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 v2.7.4 adds SuperserveTools, PlivoTools, The Context Company observability, and expanded Telegram/Tavily/Google toolkit methods.
└──▷ GET THIS VERSION
$ git clone --branch v2.7.4 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:$ git checkout v2.7.4
└──▷ USE IT
Run agent-generated code in an isolated Firecracker sandbox for a long-running agent task.
python
from agno.agent import Agent
from agno.tools.superserve import SuperserveTools
agent = Agent(tools=[SuperserveTools()])
agent.print_response('Write and execute a Python script that processes this dataset')
Send an SMS or make a voice call from an agent using Plivo credentials.
python
from agno.agent import Agent
from agno.tools.plivo import PlivoTools
agent = Agent(tools=[PlivoTools()])
agent.print_response('Send an SMS to +15551234567 saying the nightly scan is complete')
›Adds SuperserveTools class to run agent-generated code and manage files inside Superserve, a Firecracker-based sandbox platform designed for long-running agents.
›Adds PlivoTools class to send SMS, make voice calls, and look up phone numbers via Plivo.
›Adds pin_message, get_chat, get_file, and react_with_emoji methods to TelegramTools, plus save_downloads and output_directory options to save downloaded files to disk.
›Adds domain, date range, topic, and country filter parameters to TavilyTools searches.
›Adds observability integration to trace agent runs with The Context Company.
+3 moreshow less
›Enhances agno create with interactive starter template and project name prompts, four new starters (Azure, Helm, Modal, Render), and automatic .env seeding.
›Enables OxylabsTools to return full page content as Markdown.
›Workflows now accept run_context in Router selectors and Condition evaluators (deprecating session_state).
PydanticAI v2.12.0 adds Kimi-K3 model support and two new agent stream events for deferred tools and enqueued messages.
└──▷ GET THIS VERSION
$ git clone --branch v2.12.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:$ git checkout v2.12.0
└──▷ USE IT
Observe deferred tool calls and their results while streaming an agent run, useful for auditing async/human-in-the-loop tool workflows.
python
async with agent.run_stream(prompt) as stream:
async for event in stream.stream_events():
if isinstance(event, DeferredToolCallEvent):
print('Tool deferred:', event)
elif isinstance(event, DeferredToolResultEvent):
print('Deferred result received:', event)
elif isinstance(event, EnqueuedMessagesEvent):
print('Enqueued messages delivered:', event)
›Adds DeferredToolCallEvent and DeferredToolResultEvent to AgentStreamEvent, enabling stream-level visibility into deferred tool call lifecycle.
›Emits EnqueuedMessagesEvent when previously enqueued messages are delivered into a run, making message-replay observable in the event stream.
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 v2.1.212 adds per-session search/subagent caps, MCP auto-backgrounding, and a redesigned /fork workflow
└──▷ GET THIS VERSION
$ git clone --branch v2.1.212 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.212
└──▷ TRY IT
Cap the number of web searches an autonomous session can make to avoid runaway search loops in long-running agents.
$ CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION=50 claude --bg 'Research all CVEs published last week and summarize the top 10'
Limit subagent delegation in a complex multi-agent run, then reset the budget mid-session with /clear if you need to continue.
$ CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION=20 claude 'Refactor the entire src/ directory using parallel subagents'
Lower the MCP auto-background threshold so slow tool calls move to the background sooner, keeping your session interactive.
$ CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS=30000 claude
›Adds CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION environment variable to tune the new per-session WebSearch call limit (default 200), preventing runaway search loops.
›Adds CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION environment variable to tune the new per-session subagent spawn cap (default 200); /clear resets the budget.
›Adds CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS environment variable to configure (or disable) the threshold at which MCP tool calls running longer than 2 minutes automatically move to the background.
›Adds claude auto-mode reset subcommand to restore the default auto-mode configuration; accepts --yes to skip the confirmation prompt.
›Redesigns /fork to copy the conversation into a new background session (its own row in claude agents) while you keep working; the previous in-session subagent behavior is now /subtask.
+7 moreshow less
›Adds /resume in the agent view as a picker of past sessions — including sessions deleted from the list — resuming the selected one as a background session.
›Enterprise forceLoginMethod is now enforced for VS Code extension, SDK, setup-token, and install-github-app logins, not just the terminal.
›Headless/SDK sessions now apply a set_model control request mid-turn, switching models on the next round-trip without waiting for the next turn.
›Agent view and claude agents --json now surface sessions waiting on a sandbox, MCP-input, or managed-settings prompt as 'Needs input' instead of 'Working'.
›Session transcripts now record the reasoning effort level on each assistant message.
›Reduces token usage in inter-agent messaging: SendMessage bodies are no longer duplicated into replayed history and tool results.
