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.
$ git clone --branch v2.6.20 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:$ git checkout v2.6.20
└──▷ USE IT
Enable native structured outputs for a LiteLLM provider that supports the feature natively.
python
from agno.models.litellm import LiteLLM
model = LiteLLM(
model="openai/gpt-4o",
supports_native_structured_outputs=True,
)
Read citations returned from an OpenAI web-search response to attribute sources in your application.
python
from agno.models.openai import OpenAIChat
from agno.agent import Agent
agent = Agent(model=OpenAIChat(id="gpt-4o"))
response = agent.run("What happened in AI news today?")
print(response.citations)
›Enables supports_native_structured_outputs and supports_json_schema_outputs per-provider flags on LiteLLM to activate native structured outputs and JSON schema outputs.
›Surfaces web-search citations on response.citations for OpenAIChat and OpenAILike providers.
›Adds ClickHouseDB as a backend for high-volume trace ingest and OLAP scans.
›Adds a new Scavio search toolkit integration.
›Removes the hard cap on quick_prompts (previously limited to 3) per agent, team, or workflow in AgentOS.
+1 moreshow less
›Supports FastAPI >= 0.137 so get_routes() lists every registered route.
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 v2.1.195 adds mouse-disable env var for fullscreen mode and improves remote session startup UX.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.195 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.195
└──▷ TRY IT
Suppress accidental mouse clicks and drags in a fullscreen Claude Code session (e.g. on a touch-enabled display or when embedding in a kiosk) while keeping scroll working.
$ CLAUDE_CODE_DISABLE_MOUSE_CLICKS=1 claude
›Adds CLAUDE_CODE_DISABLE_MOUSE_CLICKS env var to suppress mouse click/drag/hover in fullscreen mode while preserving wheel scroll.
›Improves claude agents completed list to fill available vertical space, with a compacting header on short terminals so live sessions stay visible.
›Improves remote session startup with a provisioning checklist displayed while the container initializes.
›Improves Linux voice mode to distinguish between 'no microphone detected' and 'SoX not installed' when SoX is present but no capture device exists.
└──▷ BREAKING ON UPGRADE
!Hook matchers now exact-match hyphenated identifiers (e.g. code-reviewer, mcp__brave-search) instead of substring-matching — existing hooks that relied on partial matches will stop firing; use mcp__brave-search__.* to match all tools from a hyphenated MCP server.
Autonomous coding agent as an SDK, IDE extension, or CLI assistant.
Cline v4.0.0 ships SDK-backed task execution, ClinePass subscriptions, a Customize marketplace for Skills/MCP/Plugins, and queued chat prompts.
└──▷ GET THIS VERSION
$ git clone --branch v4.0.0 https://github.com/cline/cline.git
# already have the repo? check out this version:$ git checkout v4.0.0
›Adds SDK-backed VS Code extension runtime covering agent turns, tools, Plan/Act mode, MCP, checkpoints, telemetry, compaction, and task history.
›Introduces ClinePass: in-extension subscription onboarding, provider selection, live model lists, entitlement/org error states, and out-of-credit prompts.
›Launches the Customize marketplace for discovering, installing, enabling, and uninstalling Skills, MCP servers, and Plugins directly from the extension.
›Adds Cline Plugins: extend Cline with custom tools, workflows, skills, and MCP-powered capabilities installable from the Customize marketplace.
›Supports queued prompts — messages submitted while Cline is working are queued, shown during the active turn stream, and cancellable before execution.
+6 moreshow less
›Adds edit-and-regenerate support for previous user messages with Reset Chat and Reset Code actions.
›Adds generic SDK provider settings and model-catalog support with shared model picker, reasoning controls, dynamic model IDs, and custom model handling.
›Expands model support: Fireworks GLM 5.2, Kimi K2.6 Fast, Kimi K2.7 Code, Qwen 3.7 Plus, MiniMax M3, SAP AI Core, LiteLLM model fetching, and Codex OAuth credentials.
›Adds MCP support for plugins with shared marketplace install/uninstall plumbing.
›Adds automatic MCP hub refresh after marketplace installs so new servers are immediately available without a manual restart.
›Reworks Plan/Act mode through SDK coordinators with automatic continuation when switching from Plan to Act.
└──▷ BREAKING ON UPGRADE
!Command auto-approval is now disabled by default for new and reset configurations — existing setups that relied on auto-approval for commands will need to re-enable it.
