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.
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 v2.1.228 hardens synced skills, improves cross-session messaging, and lets newer models overwrite unread files.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.228 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.228
›Hardened skills synced from claude.ai: they no longer shadow local commands or MCP prompts, descriptions are sanitized and labeled, and skill bodies no longer run ! commands or expand @ files on your machine.
›Improved cross-session messages: sender and body now display inline instead of a collapsed line, and messages to Remote Control sessions on other machines show your Remote Control session name as the sender.
›The Write tool now allows newer models to overwrite an existing file they haven't read this session, matching the Edit tool's rules; older models still require a read first.
›Improved Vertex AI credential handling: expired or missing Google Cloud credentials now fail within seconds instead of retrying for minutes.
›Improved compaction progress: the retry countdown and stall hint now appear during compaction instead of only a progress bar.
Cline v4.1.8 adds manual Vertex model ID entry, Fable 5 support, and consolidates auto-approve controls.
└──▷ GET THIS VERSION
$ git clone --branch v4.1.8 https://github.com/cline/cline.git
# already have the repo? check out this version:$ git checkout v4.1.8
›Allows entering any Vertex model ID by hand, including models not yet listed in the catalog.
›Supports Fable 5 on Vertex.
›Shows the full model catalog for every Vertex region instead of a hardcoded global-endpoint list, with request-time failure and recovery guidance when a selected model is unavailable in the chosen region.
›Makes the auto-approve menu the single source of truth for unattended runs; the Yolo Mode toggle is removed as it had no effect on the approval path.
└──▷ BREAKING ON UPGRADE
!The Yolo Mode toggle is removed; existing setups that had Yolo Mode or auto-approve-all enabled are automatically migrated to auto-approving every action.
Lightweight coding agent that runs in your terminal
Codex CLI alpha.7 adds configurable goal token budget limits, session ID exposure to shell commands, and configurable Responses API request metadata.
└──▷ GET THIS VERSION
$ git clone --branch rust-v0.148.0-alpha.7 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.148.0-alpha.7
›Exposes the session ID to shell commands, enabling scripts and tools to reference the current session context.
›Adds configurable Responses API request metadata, allowing custom metadata to be attached to API requests.
›Supports MCP form input in full-access user threads, expanding interactive MCP workflows.
›Adds packagedDefaults as a new lowest-precedence configuration layer, with its source path reported through config diagnostics and the app-server protocol; returns an error when a configured packaged defaults file is missing.
›Honors per-directory bundled skill settings in skills/list requests.
›Simplifies skills.read so that package is the only required argument, resolving the owning orchestrator or executor catalog automatically and defaulting omitted resource values to the package's main SKILL.md.
›Adds the codex.turn.unified_exec.running_processes counter to track background terminal processes remaining at turn completion.
›Validates images before returning view_image output.
›Honors the configured Windows sandbox level for managed networking.
›Runs plugin and skill tests on Windows, extending platform coverage.
└──▷ BREAKING ON UPGRADE
!Config lockfile support has been removed (Remove config lockfile support).
Qwen Code v0.21.10 adds ACP reasoning effort configuration and Web Shell image preview with download support.
└──▷ GET THIS VERSION
$ git clone --branch v0.21.10 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout v0.21.10
›Adds ACP support for configuring reasoning effort levels (Default to Max) via session configuration.
›Web Shell image preview now opens uploaded or pasted images in the artifact panel with download support.
›OpenAI API log cleanup now applies to non-interactive sessions, headless invocations, and daemon modes to reduce disk usage.
›Improves CLI performance by caching persisted session catalogs for organized and filtered daemon lists, with automatic invalidation on metadata changes.
›Web Shell session lists now share cached data across components to reduce network requests.
Qwen Code v0.1.1 adds SSE observability, Mem0 memory writes, voice ASR URLs, inline terminal images, and workflow pause/resume.
└──▷ GET THIS VERSION
$ git clone --branch live-host-v0.1.1 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout live-host-v0.1.1
›Adds SSE stream and client observability to the daemon (feat(daemon): Add SSE stream and client observability).
›Adds a batch skill toggle API to the daemon (feat(daemon): add batch skill toggle API).
LangChain 1.3.15 adds trace_policy on AgentMiddleware, state_schema on wrap_tool_call, LangSmith provider in init_chat_model, and reasoning_effort as a standard chat model parameter.
└──▷ GET THIS VERSION
$ git clone --branch langchain==1.3.15 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain==1.3.15
└──▷ USE IT
Initialize a LangSmith-hosted model directly without provider-specific boilerplate.
python
from langchain.chat_models import init_chat_model
model = init_chat_model(model='langsmith/<your-model>', provider='langsmith')
›Exposes trace_policy parameter on AgentMiddleware to control tracing behavior per agent.
