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.
Re-enable AgentOS access logs after upgrading, since they are now off by default.
python
serve(access_log=True)
›Adds store_history_messages and store_tool_messages flags to Agent and Team to control whether history and tool messages are persisted on run output.
›Adds stream_executor_events to workflows for filtering events emitted by agents, teams, or custom functions — complementing the existing stream_intermediate_events (for workflow-level events like WorkflowStarted, StepStarted) and stream_member_events on Team.
›Adds access_log=True parameter to serve() to re-enable AgentOS access logs, which are now off by default.
›Adds async Postgres support across the library for non-blocking database operations.
›Adds knowledge search endpoints to the AgentOS API, enabling vector database searches via the AgentOS API.
+3 moreshow less
›Adds service account authentication support to GoogleSheetsTools.
›Adds native reasoning model support for Gemini 2.5+, Anthropic Claude, and VertexAI Claude when used as reasoning models with Agents.
›Allows MCP server URLs without the conventional /mcp path segment when using MCPToolbox.
└──▷ BREAKING ON UPGRADE
!AgentOS access logs are now disabled by default; existing setups relying on access logging must explicitly pass access_log=True to serve() to restore the previous behavior.
langchain-anthropic 1.0.0a5 adds async middleware, PDF ToolMessage inputs, memory/context management, web fetch, MCP connector, files API, code execution, and more.
└──▷ GET THIS VERSION
$ git clone --branch langchain-anthropic==1.0.0a5 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain-anthropic==1.0.0a5
└──▷ USE IT
Pass cache_control to a specific message block to enable prompt caching on expensive context.
python
from langchain_anthropic import ChatAnthropic
model = ChatAnthropic(model="claude-3-5-sonnet-20241022")
response = model.invoke(
[{"role": "user", "content": "Summarise this document."}],
cache_control={"type": "ephemeral"}
)
Enable parallel tool calls so Claude can invoke multiple tools concurrently in a single turn.
python
from langchain_anthropic import ChatAnthropic
from langchain_core.tools import tool
@tool
def get_weather(city: str) -> str:
"""Get weather for a city."""
return f"Sunny in {city}"
model = ChatAnthropic(model="claude-3-5-sonnet-20241022", parallel_tool_calls=True)
model_with_tools = model.bind_tools([get_weather])
response = model_with_tools.invoke("What is the weather in Paris and London?")
›Adds cache_control as a passthrough kwarg on ChatAnthropic invocations for fine-grained prompt caching control.
›Adds parallel_tool_calls parameter support to ChatAnthropic for controlling concurrent tool execution.
›Supports urls as input to ChatAnthropic, enabling direct URL references in multimodal messages.
›Adds web fetch beta tool support to ChatAnthropic, allowing the model to retrieve content from the web during inference.
LangChain 1.0.0a15 adds async agent support, a middleware pipeline with PII/HITL/retry/tool-limit hooks, and wrap_model_call/wrap_tool_call decorators.
└──▷ GET THIS VERSION
$ git clone --branch langchain==1.0.0a15 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain==1.0.0a15
└──▷ USE IT
Add PII redaction and a tool-call cap to an agent so sensitive data never reaches tools and runaway loops are prevented.
python
from langchain_v1.agents.middleware import PIIMiddleware, ToolCallLimitMiddleware
from langchain_v1 import create_agent
agent = create_agent(
model=model,
tools=[search, calculator],
middleware=[PIIMiddleware(), ToolCallLimitMiddleware(max_calls=10)],
)
›Adds wrap_model_call and wrap_tool_call middleware decorator hooks (with both sync and async implementations) to intercept and modify model and tool invocations inside create_agent.
›Adds before_agent and after_agent lifecycle hooks for middleware.
›Adds retry_model_request middleware hook and ModelFallbackMiddleware for automatic model fallback on failure.
›Adds ToolCallLimitMiddleware to cap the number of tool calls an agent can make.
›Adds PIIMiddleware to detect and redact personally identifiable information in agent inputs/outputs.
+16 moreshow less
›Adds LLM-selection middleware (add llm selection middleware) enabling dynamic model routing at runtime.
›Adds Context Editing Middleware for runtime modification of the agent's context window.
›Adds TodoListMiddleware (formerly PlanningMiddleware) for structured task-planning within the agent loop.
›Adds description generator for HITL (human-in-the-loop) middleware to auto-generate interrupt descriptions.
