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.
Agno v2.6.15 adds identity-aware, scoped MCP tool registration via a single MCPServerConfig object
└──▷ GET THIS VERSION
$ git clone --branch v2.6.15 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:$ git checkout v2.6.15
›Adds MCPServerConfig to configure the AgentOS MCP server (/mcp): register custom tools (plain callables or Agno @tool/Functions), scope built-ins with enable_builtin_tools=False, filter with include_tags/exclude_tags, inject the authenticated caller's JWT subject via a declared user_id parameter (hidden from the client schema), gate calls with an authorize function, and enable DNS-rebinding protection via allowed_hosts/allowed_origins — all in data, no custom middleware classes required.
HolmesGPT 0.33.0 adds cross-cluster remote tool execution and per-request tracing with span token metrics.
└──▷ GET THIS VERSION
$ git clone --branch 0.33.0 https://github.com/HolmesGPT/holmesgpt.git
# already have the repo? check out this version:$ git checkout 0.33.0
›Adds cross-cluster remote tool execution via an executor worker, toolset publishing, and caller headers, enabling tools to run against remote clusters from a central Holmes instance.
›Adds per-request tracing experiment routing and span token metrics to the server for observability into LLM completion costs per request.
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 v2.1.178 adds parameter-matching permission rules, smarter nested skill/directory scoping, and tighter subagent safety checks.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.178 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.178
›Adds Tool(param:value) syntax for permission rules to match on a tool's input parameters, with * wildcard support (e.g. Agent(model:opus) to block Opus subagents).
›Nested .claude/skills directories now load automatically when working in their directory; name clashes surface as <dir>:<name> so both skills remain accessible.
›Nearest-wins scoping for nested .claude/ directories: agent, workflow, and output-style configs closest to the working directory take precedence; project-scope workflow saves target the closest existing .claude/workflows/.
›Auto mode now evaluates subagent spawns through the classifier before launch, preventing a subagent from requesting a blocked action without review.
›Improves /doctor with a consistent flat tree layout, clearer section status icons, and highlighted command names.
+4 moreshow less
›Improves the skill listing truncation warning to show how many skill descriptions are affected.
›Changes the workflow prompt keyword to a purple shimmer highlight, triggering only on explicit phrases like "run a workflow" or "workflow:" rather than any mention of the word.
›Improves Remote Control error messages: connection failures show a persistent red "/rc failed" footer indicator, and "not yet enabled" errors now explain whether the cause is a gate, check failure, stale entitlement, or org policy.
›/bug now requires a description before submission and no longer uses model-refusal text as the GitHub issue title.
Qwen Code v0.18.1 adds durable cron jobs, project-scoped .mcp.json, Claude MCP server import, cross-platform Computer Use, and a wave of web-shell improvements.
└──▷ GET THIS VERSION
$ git clone --branch v0.18.1 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout v0.18.1
└──▷ USE IT
Configure a project-level MCP server so all contributors share the same server definition without touching global settings.
json
# Place at the root of your repository
# .mcp.json
{
"mcpServers": {
"my-server": {
"command": "npx",
"args": ["-y", "my-mcp-server"]
}
}
}
Run a workflow agent in an isolated git worktree so it cannot corrupt the main working tree while autonomously editing files.
›Adds isolation:'worktree' parameter to agent({schema, agentType, model, isolation:'worktree'}) in Workflow P3, enabling agents that run in isolated git worktrees.
›Adds project-scoped .mcp.json file for MCP server configuration, with workspace approval gating and aligned scope precedence.
›Adds feat(cli): import Claude MCP servers — practitioners can now import MCP server definitions from Claude directly into Qwen Code.
›Adds configurable screenshot max dimension for Computer Use via both a settings key and an environment variable.
›Adds durable /loop cron jobs that persist and survive daemon restarts.
+18 moreshow less
›Gates direct session shell behind an explicit opt-in, requiring deliberate configuration before the shell surface is exposed.
›Persists oversized tool results to disk instead of holding them in memory, enabling larger tool outputs without OOM.