›Adds state_schema parameter to wrap_tool_call for typed state passing in tool calls.
›Adds LangSmith as a supported provider in init_chat_model for direct model initialization.
›Adds reasoning_effort as a standard chat model parameter across providers.
›Filters internal middleware model calls from the messages projection, keeping conversation history clean.
OpenAI Agents SDK v0.20.0 switches the default model to gpt-5.6-luna, adds RunState.add_input() for durable pending input, MCP SDK v2 support, and GA realtime transcription settings.
└──▷ GET THIS VERSION
$ git clone --branch v0.20.0 https://github.com/openai/openai-agents-python.git
# already have the repo? check out this version:$ git checkout v0.20.0
›Adds RunState.add_input() to stage durable user input before a resumed model call, with guardrail, persistence, and serialization support.
›Adds explicit mount credential-exposure acknowledgements to sandbox mount validation, with redacted error contracts that do not serialize credential authority.
›Supports MCP Python SDK v2 alongside v1 across stdio, SSE, and Streamable HTTP transports for local MCP connections.
›Supports GA transcription settings for gpt-live-transcribe, gpt-transcribe, and gpt-realtime-whisper in realtime input transcription.
›Passes run context to custom session implementations.
+3 moreshow less
›Preserves raw usage payloads from provider responses.
›Allows applications to approve unsafe replays explicitly.
›Changes the implicit default model to gpt-5.6-luna; explicit models, run-level overrides, and the OPENAI_DEFAULT_MODEL environment variable continue to take precedence.
└──▷ BREAKING ON UPGRADE
!The implicit default model is now gpt-5.6-luna instead of the previous default; applications that relied on the old implicit default will use a different model on upgrade.
!Applications using custom MCP HTTP authentication or client factories must use the HTTP types owned by the installed MCP major version (v1 or v2), or pin mcp<2, due to the MCP SDK v2 dependency migration.
Get up and running with Kimi-K2.6, GLM-5.2, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
Ollama v0.32.9 adds NVIDIA Nemotron 3.5 Lightning, a 30B MoE model with 3B active parameters for always-on agents.
└──▷ GET THIS VERSION
$ git clone --branch v0.32.9 https://github.com/ollama/ollama.git
# already have the repo? check out this version:$ git checkout v0.32.9
└──▷ TRY IT
Run the new NVIDIA Nemotron 3.5 Lightning model locally for always-on agent workloads.
$ ollama run nemotron-3.5-lightning
›Adds nemotron-3.5-lightning model support via ollama run nemotron-3.5-lightning — a 30B mixture-of-experts model with 3B active parameters designed for always-on agent harnesses.
›Adds the Nemotron 3 model architecture to Ollama's backend, enabling the new model family.
vMLX 1.6.27 adds Muse Glimmer 30B with vision/video support, a 262,144-token prefix cache, and working Anthropic/Ollama reasoning controls.
└──▷ GET THIS VERSION
$ git clone --branch v1.6.27 https://github.com/jjang-ai/vmlx.git
# already have the repo? check out this version:$ git checkout v1.6.27
›Adds support for Anthropic thinking.budget_tokens to actually cap the reasoning chain length at runtime (previously the budget was set only as a template hint and never consumed).
›Adds support for Ollama think string levels ('low', 'medium', 'high') — clients sending these values now enable thinking and select the correct effort level instead of being silently ignored.
›Adds Muse Glimmer 30B support across all three bundles (JANG_2D / 4M / 6M), including its text tower, windowed vision tower, recipient-routed reasoning rail, and ATEM tool dialect.
›Adds single- and multi-image prompt handling for Muse Glimmer 30B, correctly identifying distinct visual elements within the same turn.
›Adds temporal-order video frame reading for Muse Glimmer 30B, so a clip fading from red to blue is answered 'red blue' rather than reversed.
+3 moreshow less
›Exposes four reasoning depth levels — Low / Medium / High / Extra High — in Chat Settings for Muse Glimmer 30B, replacing the previous binary Auto/On toggle.
›Expands the prefix cache index from ~64,000 tokens to 262,144 tokens, enabling long conversations to get cache reuse instead of falling off the end of the index; existing sessions are migrated automatically.
›Aligns paged RAM usage between the app and the command line for vision models, so both code paths now share the same cache behavior.
$ git clone --branch arize-phoenix-client-v3.1.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-client-v3.1.0
›Adds agent session persistence, enabling agent state to be saved and resumed across interactions.