›Adds async support to create_agent, enabling fully asynchronous agent execution.
›Adds dynamic system prompt middleware for runtime prompt injection.
›Adds tool emulator capability for simulating tool responses without real tool execution.
›Expands the messages namespace exports, including RemoveMessage, for richer message manipulation.
›Adds ModelResponse export from agents.middleware.
›Adds PEP 604 (| union syntax) support in tool node error handlers.
›Adds decorator pattern for dynamically generated middleware.
›Adds minimal and verbosity options to the OpenAI integration.
›Enables stream_usage by default when using the default base URL and client in the OpenAI integration.
›Adds stuff and map_reduce chains.
›Exposes rate_limiters from langchain_core in the langchain_v1 namespace.
›Migrates Anthropic middleware to langchain_anthropic package.
└──▷ BREAKING ON UPGRADE
!Globals removed from langchain-v1; globals in langchain-classic and langchain-core are updated — code relying on langchain-v1 globals will break.
!ToolNode removed from agents namespace in langchain_v1; it is now located in the tools namespace.
!PlanningMiddleware renamed to TodoListMiddleware — any code referencing PlanningMiddleware will fail to import.
!Python 3.9 support dropped for langchain_v1.
!create_react_agent renamed to create_agent — existing calls to create_react_agent will break.
!model_request node renamed to model — graph configurations referencing the model_request node name will break.
langchain-core 1.0.0rc1 adds PDF tool message support, AWS Bedrock document blocks, OpenAI web_search tool, and more new capabilities.
└──▷ GET THIS VERSION
$ git clone --branch langchain-core==1.0.0rc1 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain-core==1.0.0rc1
└──▷ USE IT
Include message IDs when converting LangChain messages to OpenAI format, useful for correlating messages across systems.
python
from langchain_core.messages.utils import convert_to_openai_messages
openai_messages = convert_to_openai_messages(messages, include_id=True)
Strip PostgreSQL NUL bytes from LLM output before inserting into a Postgres database to avoid DataError.
python
from langchain_core.utils import sanitize_for_postgres
clean_text = sanitize_for_postgres(llm_output)
›Adds include_id optional parameter to convert_to_openai_messages function to control whether message IDs are included in OpenAI-formatted output.
›Adds id field to Document objects passed to the filter callback in InMemoryVectorStore similarity search.
›Adds web_search to the list of recognized built-in OpenAI tools.
›Adds image_generation tool to the list of known OpenAI tools.
›Supports PDF inputs in ToolMessage content blocks.
+10 moreshow less
›Supports AWS Bedrock document content blocks in msg_content_output.
›Supports adding PromptTemplates with formats other than f-string.
›Allows overriding ls_model_name from kwargs when tracing model calls.
›Allows custom Mermaid diagram URL via the new custom URL override capability.
›Adds sanitize_for_postgres utility function to remove PostgreSQL NUL bytes that cause DataError.
›Adds an option to make deserialization more permissive.
›Zeros out token costs for cache hits in token usage tracking.
›Adds additional hashing options to the indexing API with a warning on SHA-1 use.
›Traces response body on error for improved observability.
›Exposes recognized block types for tool messages.
└──▷ BREAKING ON UPGRADE
!BaseMemory is removed from langchain-core and moved to langchain-classic.
!Items marked for removal in schemas.py have been deleted.
!function_calling.py utilities previously marked for removal have been deleted.
!The pydantic_v1/ compatibility shim has been deleted from langchain-core.
!get_relevant_documents has been removed.
!Global state previously in langchain-v1 has been removed; globals are now only in langchain-classic and langchain-core.
v0.14.5 adds SGLang LLM integration, SignNow MCP tools, Tavily URL extraction, Azure PostgreSQL hybrid search, and new model support across Anthropic, Bedrock, OpenAI, and Fireworks.
└──▷ GET THIS VERSION
$ git clone --branch v0.14.5 https://github.com/run-llama/llama_index.git
# already have the repo? check out this version:$ git checkout v0.14.5
└──▷ USE IT
Run inference through a local SGLang server using the new first-class SGLang LLM integration.
python
from llama_index.llms.sglang import SGLang
llm = SGLang(model="meta-llama/Llama-3.1-8B-Instruct")
response = llm.complete("Explain prompt injection attacks in one paragraph.")
print(response)
›Adds llama-index-llms-sglang (v0.1.0) — a new SGLang LLM integration for running local inference via SGLang.
