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.
Copilot CLI v1.0.41 adds experimental MCP Tasks, --attachment flag, auto shell completions, and prompt-mode extension loading.
└──▷ GET THIS VERSION
$ git clone --branch v1.0.41 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:$ git checkout v1.0.41
└──▷ TRY IT
Attach a screenshot or document to a non-interactive Copilot prompt — useful in CI pipelines or scripts where you need to feed visual or binary content alongside a question.
$ copilot -p 'Explain the architecture shown in this diagram' --attachment ./architecture.png
Enable experimental MCP Tasks so long-running MCP tools execute as background agents you can track without blocking your session.
$ copilot --experimental
Allow project extensions and management tools to load in prompt mode — set before invoking a non-interactive Copilot command in CI.
$ GITHUB_COPILOT_PROMPT_MODE_EXTENSIONS=true copilot -p 'Run the project lint extension and summarise findings'
›Adds --attachment flag in non-interactive (-p/--prompt) mode to attach files (images or native documents) to the initial prompt.
›Adds experimental MCP Tasks support: MCP tools with taskSupport: "required" run as non-blocking background agents trackable via list_agents and read_agent — available when experimental mode is enabled via /experimental on or the --experimental flag.
›Extensions now load in prompt mode (-p); user extensions load by default, while project extensions and management tools require the GITHUB_COPILOT_PROMPT_MODE_EXTENSIONS=true environment variable.
›Shell completions (bash, zsh, fish) are automatically installed on first run and updated after copilot update.
›Slash command picker now searches command descriptions and underlines matched characters.
+5 moreshow less
›Remote session connection errors now show your logged-in account and tailored remediation steps.
›Memory tool confirmation prompt now shows the scope (repository or user) when requesting permission to store a memory.
›Markdown formatting now renders in ask-user prompt questions.
›Tab-completing slash commands that accept arguments now adds a trailing space automatically.
›@-mention completion supports ./ paths, no longer adds a trailing space on directories, and shows project files before workspace roots.
LangGraph SQLite checkpointer gains a public get_writes_history API and streaming delta channel history.
└──▷ GET THIS VERSION
$ git clone --branch checkpointsqlite==3.1.0a1 https://github.com/langchain-ai/langgraph.git
# already have the repo? check out this version:$ git checkout checkpointsqlite==3.1.0a1
›Adds public get_writes_history saver API for retrieving write history with reworked delta cadence.
›Overrides get_delta_channel_history with a streaming walk implementation for more efficient history retrieval.
$ git clone --branch sdk==0.3.14 https://github.com/langchain-ai/langgraph.git
# already have the repo? check out this version:$ git checkout sdk==0.3.14
└──▷ USE IT
Reduce response payload size when updating a thread — useful in high-throughput pipelines where the full thread object is not needed.
OpenClaw v2026.5.4 adds Gemini voice bridge for Google Meet, new auth/model/status CLI commands, OpenRouter caching, WhatsApp Newsletter targets, and tree-sitter shell explainer.
└──▷ GET THIS VERSION
$ git clone --branch v2026.5.4 https://github.com/openclaw/openclaw.git
# already have the repo? check out this version:$ git checkout v2026.5.4
└──▷ TRY IT
Audit saved per-agent auth profiles for a specific provider without exposing secret values.
$ openclaw models auth list --provider openai --json
Check Discord channel health including degraded transport and event-loop starvation signals before blaming the network.
$ openclaw status --deep
Run Slack live QA in a Crabbox VNC desktop and capture screenshots alongside QA artifacts.
$ pnpm openclaw qa mantis slack-desktop-smoke
›Adds Gemini realtime voice bridge for Google Meet/Twilio dial-in with paced audio streaming, backpressure-aware buffering, and barge-in queue clearing for snappier voice agents.
›Adds openclaw models auth list [--provider <id>] [--json] to inspect saved per-agent auth profiles without exposing secrets.
›Adds openclaw qa mantis slack-desktop-smoke to run Slack live QA inside a Crabbox VNC desktop with screenshot capture.
›Adds degraded Discord transport and gateway event-loop starvation signals to openclaw channels status and openclaw status --deep.
›Adds opt-in OpenRouter response caching via X-OpenRouter-Cache, X-OpenRouter-Cache-TTL, and cache-clear headers on verified OpenRouter routes.
›Adds a tree-sitter-backed shell command explainer for approval and command-review surfaces.
›Adds before_agent_finalize bounded retry instructions in the Plugin SDK so workflow plugins can request one additional model pass.
›Adds registerIfAbsent to the plugin runtime-state API for atomic keyed-store dedupe claims.
