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 v2.6.8 adds Antigravity API support, Gemini managed agents (Deep Research + Antigravity), and centralized path-safety utilities.
└──▷ GET THIS VERSION
$ git clone --branch v2.6.8 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:$ git checkout v2.6.8
└──▷ USE IT
Use safe_join to safely resolve user-supplied paths and avoid path-traversal vulnerabilities in custom tools.
python
from agno.utils.path_safety import safe_join, PathSecurityError
try:
safe_path = safe_join("/var/app/uploads", user_input_filename)
except PathSecurityError as e:
print(f"Blocked unsafe path: {e}")
›Adds AntigravityAgent (a BaseExternalAgent served through AgentOS with native sessions, streaming, and UI) for first-party Google Antigravity API integration.
›Adds AntigravityTools (a Toolkit) so any Agno agent can delegate sub-tasks to a managed Antigravity sandbox.
›Adds GeminiInteractions support for Google's managed Deep Research agent — autonomous research with citations, background streaming with reconnect, and last_event_id resume.
›Adds GeminiInteractions support for Google's managed Antigravity agent — general-purpose agent running in a managed Linux sandbox.
›Adds agent, agent_config, and environment fields to GeminiInteractions for selecting and configuring managed agents, with per-agent forcing of background and store.
+5 moreshow less
›Adds mcp_servers and file_search_store_names support on the GeminiInteractions agent path.
›Introduces agno.utils.path_safety module with safe_join and safe_join_subpath, hardening FileGenerationTools, SlackTools, Toolkit._check_path, agno.skills.utils.is_safe_path, and FileTools.check_escape against path traversal, symlink escape, control-char injection, Windows MagicDot, and Unicode normalization attacks.
›Introduces PathSecurityError (raised on path-safety violations); FileGenerationSecurityError is kept as a deprecation alias.
›Adds 18 self-contained data-labeling workflows to cookbook/data_labeling/, covering text, image, audio, video, document, and composed (LLM-as-judge, quality review) labeling primitives.
›Adds a deterministic Slack HITL incident-commander demo to cookbook/, demonstrating structured pauses via tool_choice='required', user-input echo, and clean termination via stop_after_tool_call=True.
└──▷ BREAKING ON UPGRADE
!The available_models field is removed from EvalsDomainConfig; the only supported source for the Evals UI dropdown is now AgentOSConfig.available_models.
PydanticAI v1.98.0 adds OpenAI Responses token counting and a unified retries parameter on Agent.
└──▷ GET THIS VERSION
$ git clone --branch v1.98.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:$ git checkout v1.98.0
└──▷ USE IT
Count input tokens for an OpenAI Responses model request before committing to the API call.
python
from pydantic_ai.models.openai import OpenAIResponsesModel
model = OpenAIResponsesModel('gpt-4o')
token_count = await model.count_tokens(messages, model_settings=None)
›Adds OpenAIResponsesModel.count_tokens method to count input tokens for OpenAI Responses model calls before sending them.
›Replaces Agent parameters tool_retries= and output_retries= with a single retries: int | AgentRetries parameter, enabling unified retry control across tools and outputs.
└──▷ BREAKING ON UPGRADE
!The Agent constructor parameters tool_retries= and output_retries= are replaced by retries: int | AgentRetries; code passing either removed keyword argument will break on upgrade.
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 v2.1.145 adds JSON agent listing, richer OTEL spans, plugin previews, and mouse support in fullscreen.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.145 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.145
└──▷ TRY IT
Pipe live agent session data into a status bar or session picker script.
›New claude agents --json command lists live Claude sessions as JSON for scripting (tmux-resurrect, status bars, session pickers).
›Adds agent_id and parent_agent_id attributes to claude_code.tool OTEL spans, with corrected trace parenting so subagent spans nest under their dispatching Agent tool span.
›Status line JSON input now includes GitHub repo and PR information when detected.
›/plugin Discover and Browse screens now preview a plugin's commands, agents, skills, hooks, and MCP/LSP servers before installation.
›claude agents terminal tab title now shows the awaiting-input count so an alt-tabbed window indicates when an agent needs attention.
