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 v2.1.8 adds class-based workflow executors, streaming post-hooks, Jira worklogs, and a renamed knowledge search endpoint.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.8 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:$ git checkout v2.1.8
›Renames the knowledge search endpoint from search_vectors to search_knowledge.
›Supports class-based executors in Workflows by defining a class that implements the __call__ method.
›Adds post-hook support on streaming flows.
›Extends JiraTools toolkit to support creating worklogs.
›Updates GoogleCalenderTools to notify attendees when creating, updating, or deleting calendar events.
└──▷ BREAKING ON UPGRADE
!The knowledge search endpoint is renamed from search_vectors to search_knowledge; any client code calling search_vectors will break and must be updated.
LangChain v1.0.0 ships a middleware-first agent API with shell, PII, retry, HITL, and tool-limit hooks plus Python 3.14 support.
└──▷ GET THIS VERSION
$ git clone --branch langchain==1.0.0 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain==1.0.0
›Adds ShellToolMiddleware and ClaudeBashToolMiddleware for intercepting and controlling shell/bash tool execution in agent pipelines.
›Adds TodoListMiddleware (formerly PlanningMiddleware) for structured task planning within the middleware chain.
›Adds PIIMiddleware for detecting and redacting personally identifiable information in model inputs/outputs.
›Adds ToolCallLimitMiddleware to cap the number of tool calls an agent can make per run.
›Adds ModelFallbackMiddleware and retry_model_request middleware hook for automatic model-level retry and fallback logic.
+16 moreshow less
›Adds file-search middleware enabling retrieval-augmented tool use inside the middleware chain.
›Adds ContextEditingMiddleware for programmatically editing the agent's context window mid-run.
›Adds wrap_model_call and wrap_tool_call decorator hooks (with async support) for intercepting and mutating model and tool calls.
›Adds before_agent and after_agent lifecycle hooks on create_agent for pre/post-agent execution logic.
›Adds ToolRuntime and generic ToolRuntime[ContextT, StateT] injection into tool nodes, accessible via the runtime argument.
›Adds LLM-based tool-selection middleware (add llm selection middleware) for dynamic routing of tool calls.
›Adds a tool emulator enabling client-side simulation of server-side tool calls.
›Adds Human-in-the-Loop (HITL) middleware with description generation, interrupt-on-approval patterns, and a refactored HITL API.
›Adds dynamic system prompt middleware for runtime prompt injection.
›Adds async support to create_agent, wrap_model_call, and wrap_tool_call.
›Adds middleware support directly in create_agent via a new decorator pattern for dynamically generated middleware.
›Adds model_call_limits capability to cap total model invocations per agent run.
›Adds PEP 604 (| union) syntax support in tool node error handlers.
›Adds Python 3.14 support.
›Renames create_react_agent to create_agent as the canonical entry point for building agents.
›Drops Python 3.9 support.
└──▷ BREAKING ON UPGRADE
!Python 3.9 is no longer supported; the minimum supported version is now Python 3.10.
!create_react_agent is renamed to create_agent; existing code calling create_react_agent will break.
!PlanningMiddleware is renamed to TodoListMiddleware; references to PlanningMiddleware will break.
!ToolNode is removed from create_agent and from the agents namespace; callers that passed a ToolNode to create_agent will break.
!Global state helpers are removed from the langchain-v1 namespace (moved to langchain-classic / langchain-core); any import of those globals from langchain_v1 will break.
!The model_request node is renamed to model; any graph or config referencing the model_request node name will break.
!The injected tool argument key changes from tool_runtime to runtime; middleware or tools reading tool_runtime from injected state will break.
$ git clone --branch langchain-openai==1.0.0 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain-openai==1.0.0
›Adds OpenAI moderation middleware, enabling content moderation to be applied as a processing layer in LangChain-OpenAI pipelines.
›Populates OpenAI service tier token details in model responses, surfacing per-tier usage metadata for cost and quota tracking.