!Subagents are temporarily disabled in the VS Code extension.
!The legacy Explain Changes feature has been removed.
!MCP files and legacy formats are migrated into the shared settings file — manual MCP config files from prior versions are consolidated automatically.
!Provider and model configuration is reworked around providers.json and the model catalog — provider-specific settings views are replaced with shared generic components.
OpenHands 1.40 adds full Git history cloning, a default sandbox spec setting, Azure DevOps webhook setup, and richer Jira DC integration.
└──▷ GET THIS VERSION
$ git clone --branch cloud-1.40.0 https://github.com/All-Hands-AI/OpenHands.git
# already have the repo? check out this version:$ git checkout cloud-1.40.0
└──▷ HOW TO FIND IT
Pin a specific sandbox environment for all your agent runs so you don't have to reconfigure it per session.
📍In the console, go to Settings › User Settings and set the Default Sandbox Spec to your preferred environment via the default_sandbox_spec_id field.
Enable full Git history cloning when your agents need to analyze commit history or run git-log-based tools.
📍In the console, go to Settings › User Settings and enable the Full Git History option.
Set up org-level Azure DevOps webhooks in one click to automatically trigger OpenHands resolvers on mentions.
📍In the console, go to Settings › Integrations › Azure DevOps and use the one-click webhook setup under your organization.
›Adds a user setting to enable full Git history cloning (instead of shallow clones) for agents that need deep commit context.
›Adds default_sandbox_spec_id to user settings, letting users pin a preferred sandbox environment across sessions.
›Adds an enterprise admin endpoint for programmatic user provisioning.
cua-driver-rs v0.6.8 ships relative-coordinate support, zero-config computer-use via MCP, Feishu channel, --worktree flag, and post-tool-batch hooks.
└──▷ GET THIS VERSION
$ git clone --branch cua-driver-rs-v0.6.8 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout cua-driver-rs-v0.6.8
└──▷ TRY IT
Enable relative-coordinate mode in the CUA driver to use normalised coordinates (0–1000 scale) instead of absolute pixels.
›Enables relative-coordinate mode in the CUA driver via CUA_DRIVER_RS_COORDINATE_SPACE=1 (default 0); optional scale tuning with CUA_DRIVER_RS_COORDINATE_SCALE=1000.
›Adds --worktree startup flag (Phase D) with symlinkDirectories support and PR refs for isolated worktree sessions.
›Adds respectUserColors and hideContextIndicator options for the statusline.
›Adds auto-dream and auto-skill defaulting to on, plus a /memory toggle command.
›Adds /compress instructions, PreCompact hook plumbing, and plan/subagent attachments to the auto-compact follow-up workflow.
+26 moreshow less
›Adds post-tool-batch hooks (post tool batch hooks) for running actions after a full tool batch completes.
›Adds user prompt expansion hooks.
›Adds PermissionDenied hook emission for AUTO classifier blocks.
›Injects context environment variables — session, agent, and prompt ID — into shell subprocesses.
›Adds CPU profiling support for Chrome DevTools analysis.
›Adds AUTO mode denial observability and caps.
›Auto-dumps memory diagnostics to disk on pressure detection.
›Adds zero-config built-in computer-use via the open-computer-use MCP server.
›Ships a codesigned and notarized @qwen-code/open-computer-use fork as a universal binary with QwenCuaDriver.app on macOS; unsigned x86_64 + arm64 binaries for Linux (glibc 2.31 floor) and Windows.
›Adds Feishu (Lark) channel adapter.
›Adds MiniMax-M3 model to the searchable model setup.
›Adds standalone auto-update support.
›Adds release-asset verification to the installer, switching public docs to a standalone entrypoint.
›Adds a virtual viewport for long conversations on Ink 7.
›Adds background housekeeping for stale file-history directories.
›Adds a simplify bundled skill to core.
›Adds agent reproduction workflows to skills.
›Adds a triage skill for issue/PR gatekeeping.
›Adds a fork subagent feature gate with 'Don't peek / Don't race' prompt discipline.
›Adds a settings JSON corrupted warning dialog in the CLI.
›Adds retry visibility for qwen-code.llm_request telemetry (Phase 4b).
›Adds background-shell completion notifications.
›Improves hooks matcher display.
›Enhances system prompts with global reasoning discipline, iterative planning, dedicated tool priority, and step-by-step communication guidance.
