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.
Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.
langchain-exa 1.1.0 changes the default Exa search type from neural to auto.
└──▷ GET THIS VERSION
$ git clone --branch langchain-exa==1.1.0 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain-exa==1.1.0
›Changes the default search type from neural to auto, enabling Exa to automatically select the best search strategy per query.
└──▷ BREAKING ON UPGRADE
!The default Exa search type is changed from neural to auto; existing integrations relying on the implicit neural default will now use auto search behavior without an explicit override.
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 v2.1.85 adds conditional hook filtering, richer PreToolUse headless flows, and expanded deep-link query support.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.85 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.85
└──▷ USE IT
Run a git-related hook only when Bash commands match git *, avoiding overhead for every other tool call.
Allow a headless CI integration to answer tool-permission questions programmatically without interactive prompts.
json
# PreToolUse hook returns both a permission decision and updated input
# so AskUserQuestion is satisfied without user interaction
{
"permissionDecision": "allow",
"updatedInput": { "answer": "yes" }
}
Enable full tool-parameter logging in OTEL tool_result events for detailed pipeline observability.
$ OTEL_LOG_TOOL_DETAILS=1 claude --mcp-config mcp.json 'audit my repo'
›Adds CLAUDE_CODE_MCP_SERVER_NAME and CLAUDE_CODE_MCP_SERVER_URL environment variables to MCP headersHelper scripts, enabling a single helper script to serve multiple MCP servers.
›Adds conditional if field to hooks using permission rule syntax (e.g., Bash(git *)) so hooks only fire for matching tool calls, reducing unnecessary process spawning.
›Adds timestamp markers in transcripts when scheduled tasks (/loop, CronCreate) fire, making it easier to audit timing of automated runs.
›Expands deep link query support (claude-cli://open?q=…) to 5,000 characters, with a 'scroll to review' warning for long pre-filled prompts.
›MCP OAuth now follows RFC 9728 Protected Resource Metadata discovery to locate the authorization server automatically.
+4 moreshow less
›Organization-policy-blocked plugins (managed-settings.json) are now hidden from marketplace views and cannot be installed or enabled.
›PreToolUse hooks can now satisfy AskUserQuestion by returning updatedInput alongside permissionDecision: "allow", enabling headless integrations to supply answers programmatically.
›Gates tool_parameters in OpenTelemetry tool_result events behind the OTEL_LOG_TOOL_DETAILS=1 environment variable to reduce log verbosity by default.
Claude Code v2.1.84 adds PowerShell tool preview, new hooks, env-var overrides for 3p model capability detection, and MCP deduplication.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.84 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.84
└──▷ TRY IT
Override capability detection when using a pinned Bedrock/Vertex/Foundry model that Claude Code doesn't recognize as supporting extended thinking.
$ ANTHROPIC_DEFAULT_OPUS_MODEL_SUPPORTS=thinking claude
Raise the streaming idle watchdog timeout to 3 minutes for slow-responding enterprise API gateways.
$ CLAUDE_STREAM_IDLE_TIMEOUT_MS=180000 claude
›Adds PowerShell tool for Windows as an opt-in preview, enabling native PowerShell execution alongside the existing bash tool.
›Adds ANTHROPIC_DEFAULT_{OPUS,SONNET,HAIKU}_MODEL_SUPPORTS env vars to override effort/thinking capability detection for pinned default models on Bedrock, Vertex, and Foundry; companion _MODEL_NAME/_DESCRIPTION vars customize the /model picker label.
›Adds CLAUDE_STREAM_IDLE_TIMEOUT_MS env var to configure the streaming idle watchdog threshold (default 90 s).
›Adds TaskCreated hook that fires when a task is created via TaskCreate.
›Adds type: "http" support for the WorktreeCreate hook — return the created worktree path via hookSpecificOutput.worktreePath in the response JSON.
+10 moreshow less
›Adds allowedChannelPlugins managed setting for team/enterprise admins to define a channel plugin allowlist.
›Adds x-client-request-id header to all API requests to aid timeout debugging.
›Adds idle-return prompt that nudges users back after 75+ minutes of inactivity to /clear, reducing unnecessary token re-caching on stale sessions.
›Deep links (claude-cli://) now open in the user's preferred terminal instead of the first terminal found in the detection list.
›Rules and skills paths: frontmatter now accepts a YAML list of globs, not just a single pattern.
›MCP servers configured both locally and via claude.ai connectors are now deduplicated — local config wins.
›Background bash tasks stuck on an interactive prompt now surface a notification after ~45 seconds.
›Global system-prompt caching now works when ToolSearch is enabled, including for users with MCP tools configured.
›[VSCode] Adds a rate-limit warning banner showing usage percentage and reset time.
›Stats screenshot (Ctrl+S in /stats) now works in all builds and renders 16× faster.
Adds useResponsesApi opt-out option and OpenRouter app-identification headers in Continue v1.2.19.
└──▷ GET THIS VERSION
$ git clone --branch v1.2.19-vscode https://github.com/continuedev/continue.git
# already have the repo? check out this version:$ git checkout v1.2.19-vscode
└──▷ USE IT
Opt out of the OpenAI Responses API when it causes compatibility issues with your deployment.
Store team-shared scan rules in .claude/settings.json and personal overrides (e.g. local model preferences) in .claude/settings.local.json, both picked up automatically per repo.
json
# .claude/settings.json — committed to the repo
{
"model": "claude-3-7-sonnet-20250219",
"reasoning_effort": "high"
}
# .claude/settings.local.json — git-ignored, personal overrides
{
"reasoning_effort": "low"
}
›Adds CLAUDE_PROJECT_DIR and CLAUDE_PLUGIN_DATA environment variables to plugin hooks, plus {{project_dir}} and {{plugin_data_dir}} template variables in hook configurations.