›Improves prompt caching: the mid-conversation system block now works behind LLM gateways and custom base URLs (Bedrock, Vertex, 1P).
└──▷ BREAKING ON UPGRADE
!The Task tool's mode parameter is deprecated and now ignored; subagents inherit the parent session's permission mode by default.
!/fork no longer launches an in-session subagent — that behavior is now /subtask. Workflows that relied on /fork for subagent spawning must switch to /subtask.
Autonomous coding agent as an SDK, IDE extension, or CLI assistant.
Cline SDK v0.0.63 emits a new telemetry event when the consecutive-mistake limit is reached.
└──▷ GET THIS VERSION
$ git clone --branch sdk/sdk/v0.0.63 https://github.com/cline/cline.git
# already have the repo? check out this version:$ git checkout sdk/sdk/v0.0.63
›Adds task.mistake_limit_reached telemetry event, emitted by the session runtime whenever the consecutive-mistake limit is hit, captured across all hosts (CLI, VS Code extension, hub daemon) including auto-stops with no host prompt configured.
An MCP that lets AI tools securely connect to your infrastructure, write IaaS code, debug issues, and assist during incidents - without risking production stability. Built for security teams to approve and infrastructure teams to experience like magic.
emisar v0.30.0 enforces tenant isolation in the database and adds signed-pack visibility over MCP
└──▷ GET THIS VERSION
$ git clone --branch v0.30.0 https://github.com/AndrewDryga/emisar.git
# already have the repo? check out this version:$ git checkout v0.30.0
›Signed-pack enforcement is now visible over MCP: a runner that requires signed packs announces that requirement to MCP clients.
›Database-enforced tenant isolation: every runner-owned child row carries a composite foreign key to its account, so cross-tenant writes are refused at the database row level rather than only in application code.
›CHECK constraints added for security-relevant enums so the database rejects out-of-range values that gate trust and dispatch, independent of the application layer.
›Content-Security-Policy headers now applied to public unauthenticated pages and error responses, matching the rest of the site.
›Sentry named as a subprocessor; events scrubbed of personal data before leaving the process.
+3 moreshow less
›Trigram index added to speed run search on large accounts, replacing three now-covered indexes.
›Steadier dispatch: a runner at capacity redelivers refused dispatches, the stale-dispatch sweep drains oldest-queued entries first, and stuck pending runs advance instead of stranding.
›Severe database and load-balancer conditions now page directly; point-in-time recovery drill measures actual RTO.
Phoenix v19 adds OAuth2 authorization server with CLI login and tightens API key creation to session-authenticated callers only.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v19.0.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v19.0.0
›Adds OAuth2 authorization server and CLI login (pxi CLI) for browser-based and unattended authentication workflows.
›Improved pxi token display makes issued tokens easier to inspect after CLI login.
›Relocates the pxi button in the UI for improved discoverability.
└──▷ BREAKING ON UPGRADE
!GraphQL createUserApiKey and createSystemApiKey mutations no longer accept API-key-authenticated callers; workflows that used an existing API key for unattended key creation must switch to a session-authenticated context or, for system keys, use PHOENIX_ADMIN_SECRET.
Phoenix 18.1.0 adds REST CRUD for API keys, a px setup CLI wizard, a user-friction evaluator, and user.id span tracing.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v18.1.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v18.1.0
└──▷ TRY IT
Onboard a new app to Phoenix and confirm traces are flowing without manual config.
$ px setup
›Adds REST CRUD endpoints for user and system API keys under /v1/users/api_keys (page size capped at 1000), enforcing a unified authority model that requires a human session to issue keys.
›Adds px setup CLI command to connect an application to Phoenix and verify that traces arrive, with an option to include the Phoenix docs MCP server during setup.
›Surfaces the user.id span attribute on spans, traces, and sessions for end-to-end user-level traceability.
›Adds a user-friction evaluator to arize-phoenix-evals for assessing friction in user interactions.
›Splits Users and API Keys into dedicated tabs in the Settings UI for clearer key management.
+2 moreshow less
›Adds browser/server tabs in the AI provider credential storage UI to clarify where credentials are held.
›Settings tabs now display vertically on large screens with a responsive fallback.
Langfuse v3.219.0 adds deep-linked monitor alerts and context-aware AI assistant quick actions.
└──▷ GET THIS VERSION
$ git clone --branch v3.219.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:$ git checkout v3.219.0
›Monitor alerts now deep-link directly to the traces/observations table filtered to the breaching time window, making it faster to investigate threshold violations.
›AI assistant quick actions are now context-aware and cover a broader range of in-app scenarios.
›Dark mode layering model updated for improved visual consistency across the UI.
THE AI TOOLCHAIN