBERT encoder backbone.
›Deepens FlashAttention 4 integration on SM100 with FP8 KV cache support and headdim-256 support.
›Adds new JIT warmup infrastructure and runner-owned Triton kernel warmup before the first request, removing first-request compilation stalls.
›Adds fault tolerance framework for DP+EP external load-balancer deployments.
›Adds async preparation for elastic EP scaling.
›Adds NIXL P/D disaggregation support for hybrid MLA+SSM models.
›Adds NIXL heterogeneous P/D block sizes for hybrid models.
›Adds gRPC control plane to the Rust frontend with engine-aware health reporting, abort control, server and model discovery, and KV event source discovery.
›Adds sm_107 target for NVIDIA Rubin with NVLink all-reduce paths on SM107.
›Enables ROCm gfx1250 architecture support.
›Expands Model Runner V2 to non-generative workloads: encoder-only attention, sequence pooling for embedding/classification, encoder token classification, encoder token embedding, and BGE-M3 pooling.
›Adds multi-layer MTP speculator to Model Runner V2.
›Adds multimodal inference on CPU via Model Runner V2.
›Adds per-request KV offload tier filtering via TierFilter/TierMatcher and self-describing KV events via TieringOffloadingSpec.
›Adds pluggable eviction policies via CachePolicyFactory for KV offloading.
›Adds CPUOffloadingSpec with single-copy MLA layout and migration onto SharedOffloadRegion.
›Adds generic P2P secondary tier with peer lookup and serving for KV offloading.
›Adds FP4 Qutlass integration for compressed-tensors quantization.
›Adds CuTeDSL MoE for ReLU2 NVFP4 quantization.
›Adds MXFP8 linear support in INC quantization.
›Adds AutoRound W4A16 MoE and MXFP4 linear/MoE on XPU.
›Adds KV quant mode for TurboQuant.
›Adds ModelOpt FP8 emulation on SM80.
›Adds compressed-tensors checkpoint support for DeepSeek-V4 and Kimi-K3.
›Adds top_k and top_p sampling for DiffusionGemma diffusion models.
›Adds weight version tagging for RL rollouts.
›Adds INT8 fused MoE kernel for Arm CPUs.
›Adds s390x inference optimization with oneDNN INT8 GEMM.
›Adds XPU QK Norm + RoPE fusion pass.
›Adds XPU FP8 o_proj with fp8_bmm and load-time scale transpose.
›Adds sequence parallelism for DeepSeek-V4 serving.
›Adds ReplaySSM caching for faster Mamba2 standard decode.
›Adds FlashInfer Mamba SSU algorithm selection.
›Adds FlexAttention epilogue hook.
›Adds audio model support to the Transformers modeling backend.
›Adds fused residual-add + RMSNorm compilation pass in the Transformers modeling backend.
›Upgrades to PyTorch 2.13.0 with torchvision 0.28.0 and Triton 3.7.1.
›Adds grammar advancement across the reasoning boundary with speculative decode for structured outputs.
└──▷ BREAKING ON UPGRADE
!PyTorch upgraded to 2.13.0 (along with torchvision 0.28.0 and Triton 3.7.1), which is a breaking environment change requiring environment rebuild for all platforms including XPU and CPU.
!FusedMoE is renamed to FusedMoEFactory in the MoE refactor.
Adds REST endpoints for experiment tag management and AI-powered natural-language session filtering.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v19.21.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v19.21.0
›Adds REST endpoints for setting tags on experiments, enabling programmatic experiment organization via the API.
›Adds AI-powered natural-language query support to the session filter in the UI, allowing practitioners to filter sessions using plain-language prompts.
Phoenix v19.20.0 adds dataset split REST endpoints and copy actions for session turns.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v19.20.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v19.20.0
›Adds REST endpoints to create, update, and delete dataset splits.
LanceDB v0.37.1 adds LSM table operations, custom FTS stop-words, blob range reads, job handles for index creation, and namespace/table existence checks.
└──▷ GET THIS VERSION
$ git clone --branch v0.37.1 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:$ git checkout v0.37.1
└──▷ USE IT
Apply custom stop-words to an FTS index so domain-specific noise terms are excluded from search.
THE AI TOOLCHAIN