›Adds plugin-owned SessionEntry slot projection and scoped trusted-policy session extension reads to the Plugin SDK.
›Adds streaming.progress.render: "rich" for Block Kit progress drafts backed by structured progress line data in Slack.
›Adds agents.defaults.toolProgressDetail: "raw" and per-agent overrides to control compact vs. raw tool summaries in /verbose and progress drafts.
›Adds startup phase spans, active work labels, stale terminal bridge markers, and default sync-I/O tracing in pnpm gateway:watch for Gateway diagnostics.
›Annotates ClawHub 429 errors with the reset window from RateLimit-Reset/Retry-After and appends a sign-in hint for unauthenticated requests.
›Supports drive-absolute Docker bind sources on Windows sandbox while keeping blocked-path and allowed-root policy comparisons case-insensitive.
›Advertises Codex audio transcription in runtime and manifest metadata, routing active Codex chat models to the OpenAI transcription default.
›Preserves realtime.introMessage: "" so realtime Chrome joins can stay silent instead of restoring the default spoken intro.
›Adds an agent-first filter to the Control UI chat session picker with responsive layout across phone/tablet/desktop widths.
›Collapses consecutive duplicate text messages in the Control UI chat into one bubble with a count.
›Makes the New Job sidebar in the Control UI cron view collapsible to reclaim space for the jobs list.
›Shows the active agent name in Control UI dashboard breadcrumbs.
›Records browser long animation frame or long task entries in the debug event log for slow dashboard render attribution.
›Expands OpenRouter app-attribution categories to include coding, programming, writing, chat, and personal-agent usage on verified routes.
Get up and running with Kimi-K2.6, GLM-5.2, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
Ollama v0.23.1 adds Gemma 4 MTP speculative decoding on Mac, delivering 2x+ speed on 31B coding tasks.
└──▷ GET THIS VERSION
$ git clone --branch v0.23.1 https://github.com/ollama/ollama.git
# already have the repo? check out this version:$ git checkout v0.23.1
└──▷ TRY IT
Run Gemma 4 31B with speculative decoding on a Mac to get 2x+ faster completions on coding tasks.
$ ollama run gemma4:31b-coding-mtp-bf16
›Supports Gemma 4 MTP (Multi-token Processing) speculative decoding on Macs via the MLX runner, enabling 2x+ speed gains for the Gemma 4 31B model on coding tasks.
SGLang v0.5.11 adds DFLASH spec-decode, decode-side radix cache for PD disaggregation, LoRA on DeepSeek-V3/Kimi-K2, and 14+ new model architectures.
└──▷ GET THIS VERSION
$ git clone --branch v0.5.11 https://github.com/sgl-project/sglang.git
# already have the repo? check out this version:$ git checkout v0.5.11
└──▷ TRY IT
Monitor speculative-decoding efficiency per request using the new split metrics instead of the old aggregate accept_length.
$ # After upgrade, query your metrics endpoint — accept_length is gone:
curl http://localhost:30000/metrics | grep -E 'num_accepted_drafts|num_accepted_tokens'
Scrape SGLang's Prometheus metrics from a gRPC-mode server (previously only available in HTTP mode).
$ curl http://localhost:30000/metrics
›Splits accept_length metric into num_accepted_drafts and num_accepted_tokens for finer speculative-decoding observability.
›Adds --page-size support with speculative decoding on AMD ROCm (previously caused a memory access fault).
›Exposes raw KV cache pool token counts as Prometheus gauges.
›Adds a Prometheus metrics endpoint for gRPC mode.
›Adds HTTP sidecar endpoints and a FlushCache gRPC RPC for gRPC mode.
+33 moreshow less
›Surfaces pending token count in prefill log and get_load.
›Supports moe_dp_size = 1 paired with arbitrary attention_cp_size so MoE and attention parallelism can be tuned independently.
›New DFLASH speculative-decoding kernel, enabled across additional model backends and AMD ROCm.
›Speculative Decoding V2 (Spec V2) with overlap scheduling is now the default, reducing per-step CPU cost for EAGLE/MTP/DFLASH paths.
›Adds adaptive speculative_num_steps for EAGLE topk=1.
›Enables piecewise CUDA graph with speculative decoding.
›Adds OpenTelemetry tracing for speculative decoding, pipeline parallelism, and DiffGenerator.
›Adds decode-side radix cache (Decode Radix Cache) for prefill/decode (PD) disaggregation, recovering cache hit rates and TTFT savings for long shared prefixes.
›Adds incremental transfer for the Mooncake transfer engine in PD disaggregation.
›Adds NIXL heterogeneous TP KV transfer for non-MLA models.