›Enables stream_usage by default when using the default OpenAI base URL and client, so token usage is reported during streaming calls.
langchain-mistralai 1.0.0 adds reasoning support and v1 content handling for Mistral models.
└──▷ GET THIS VERSION
$ git clone --branch langchain-mistralai==1.0.0 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain-mistralai==1.0.0
›Supports Mistral reasoning feature and v1 content format in langchain-mistralai.
langchain-groq 1.0.0 adds support for built-in tools in message content.
└──▷ GET THIS VERSION
$ git clone --branch langchain-groq==1.0.0 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain-groq==1.0.0
›Supports built-in tools in message content for Groq integrations, enabling tool-use responses to be parsed directly from message content blocks.
›Allows overriding ls_model_name from kwargs at invocation time across LangChain core.
langchain-anthropic 1.0.0 adds ShellToolMiddleware, ClaudeBashToolMiddleware, and async middleware support.
└──▷ GET THIS VERSION
$ git clone --branch langchain-anthropic==1.0.0 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain-anthropic==1.0.0
›Adds ShellToolMiddleware and ClaudeBashToolMiddleware classes for intercepting and controlling shell/bash tool calls in Anthropic-powered agents.
›Adds async implementation to middleware, enabling non-blocking middleware execution in async LangChain pipelines.
›Expands the middleware surface with additional Anthropic-specific middleware options migrated into langchain_anthropic.
LangChain standard-tests 1.0.0 adds parametrized tool-calling tests and configurable output_version for integration test suites.
└──▷ GET THIS VERSION
$ git clone --branch langchain-tests==1.0.0 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain-tests==1.0.0
›Adds parametrization of the tool-calling test in the standard test suite, allowing integration test authors to cover multiple tool-calling scenarios in a single test run.
›Enables parametrization of output_version in standard tests, letting library authors test against multiple output format versions without duplicating test classes.
LangChain 1.0.0rc2 ships middleware hooks, injected runtime, HITL patterns, PII/retry/fallback middleware, and async agent support via create_agent.
└──▷ GET THIS VERSION
$ git clone --branch langchain==1.0.0rc2 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain==1.0.0rc2
›Adds ToolRuntime and generic ToolRuntime[ContextT, StateT] injection into agent tool nodes, configurable via the runtime argument.
›Adds tool retry middleware, exposing a retry hook in the middleware chain for transient tool failures.
›Adds wrap_model_call and wrap_tool_call middleware hooks (both sync and async) for intercepting and modifying model and tool invocations.
›Adds before_agent and after_agent lifecycle hooks to the middleware API.
›Adds retry_model_request middleware hook and ModelFallbackMiddleware for model-level fallback on failure.
+18 moreshow less
›Adds ToolCallLimitMiddleware to enforce per-session tool call limits.
›Adds PIIMiddleware for detecting and handling PII in model I/O.
›Adds LLM-selection middleware (add llm selection middleware) for dynamic model routing.
›Adds Context Editing Middleware for in-flight context manipulation.
›Adds TodoListMiddleware (formerly PlanningMiddleware) for multi-step planning inside the agent graph.
›Adds dynamic system prompt middleware, allowing prompts to be generated or modified at runtime.
›Adds async support for create_agent, enabling fully async agent graphs.
LangChain Core 1.0.0rc3 adds PDF tool messages, AWS Bedrock document blocks, VertexAI content, and several new utility capabilities.
└──▷ GET THIS VERSION
$ git clone --branch langchain-core==1.0.0rc3 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain-core==1.0.0rc3
└──▷ USE IT
Include message IDs when converting LangChain messages to OpenAI format, useful for correlating tool call results.
python
from langchain_core.messages.utils import convert_to_openai_messages
openai_messages = convert_to_openai_messages(messages, include_id=True)
Sanitize text before inserting into PostgreSQL to avoid NUL byte DataErrors.
python
from langchain_core.utils import sanitize_for_postgres
clean_text = sanitize_for_postgres(raw_text)
cursor.execute('INSERT INTO docs (content) VALUES (%s)', (clean_text,))
›Adds include_id optional parameter to convert_to_openai_messages() to control whether message IDs are included in OpenAI-format output.