›Adds llama-index-tools-signnow (v0.1.0) — a new SignNow MCP tools integration for document signing workflows.
›Adds a Tavily extract function in llama-index-tools-tavily-research for URL content extraction.
›Adds hybrid search support to llama-index-vector-stores-azurepostgresql.
›Adds prompt caching model validation utilities to llama-index-llms-anthropic.
+8 moreshow less
›Adds support for custom models in llama-index-llms-fireworks.
›Adds support for xAI models in llama-index-llms-oci-genai.
›Adds haiku 4.5 model support to llama-index-llms-anthropic and llama-index-llms-bedrock-converse.
›Adds Claude Sonnet 4.5 as a reasoning model and Opus 4.1 function-calling model support in llama-index-llms-bedrock-converse.
›Adds support for global cross-region inference profile prefix in llama-index-llms-bedrock-converse.
›Adds GPT-5 and GPT-5 Pro model support (including JSON_SCHEMA_MODELS) in llama-index-llms-openai.
›Adds pagination parameters for repository tree and issues in llama-index-readers-gitlab.
›Adds a progress bar for multiprocess document loading in llama-index-core.
$ git clone --branch v1.2.9-vscode https://github.com/continuedev/continue.git
# already have the repo? check out this version:$ git checkout v1.2.9-vscode
└──▷ USE IT
Authenticate an MCP server using a Bearer token directly in your config instead of relying on environment credentials.
Continue 1.25.0 renames workflows to agent files and overhauled onboarding for updated billing.
└──▷ GET THIS VERSION
$ git clone --branch @continuedev/[email protected] https://github.com/continuedev/continue.git
# already have the repo? check out this version:$ git checkout @continuedev/[email protected]
›Renames 'workflows' to 'agent files' across the product, reflecting a new operational surface name for reusable agent task definitions.
›Overhauled onboarding flow updated for revised billing tiers and free trial handling.
Show a git status icon in the title bar to see repo state at a glance without opening the git panel.
json
{
"title_bar": {
"show_branch_icon": true
}
}
›Introduces agent_buffer_font_size setting for granular buffer font size control in the agent panel, separate from regular editors.
›agent_font_size is renamed to agent_ui_font_size for agent UI font size control.
›Adds "title_bar": {"show_branch_icon": true} setting to display a git status indicator icon in the title bar.
›Adds "terminal": {"keep_selection_on_copy": false} setting to opt out of the new default behavior of retaining terminal selection after copying.
›Adds experimental "status_bar": {"experimental.show": false} setting to hide the status bar.
+14 moreshow less
›Adds granted_extension_capabilities setting to control the capabilities granted to extensions.
›Supports action sequences in keymap bindings via "action::Sequence", enabling chaining of multiple actions — e.g., "cmd-alt-a": ["action::Sequence", ["editor::SelectLargerSyntaxNode", "editor::Copy", "editor::UndoSelection"]].
›Adds support for OpenAI Codex as an ACP agent.
›Adds support for pasting TIFF and BMP images in the agent panel.
›Adds support for HTML tables and HTML block quotes in Markdown Preview.
›Adds comment injections for Rust.
›Enables the outline modal in channel notes.
›Revamps project panel entry refresh for significantly smoother performance in large projects.
›Adds SelectPrevious and SelectAllMatches items to the Selection app menu.
›Adds Close Multibuffers pane context menu entry.
›Improves Collab panel by showing display names and GitHub handles.
›Adds support for file extension icons for patterns such as stories.tsx and stories.svelte.
›Adds graceful autohiding to scrollbars outside of the editor.
›Terminals now keep the selection after copying text by default, matching Terminal.app, Ghostty, and VS Code terminal behavior.
└──▷ BREAKING ON UPGRADE
!agent_font_size is renamed to agent_ui_font_size; existing configs using agent_font_size must be updated.
!format_on_save is now restricted to the values "on" and "off" only — format steps previously placed in format_on_save must be moved to the formatter array. Having format steps in both format_on_save and formatter is no longer supported.
Codebase investigator now enabled by default in gemini-cli nightly.
└──▷ GET THIS VERSION
$ git clone --branch v0.10.0-nightly.20251015.996c9f59 https://github.com/google-gemini/gemini-cli.git
# already have the repo? check out this version:$ git checkout v0.10.0-nightly.20251015.996c9f59
›Enables the codebase investigator feature by default ahead of the next preview release.
THE AI TOOLCHAIN