›Bubbles background sub-agent permission prompts to the parent session so they can be approved interactively.
›Adds dedicated agent permission dialog via _meta.toolName for clearer per-agent permission handling in ACP.
›Introduces DaemonTransport abstraction with ACP standard compliance, exposing a stable SDK/serve interface.
›Delivers A2UI surfaces over MCP with a bridge extraction and action endpoint.
›Supports user-invocable frontmatter in skills files.
›Web-shell: revamps floating todo panel interactions.
›Web-shell: shows message timestamps on hover and per-task token and time detail on completed todos.
›Web-shell: reveals full tool detail with auto-collapse of finished tools.
›Web-shell: makes input shortcuts discoverable and clickable.
›Web-shell: adds collapsible TodoWrite history with status diff.
›Web-shell: collapses completed turns to prompt and final answer for a cleaner conversation view.
›Desktop app: shows git branch in the working directory badge.
└──▷ BREAKING ON UPGRADE
!Direct session shell is now gated behind an explicit opt-in; existing setups that relied on it being available by default will lose shell access until the opt-in is configured.
Qwen Code desktop-v0.0.4 adds zero-config computer-use, worktree support, Feishu channel, post-tool-batch hooks, and a /skills picker dialog.
└──▷ GET THIS VERSION
$ git clone --branch desktop-v0.0.4 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout desktop-v0.0.4
└──▷ TRY IT
Start a session in an isolated git worktree to safely run agent tasks in parallel without polluting the main branch.
$ qwen-code --worktree
Toggle memory (auto-dream/auto-skill) off mid-session when working with sensitive code you don't want persisted.
$ /memory
Customise the statusline to respect your terminal colour scheme and hide the context indicator for a cleaner UI.
llama.cpp b9654 adds a post-decode callback to the multimodal (mtmd) subsystem.
└──▷ GET THIS VERSION
$ git clone --branch b9654 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b9654
›Adds a post-decode callback to the mtmd (multimodal) subsystem, enabling callers to hook into the decode pipeline after each multimodal token batch is processed.
vLLM v0.23.0 adds a maturing Rust frontend, multi-tier KV offloading, unified parser, and broad new model support across backends.
└──▷ GET THIS VERSION
$ git clone --branch v0.23.0 https://github.com/vllm-project/vllm.git
# already have the repo? check out this version:$ git checkout v0.23.0
└──▷ TRY IT
Enable request-ID headers in the Rust frontend so downstream systems can correlate responses to originating requests.
$ vllm serve <model> --enable-request-id-headers
›Rust frontend adds a streaming generate endpoint, dynamic LoRA endpoints, /version and /server_info endpoints, a server-router extension hook, and --enable-request-id-headers flag.
›Adds per-request KV cache offloading policy via the on_new_request lifecycle hook and on_schedule_end() hook, plus an object-store secondary offloading tier.
›Introduces scheduler_block_size threaded into KVCacheManager/Coordinator and max_concurrent_batches moved to VllmConfig.
›Adds pluggable KVCacheSpec for configurable KV cache specifications.
›Unifies reasoning and tool-call parsing behind a single Parser.parse() interface, with the Responses parser migrated to it.
+35 moreshow less
›Adds system_fingerprint field to the OpenAI/Responses API.
›Adds chat_template_kwargs support in the Responses API.
›Supports streaming tool/function calling with required tool choice in the OpenAI-compatible API.
›Adds tool_choice="none" honored in streaming responses.
›Adds Anthropic Messages API support for structured output, effort, and system-role messages inside the messages array.
›Model Runner V2 (MRv2) is now the default for Llama and Mistral dense models, gaining a FlashInfer sampler, breakable CUDA graphs, and pipeline-parallel bubble elimination.
›Adds new tool parsers for InternLM2, hy_v3, Phi-4-mini, and Gemma4 in the Rust frontend.
›Adds new tool parsers: MiniCPM5 XML, Qwen3 XML JSON-args-first, and DeepSeek DSML incremental streaming.