$ git clone --branch arize-phoenix-v20.0.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v20.0.0
›Supports trace_annotations in the span filter DSL for server-side filtering of spans by their annotations.
›Replaces the google-generativeai formatter with the google-genai formatter in the client.
›Adds agent session persistence, enabling multi-turn agent sessions to be stored and resumed.
›Adds semantic coloring to project metric tooltips in the UI.
└──▷ BREAKING ON UPGRADE
!The google-generativeai formatter is replaced by the google-genai formatter; any client code relying on the old formatter will break on upgrade. See MIGRATION.md for details.
Phoenix client v3 adds REST endpoints for API keys, experiment tags, dataset splits, and an OpenAI-compatible chat completions proxy.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-client-v3.0.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-client-v3.0.0
›Adds span_ids filter parameter to get_spans in the Python and TypeScript clients, enabling targeted span retrieval by ID.
›Adds REST CRUD endpoints for user and system API keys with a unified authority model.
›Adds REST endpoints for setting tags on experiments.
›Adds dataset split create, update, and delete REST endpoints.
›Adds a REST endpoint for prompt metadata updates.
+7 moreshow less
›Adds support for updating prompt description and metadata via the client.
›Adds an OpenAI-compatible v1/chat/completions proxy endpoint with server-side credentials, enabling LLM calls without exposing client-side API keys.
›Makes PHOENIX_ENDPOINT the canonical API-access environment variable in the JavaScript client.
›Adds evaluator trace isolation and experiment metadata support to the pytest plugin.
›Replaces the google-generativeai formatter with the google-genai formatter in the client.
└──▷ BREAKING ON UPGRADE
!The google-generativeai formatter has been replaced with google-genai; any code or configuration depending on the google-generativeai formatter will break on upgrade.
Langfuse v4.9.0 adds an instance switcher for self-hosted deployments.
└──▷ GET THIS VERSION
$ git clone --branch v4.9.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:$ git checkout v4.9.0
›Adds a self-hosted instance switcher to the UI, enabling practitioners to navigate between multiple self-hosted Langfuse instances without re-configuring.
Langfuse v4.8.0 lets users approve an agent tool once for an entire conversation session.
└──▷ GET THIS VERSION
$ git clone --branch v4.8.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:$ git checkout v4.8.0
›In the in-app agent, a tool approval can now be granted once for the duration of a whole conversation, removing the need to re-approve on every tool call.
Composio CLI 0.3.3 bakes the toolkit slug catalog locally for faster, offline toolkit resolution.
└──▷ 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]
›Bakes the toolkit slug catalog directly into the CLI binary, enabling toolkit resolution from local knowledge without a catalog network call — unlocking offline and low-latency workflows.
Composio @composio/[email protected] extends ToolSchema to support free-form object roots, patternProperties, and schema-valued additionalProperties.
└──▷ 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 support for free-form object roots, patternProperties, and schema-valued additionalProperties in ToolSchema.parse, so tools that declare dynamic keys now pass those constraints through to every provider's inputParameters instead of having them silently stripped.
›Makes the properties field optional on the public ToolSchema type to reflect valid property-less object schemas.
›Automatically injects type: 'object' into nested JSON Schema nodes that carry properties without an explicit type, ensuring compatibility with strict OpenAPI 3.0 consumers such as Google Gemini.
›Reuses fetched tool schemas when provider-wrapped tools execute, avoiding a redundant retrieval request per tool call.
└──▷ BREAKING ON UPGRADE
!Code that assumed additionalProperties in inputParameters is always a boolean must be widened — ToolSchema.parse now accepts and preserves additionalProperties as a schema object (e.g. { type: 'number' }), and patternProperties can also appear on a parsed schema.
json-schema-to-zod now accepts arbitrary content in property-less objects and correctly scopes patternProperties and additionalProperties validation.
└──▷ 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]
└──▷ USE IT
Preserve arbitrary payload content when converting a free-form object schema, instead of having it rejected.
›Property-less object schemas ({ "type": "object" } or { "properties": {} }) are now open by default, accepting and preserving arbitrary content at the root, nested inside objects, and inside array items — instead of rejecting everything as z.object({}).strict().
›Dynamic key routing now correctly applies each patternProperties pattern only to keys it actually claims, and additionalProperties (schema-valued) applies only to keys unmatched by any declared property or pattern.
└──▷ BREAKING ON UPGRADE
!Converting a property-less object schema ({ "type": "object" } or { "properties": {} }) no longer produces a schema that rejects all non-empty payloads; callers that relied on that closed behavior must now explicitly set additionalProperties: false in their JSON Schema to restore it.
THE AI TOOLCHAIN