└──▷ BREAKING ON UPGRADE
!The refactor(core): tail-preservation compaction is replaced with summary + restoration attachments, which changes how context is compacted on resume.
›Adds --ep-dispatch-algorithm=lp to enable the new LPLB linear-programming load balancer for MoE expert parallelism, balancing token routing across redundant expert replicas by solving a per-layer LP; default behavior unchanged.
›Adds --moe-runner-backend flashinfer_trtllm_routed to enable NVFP4 MoE quantization for DeepSeek-V4 on Blackwell GPUs for higher MoE throughput.
›Adds --flashinfer-allreduce-fusion-backend flag (values: auto, trtllm, mnnvl) to control the FlashInfer fused allreduce + residual + RMSNorm MNNVL backend.
›Adds SGLANG_ENABLE_WAR_BARRIER environment variable to force-enable the overlap scheduler WAR barrier on non-CUDA platforms such as AMD ROCm.
›Adds support for online NVFP4 quantization via new online nvfp4 quantization path.
+41 moreshow less
›Adds support for asymmetric compressed-tensors MoE.
›Adds Mooncake group semantics for disaggregated serving parallelism.
›Adds bucketed multi-directory layout for NIXL file storage.
›Adds get_parallel() structured accessor for parallel-topology state.
›Adds graceful scheduler shutdown with hisparse host buffer freed on exit.
›Adds support for GLM-4.7 function calling via structural tags.
›Adds request/TTFT/worker metrics and a Grafana dashboard to the experimental sgl-router.
›Adds W8A8 MXFP8 quantization support for Qwen3 Dense on Ascend NPU.
›Adds Ascend NPU support for DeepSeek-V4.
›Adds Gemma4 Sliding Window Attention support on the Ascend backend.
›Adds LoRA support for DSA indexer LoRA targets for GLM-5.1 and DeepSeek-V3.2-family models.
›Adds rejection sampling for MTP speculative decoding.
›Adds sync-free fast_prefill_plan for EAGLE draft-extend CUDA graph.
›Adds FlashInfer CUDA graph support for EAGLE draft-extend.
›Adds sliding window attention draft layer support for Dflash.
›Adds piecewise CUDA graph support for Dflash.
›Adds CuteDSL prefill kernel on SM100 (Blackwell) for Kimi-Linear (KDA), delivering 1.08–1.52x faster prefill than the Triton path via a reusable scratch workspace.
›Adds int8 checkpoint pool for the linear-attention prefix cache (KDA/GDN models), substantially increasing prefix-cache capacity.
›Deduplicates speculative conv-window intermediate cache via a sliding-window layout, halving its footprint with no numerical change.
›Enables breakable CUDA graph execution on AMD ROCm/HIP.
›Migrates MSCCL++ to the upstream mscclpp Python package (Executor + DSL compiler) with auto-tuned collectives for TP=8 single-node and TP=16 two-node configurations.
›Adds context parallel strategy abstractions for new parallelism configurations.
›Adds prefill context parallel support for DeepSeek-V4 unified KV attention on AMD.
›Adds EPD disaggregated encode tracing.
›Adds MPServer and embedded server support for granian to enable multiple tokenizer workers.
›Adds Nemotron DP attention and MTP support for the hybrid Nemotron-H architecture.
›Adds GB10 FP8 fused MoE Triton configuration.
›Adds SwiGLU activation fused into gate gather_qmv for SwitchGLU MoE blocks on Apple Silicon (MLX).
›Adds MTP support for DeepSeek-V4 Online Compress.
›Adds MLA decode q-head padding to 64 under attention-TP, dispatching the cheaper FlashMLA head64 kernel instead of head128.
›Prewarms the MHC prenorm kernel at startup to eliminate first-run JIT slowdown.
›Supports BF16 mixed-dtype compression states on C4/C128 paths for DeepSeek-V4.
›Adds new model support for GLM-5.2, LiquidAI LFM2.5, Kimi-K2.7-Code, Poolside Laguna-M.1, DiffusionGemma, Zyphra ZAYA1, and MiMo-V2-ASR.
›Adds DeepSeek-V4 MTP acceptance length checks.
›Fuses sigmoid + mul attention output gate into a single Triton kernel on AMD.
›Enables fused GDN QKV split Triton kernel on HIP (AMD).
›Fuses sigmoid + mul into a single Triton kernel for shared expert gating on AMD.