›Extends /allow-all (alias /yolo) with on, off, and show subcommands to enable, disable, or check allow-all mode from the CLI.
›Reads .claude/settings.json and .claude/settings.local.json as additional repository-level config sources.
›OTEL hook executions are now recorded as span events instead of child spans, reducing trace clutter in distributed tracing backends.
›/session rename auto-generates a session name from conversation history when called without a name argument.
+7 moreshow less
›Status line payload now includes a session_name field alongside the existing session_id.
›Model picker opens in full-screen view with inline reasoning effort adjustment using ← / → arrow keys; active reasoning effort level (e.g. '(high)') appears in the model display header.
›/yolo path permissions now persist after using /clear to start a new session.
›Ctrl+Y in plan mode opens the most recent research report when no plan exists yet.
›OSC 8 hyperlinks are now clickable in VS Code terminals.
›/clear preserves MCP server state into the new session.
›Workspace MCP servers are now correctly loaded and visible to the agent.
└──▷ BREAKING ON UPGRADE
!The --alt-screen flag and alt_screen setting are removed; alt screen is now always enabled and cannot be disabled.
Lightweight coding agent that runs in your terminal
Codex CLI gains first-class plugin management, path-based multi-agent addressing, and richer app-server capabilities in v0.117.0.
└──▷ GET THIS VERSION
$ git clone --branch rust-v0.117.0 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.117.0
└──▷ TRY IT
Browse, install, or remove plugins from within an active Codex session.
$ /plugins
Label a terminal session with a custom title so parallel Codex sessions are easy to tell apart.
$ /title
›Adds first-class plugin workflow: Codex syncs product-scoped plugins at startup, lets you browse them via /plugins, and install or remove them with auth/setup handling built in.
›Introduces path-based sub-agent addressing (e.g. /root/agent_a) with structured inter-agent messaging and agent listing for multi-agent v2 workflows.
›Adds /title terminal-title picker to both the classic TUI and the app-server TUI, making parallel sessions easier to distinguish.
›App-server clients can now send ! shell commands, watch filesystem changes, and connect to remote WebSocket servers with bearer-token authentication.
›Enables prompt history recall in the app-server TUI, persisting across sessions.
+1 moreshow less
›view_image now returns image URLs for code mode, generated images are reopenable from the TUI, and image-generation history survives session resume.
└──▷ BREAKING ON UPGRADE
!The legacy artifact tool has been removed, and the old read_file and grep_files handlers have been retired — any workflow relying on these tool surfaces will break on upgrade.
!The app-server-backed TUI is now enabled by default, and plugin/app rollout flags have been flipped on in normal builds — setups that depended on the previous defaults may behave differently.
llama.cpp b8532 adds F32 kernel type support for CONV_TRANSPOSE_2D on CUDA and CPU backends.
└──▷ GET THIS VERSION
$ git clone --branch b8532 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b8532
›Adds F32 kernel type support for CONV_TRANSPOSE_2D on CUDA and CPU backends, alongside the existing F16 type, enabling full-precision 2D transposed convolution operations.
llama.cpp b8530 adds DeepSeek-OCR multimodal support with dynamic resolution and flash-attn compatibility in llama-cli.
└──▷ GET THIS VERSION
$ git clone --branch b8530 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b8530
›Adds DeepSeek-OCR model support via llama-mtmd-cli, including CLIP-ViT vision encoder, SAM-based image encoding, and DeepSeek2 language model backend with standard attention.
›Supports dynamic and native resolution preprocessing for DeepSeek-OCR, with configurable min/base/max resolution tiers (640/1024/1280) for image quality vs. speed trade-offs.
›Enables --flash-attn option compatibility with DeepSeek-OCR inference in llama-cli.
›Adds convert_hf_to_gguf.py conversion support for DeepSeek-OCR models, outputting FP16 GGUF weights with corrected tensor names including mm_fc_w and image newline/separator tensors.
›Adds a default Jinja chat template for DeepSeek-OCR for use with the llama.cpp server.
Phoenix v13.19.0 adds a PromptInput compound component, collapsible agent tool calls, session management controls, and an improved project creation flow.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v13.19.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v13.19.0
›Adds PromptInput compound component for building AI interaction UIs.
›Agent chat now collapses tool calls into an expandable pool with streaming performance improvements.
›Agent sessions gain a switcher with delete controls and async summary generation.
›Project creation flow updated with a color picker and description field.
Composio CLI 0.2.9 adds parallel tool execution, ACP-backed subagent runs, and a new dev namespace for manage commands.
└──▷ 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 ACP-backed subagent execution to composio run, enabling agent-driven tool invocation via the CLI.
›Moves CLI manage commands under the dev namespace (e.g., composio dev ...).
›Adds parallel tool execution support in the CLI.
└──▷ BREAKING ON UPGRADE
!CLI manage commands have been moved under the dev namespace; any scripts or workflows invoking those commands directly will need to be updated to use composio dev <command>.
Weaviate v1.35.16 adds token source authentication for GCS backup and usage integrations.
└──▷ GET THIS VERSION
$ git clone --branch v1.35.16 https://github.com/weaviate/weaviate.git
# already have the repo? check out this version:$ git checkout v1.35.16
›Introduces token source authentication support for the backups-gcs and usage-gcs integrations, enabling credential-less auth flows for GCS-backed operations.
THE AI TOOLCHAIN