›Adds id field to Document objects passed to filter callbacks in InMemoryVectorStore similarity search.
›Adds web_search to the recognized OpenAI tools list.
›Adds sanitize_for_postgres utility function to strip PostgreSQL NUL bytes that cause DataError.
›Adds support for PDF inputs in ToolMessage content blocks (via standard-tests integration).
+12 moreshow less
›Adds support for AWS Bedrock document content blocks in msg_content_output.
›Adds support for VertexAI standard content blocks in message handling.
›Includes original block type in server tool results for google-genai integrations.
›Adds ls_model_name override capability from kwargs in model tracing.
›Allows custom Mermaid diagram URL via overridable parameter in graph visualization.
›Adds a permissive deserialization option to handle looser object structures.
›Supports PromptTemplate addition for formats other than f-string.
›Exposes recognized block types for ToolMessage to consumers.
›Adds SHA-1 warning and additional hashing options to the indexing API.
›Zeroes out token costs for cache hits in token usage accounting.
›Traces response body on error for improved observability.
›Injects ToolRuntime and generic ToolRuntime[ContextT, StateT] into tool execution context.
└──▷ BREAKING ON UPGRADE
!BaseMemory has been deleted from langchain-core and moved to langchain-classic.
!Items previously 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 deleted.
!Global state previously in langchain-v1 has been removed; globals updated in langchain-classic and langchain-core.
!Deprecated items (marked for removal) across the codebase have been deleted.
Continue 1.3.0 adds background agents, full CLI secrets parsing, and Anthropic cache-write token reporting.
└──▷ 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]
›Adds full secrets parsing for all CLI flags, so secrets stored in the secrets manager are expanded across every command-line option.
›Adds background agent creation and execution, enabling agents to run without an active interactive session.
›Reports cache write tokens from the Anthropic OpenAI adapter, surfacing prompt-caching cost and usage data.
Continue 1.26.0 adds background agents, full CLI secrets parsing, and Anthropic cache-write token reporting.
└──▷ 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]
›Adds full secrets parsing for all CLI flags, so secret values are resolved across every flag at invocation time.
›Adds background agent creation and execution, enabling agents to run asynchronously without blocking the foreground session.
›Reports cache write tokens from the Anthropic OpenAI-compatible adapter, surfacing prompt-cache cost data previously omitted.
An open-source AI agent that brings the power of Gemini directly into your terminal.
gemini-cli v0.10.0-preview.0 adds failed-response retry, context-overflow protection, GOOGLE_CLOUD_PROJECT_ID fallback, and a new code guide command.
└──▷ GET THIS VERSION
$ git clone --branch v0.10.0-preview.0 https://github.com/google-gemini/gemini-cli.git
# already have the repo? check out this version:$ git checkout v0.10.0-preview.0
└──▷ TRY IT
Use GOOGLE_CLOUD_PROJECT_ID as a fallback when GOOGLE_CLOUD_PROJECT is not set, avoiding manual flag overrides.
Weaviate v1.33.1 adds debug endpoints for shard/lock monitoring, image support in generative-Cohere, slow-query sampling, and a broad new set of internal observability metrics.
└──▷ GET THIS VERSION
$ git clone --branch v1.33.1 https://github.com/weaviate/weaviate.git
# already have the repo? check out this version:$ git checkout v1.33.1
›Adds debug endpoints for shard and lock status monitoring (PR #9402).
›Adds image support in the generative-cohere module, enabling multimodal generative queries.
›Adds 1% sampled queries to the slow query log for lightweight production query tracing.
THE AI TOOLCHAIN