›Adds LocalAttention for mistral3 encoder in diffusion models.
›Defers MoE finalize and fuses it with main-stream add for DeepSeek-V3.
›Fuses UE8M0 scale rounding directly into the FP8 group quantization kernel, removing a separate rounding pass.
llama.cpp b9813 adds Vulkan cooperative matrix support for Intel Xe-LPG Plus (Xe1-ARLH) GPUs
└──▷ GET THIS VERSION
$ git clone --branch b9813 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b9813
›Adds INTEL_XE1 architecture enum to the Vulkan backend and enables coopmat1 (cooperative matrix) acceleration on Intel Xe-LPG Plus (Xe1-ARLH) hardware, including a Windows driver compatibility check.
$ git clone --branch v2.70.0 https://github.com/triton-inference-server/server.git
# already have the repo? check out this version:$ git checkout v2.70.0
›Adds multi-device (multi-GPU) inference support to the TensorRT backend.
›Enables PyTorch 2 dynamic batching in the PyTorch backend.
›Adds an HTTP request body size limit to the OpenAI-compatible frontend.
›Hardens model environment unpacking and file-path handling against path traversal attacks.
›Changes the container install tree under /opt/tritonserver/ to root-owned, with the triton-server user accessing it via group/other permissions.
└──▷ BREAKING ON UPGRADE
!Windows server build is removed; Windows is no longer supported in the core build or documentation.
!The Python client now requires ml_dtypes.bfloat16 arrays for BF16 input/output tensors and no longer casts to/from np.float32; scripts passing BF16 data as np.float32 arrays must switch to ml_dtypes.bfloat16.
Phoenix client v2.10.0 adds a pytest plugin for eval CI, a direct server agent endpoint, session context for agents, and prompt route contexts.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-client-v2.10.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-client-v2.10.0
›Adds a pytest plugin for running eval CI directly from the phoenix-client library.
›Adds a direct server agent endpoint for communicating with agents.
›Adds session context support for agents, enabling stateful agent interactions.
›Adds prompt route contexts (pxi) for managing prompt routing.
›Adds experiment editing and eval skills for agents.
@composio/vercel 0.11.0 adds AI SDK 7 support, with peer dependency range now covering both v6 and v7.
└──▷ GET THIS VERSION
$ git clone --branch @composio/[email protected] https://github.com/ComposioHQ/composio.git
# already have the repo? check out this version:$ git checkout @composio/[email protected]
›Adds AI SDK 7 support to @composio/vercel; the ai peer dependency range is now ^6.0.0 || ^7.0.0, covering both v6 and v7 with e2e compatibility tests.
└──▷ BREAKING ON UPGRADE
!AI SDK 5 is no longer supported by @composio/vercel; the ai peer dependency range drops v5 and now requires ^6.0.0 || ^7.0.0.
Composio @composio/[email protected] adds first-class session and trigger APIs, MCP opt-in, and workbench config surfacing.
└──▷ GET THIS VERSION
$ git clone --branch @composio/[email protected] https://github.com/ComposioHQ/composio.git
# already have the repo? check out this version:$ git checkout @composio/[email protected]
└──▷ USE IT
Create a session with the hosted MCP endpoint available, now that MCP is opt-in.
›Adds composio.sessions.create() as the new first-class session creation API, with composio.create() retained as an alias.
›Adds connectedAccounts.updateAcl() as a stable alias for the experimental shared-connection ACL patch helper, previously only reachable via experimental.updateAcl().
›Makes MCP opt-in: sessions now return SessionWithoutMcp by default; pass { mcp: true } to sessions.create() to surface the hosted MCP endpoint on the type.
›Exposes Session.workbench populated from the API response on create, retrieve, attach, and update, including session.workbench?.enable, enabling callers to pass { workbench: { enable: false } } and detect sandbox state.
›Adds triggers.parse() to parse and optionally verify incoming webhook requests.
+2 moreshow less
›Adds triggers.setWebhookSubscription() to create or update a project webhook subscription from the TypeScript SDK.
›Accepts sandbox as the preferred key for session code-execution configuration, with workbench continuing to work as an alias.
└──▷ BREAKING ON UPGRADE
!The default sessions.create() / use() now returns SessionWithoutMcp — session.mcp is no longer present on the type unless the session is created with { mcp: true }. Code that reads session.mcp without passing { mcp: true } will encounter a type error.
THE AI TOOLCHAIN