›Adds NIXL Mamba state slice transfer for heterogeneous TP.
›Enables LoRA on DeepSeek-V3 MLA (including quantization) and Kimi-K2.
›Adds LoRADrainer to reduce high P99 TTFT under LoRA workloads.
›Adds decoupled LoRA MoE backend with Marlin support.
›Adds dual MoE CUDA graph capture for lora/nolora batches.
›Adds CSGMV kernel offline auto-tuning for LoRA.
›New FlashInferCuteDslMoE layer for the standard FP4 MoE path.
›Integrates community-contributed FA3 kernels alongside FA4 as a high-performance alternative.
›Adds NVFP4 KV cache with a quantization strategy abstraction and kernel.
›Adds GPTQ / AWQ 4-bit quantization on CPU.
›Adds GGUF quantization for Ascend NPU (dense and MoE models).
›Adds TP communications compression for Qwen3 on Ascend NPU.
›Adds pipeline parallelism support on Intel XPU.
›New SequenceClassification model architecture, powering the Score API.
›Adds day-0 support for Gemma 4, GLM-5.1, Qwen3.6, MiMo-V2.5/V2.5-Pro, Ling-2.6-Flash, Mistral Medium 3.5, Kimi-K2.6, Hunyuan v3 (preview), Voxtral, Parakeet, Moss-VL, and Qwen3-ASR (chunk-based streaming).
›Adds diffusion model support for LTX-2.3 (including two-stage and TI2V), ERNIE-Image, FLUX.2-small-decoder, JoyAI-Image-Edit, FLUX.1-dev ModelOpt NVFP4, Qwen Image ModelOpt FP8, and Stable Diffusion 3 medium.
›Adds ModelOpt diffusion FP8 support for Flux1/Flux2 and Wan2.2.
›Adds Standalone Rollout API, Denoising Environment Backpass, and SP-Aligned Log-Prob for T2I post-training in SGLang-Diffusion.
›Adds disaggregated diffusion support.
›Adds dynamic batching v0 for SGLang-Diffusion.
›Adds CPU platform support for SGLang-Diffusion.
›Enables in-memory loading for URL/base64 image inputs in SGLang-Diffusion (now default).
›Default CUDA version moves to 13.0 across SGLang, sgl-kernel, and Docker images; PyTorch upgraded from 2.9 to 2.11.
Phoenix 15.4.0 adds filter-based DELETE annotation endpoints, token counts in REST payloads, and vendor passthrough tool support.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v15.4.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v15.4.0
›Adds filter-based DELETE endpoints for span, trace, and session annotations via the REST API, enabling bulk deletion without enumerating individual IDs.
›REST API span/trace/session payloads now include token counts, surfacing LLM usage data directly in API responses.
›Adds set_time_range tool for agents with hardened context injection, enabling time-scoped agent queries.
›Adds vendor passthrough tools support, allowing tool definitions to be forwarded as-is to underlying model providers.
›Adds ToolPart styles and subcomponents for richer agent tool rendering in the UI.
+1 moreshow less
›Simplifies trace/span status icons and introduces a status badge in panel views.
Chroma 1.5.9 adds maxscore sparse search, sharded group-by, read-only failover, and a new Tilt fault-injection CLI.
└──▷ GET THIS VERSION
$ git clone --branch 1.5.9 https://github.com/chroma-core/chroma.git
# already have the repo? check out this version:$ git checkout 1.5.9
›Adds maxscore option in the collection schema, wiring a new MaxScore sparse-search index (backed by SparsePostingBlock) with SIMD acceleration through the metadata segment and search path for faster full-text-style scoring.
›Adds SparsePostingBlock, a maxscore writer/reader, and a batch-loading lazy cursor as foundational components of the new sparse posting index.
›Adds read-only backend failover to the Rust client (rust-client), enabling automatic fallback when the primary backend is unavailable.
›Enables group by support for sharded collections.
›Enables index rebuilds for sharded collections.
+8 moreshow less
›Adds a Tilt fault-injection CLI for chaos/fault testing of distributed deployments.
›Adds a spanner-cli wrapper binary for interacting with Spanner from the Chroma toolchain.
›Adds client-header propagation to the Gemini embedding functions.
›Adds MCMR (multi-collection/multi-region) support for log garbage collection, including GC of empty MCMR collections.
›Adds a workflow to build and publish service container images to both GitHub Container Registry (:1.5.9) and DockerHub (:1.5.9).
›Defers Spanner initialization in the log service to first use, reducing startup latency.
›Names and sizes all worker threads for improved observability in system diagnostics.
›Sealing a shard now redistributes lower offset IDs to the previous active shard.
THE AI TOOLCHAIN