+2 moreshow less
›Slash command and @-mention suggestion list now supports mouse hover and click in fullscreen mode.
›Read tool now returns a truncated first page with a 'PARTIAL view' notice instead of a hard error when a whole-file read exceeds the token limit.
›Plugin browse and discover panes now display when each plugin was last updated.
›/model now changes the model for the current session only; press d in the model picker to set a default for all new sessions.
›Adds claude --bg --name <label> echoing the session name in the post-spawn confirmation message.
+8 moreshow less
›Adds claude agents Ctrl+R session rename that immediately updates the attached session's banner.
›/doctor now shows an exec-form example when a command hook is missing the command field.
›Skill-listing truncation moved out of startup notifications — full breakdown now available via /doctor.
›Improved SDK/headless MCP startup: pre-wait now overlaps with startup instead of blocking before the first turn (up to 2 s faster with slow MCP servers).
›Improved recovery from rare pre-response stream stalls — now retries streaming once instead of falling back to a slower non-streaming request.
›claude agents / --bg rejection messages now name the specific gate (non-TTY, env var, or setting) instead of a generic message.
›Plugin marketplace add/update now respects the CLAUDE_CODE_PLUGIN_PREFER_HTTPS environment variable.
›/plugin now returns to the Installed list after enabling, disabling, or uninstalling a plugin.
›Adds auth.redirectPort config option to pin the OAuth callback to a fixed port for MCP servers.
›Adds COPILOT_PLUGIN_DIR_ONLY environment variable to disable automatic plugin discovery, enabling deterministic plugin sets when combined with --plugin-dir.
›Adds copilot plugin update --all to update all installed plugins in a single command.
›Adds /chronicle search subcommand to search all session content by keyword or topic.
›Adds /session id subcommand to display the current session ID and copy it to the clipboard.
+23 moreshow less
›Adds /memory on|off|show slash command to enable, disable, or view persistent memory status.
›Adds /exit print option to print the session transcript to the terminal before exiting.
›Adds /rubber-duck command (experimental) to invoke an independent critique of the agent's current work.
›Adds experimental /mcp search command to search and install MCP servers from a registry.
›Adds experimental tool search with deferred loading for MCP and external tools.
›Adds 'None' reasoning effort option in the reasoning effort picker to disable model reasoning entirely.
›Extends --plugin-dir and --additional-mcp-config to work in --server / --headless mode.
›Injects postToolUse hook additionalContext as a system message to the model instead of silently discarding it.
›Hooks (preToolUse, postToolUse, subagentStart, subagentStop) now fire correctly for sub-agent tool calls.
›Plugins loaded via --plugin-dir now correctly register their agents as available task(agent_type=...) subagents in prompt mode.
›Repo hooks in .github/hooks/ now load automatically in prompt mode (-p) when the folder is already trusted.
›Prompt mode (-p) automatically loads workspace MCP sources when the current folder is already trusted.
›Adds support for running the CLI on Alpine Linux (musl libc).
›Memory permission prompts now name the scope of a stored memory — user scope or the specific owner/repo for repository scope — and timeline entries show (for user) / (shared with repository collaborators).
›Auto-links GitHub issue and PR references (owner/repo#number) in assistant responses.
›Auto-update downloads the smaller platform-specific package instead of the universal one when available.
›Content-filtered model responses now display an explanation instead of a blank assistant turn.
›Reduces PowerShell syntax errors on Windows by avoiding && chaining when using legacy PowerShell 5.x.
›PromptFrame UI now renders inside tmux when the outer terminal is ghostty, WezTerm, or kitty (detected via tmux list-clients).
›Document attachment paths are included in context so the agent can reference pasted file paths, including Windows 'Copy as path' inputs.
›MCP stdio servers now display type as 'stdio' instead of 'local' for consistency.
›MCP server configurations with no args field are now accepted and treated as an empty args list.
›Input prompt collapses to a single line when empty and grows naturally as text is entered.
$ git clone --branch arize-phoenix-v15.11.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v15.11.0
›Adds generative UI rendering within agent chats, enabling dynamic visual components to be displayed inline during agent conversations.
THE AI TOOLCHAIN