›Adds support for new models: Step-3.7-Flash, Cosmos3 Reasoner, Gemma 4 Unified (encoder-free), JetBrains Mellum v2, Granite Speech Plus, and Cohere Mini Code.
›Adds Gemma 4 MTP and encoder-free Unified support.
›Adds LMCacheMPConnector for LMCache in disaggregated serving.
›Enables async EPLB by default and adds EPLB support for DeepSeek-V4 Mega-MoE.
›Adds SSL support for the data-parallel supervisor.
›Enables Triton MoE backend on Hopper by default.
›Adds native W4A16 and fused-MoE W4A16 HIP kernels for RDNA3 (gfx1100) on AMD ROCm.
›Adds AITER top-k/top-p sampler by default on AMD ROCm.
›Adds Intel XPU support for CPU/tiering offloading, block_fp8_moe, block-scaled W8A8 FP8 path, and transparent sleep mode.
›Adds zentorch-accelerated W8A8/W4A16 quantization on AMD Zen CPUs.
›Adds RISC-V RVV WNA16 helpers and PowerPC SHM communicator.
›Adds arm64 CI image.
›Adds ModelOpt LM-head quantization and MXFP8 non-gated MoE support.
›Adds compressed-tensors support for WNA8O8Int linears, WNInt embeddings, and asymmetric MoE WNA16 Marlin.
›Adds Triton W4A16 as a CUDA fallback for non-Marlin-aligned shapes.
›Adds thinking_token_budget validation in the frontend.
›Adds NUMA auto-binding on DGX B300.
›Adds extra_repr() for pooler classes in pooling/classification models.
›Adds auto-selection of registered video loader for VLMs.
›Adds Nixl zero-copy EPLB transfers and Nixl Mamba prefix-caching mode.
›Adds Transformers v5 compatibility, including vendored MiniCPM-V/O processors and Voxtral fetch_audio for transformers≥5.10.
›Adds DeepSeek-V4 TRTLLM-gen attention kernel, EPLB support, selective prefix-cache retention for sliding-window KV cache, and index-share feature for DSA MTP.
›Adds sparse NCCL weight transfer for in-place model updates.
›Adds per-GPU-worker RDMA NIC selection for data-parallel deployments.
Phoenix v17.6.0 adds experiment editing, annotation score time series, and live-streaming pan/zoom controls.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v17.6.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v17.6.0
›Adds trace and session annotation score time series to the metrics view, enabling trend analysis of annotation scores over time.
›Adds pan and zoom time range controls to the UI with a live streaming toggle for real-time monitoring.
›Adds experiment editing and eval skills for agents, allowing in-place modification of experiments and evaluations.
Weave v0.52.43 adds OpenAI Agents SDK span emission, aggregate feedback endpoint, set_attributes/add_event on span types, and playground trace routing.
└──▷ GET THIS VERSION
$ git clone --branch v0.52.43 https://github.com/wandb/weave.git
# already have the repo? check out this version:$ git checkout v0.52.43
›Adds set_attributes and add_event methods to Tool, LLM, SubAgent, and Turn span types in both the Python and TypeScript SDKs.
›New aggregate feedback API endpoint for summarising feedback across calls.
›Emits invoke_agent, execute_tool, chat, handoff, guardrail, transcription, speech, speech_group, mcp_list_tools, and custom spans for the OpenAI Agents SDK (TypeScript).
›Emits message data on chat spans for the OpenAI Agents SDK (TypeScript).
›Supports post-hoc start/end time overrides on GenAI spans (TypeScript).
+5 moreshow less
›Pushes traces from playgrounds to the spans table.
›Adds a database migration to include agent columns in the feedback table.
›Formalises integration-tracking call attributes for consistent attribution across integrations.
›Adds claude-fable-5 to model providers and cost tracking.
›Performance improvement: bounded conversation message previews for grouped spans query, extending the calls_query_stats fast path with a time-window filter, and a flat-sum fast path for unfiltered storage stats.
THE AI TOOLCHAIN