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.
AutoGPT Platform v0.6.65 adds global Cmd+K search, read-only graph viewing, webhook triggers, and MCP argument expansion.
└──▷ GET THIS VERSION
$ git clone --branch autogpt-platform-beta-v0.6.65 https://github.com/Significant-Gravitas/AutoGPT.git
# already have the repo? check out this version:$ git checkout autogpt-platform-beta-v0.6.65
Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.
CrewAI 1.15.0 adds declarative Flow definitions, DMN mode, composite each actions, and full CLI TUI support for conversational flows.
└──▷ GET THIS VERSION
$ git clone --branch 1.15.0 https://github.com/crewAIInc/crewAI.git
# already have the repo? check out this version:$ git checkout 1.15.0
›Adds unified declarative flow loading so flows can be defined and loaded from config rather than pure code.
›Adds declarative Flow CLI support, enabling flow kickoff and management via CLI commands.
›Adds each composite action to FlowDefinition for iterating over steps within a flow.
›Adds optional if expression to each.do steps for conditional iteration logic inside flows.
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 v2.1.193 adds shell classifier routing, OTEL response logging, live path autocomplete, and memory-pressure reaping for background shells.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.193 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.193
└──▷ USE IT
Enforce auto-mode classification on every shell command — useful in enterprise deployments that want consistent policy enforcement across all Bash/PowerShell invocations, not just detected code-execution patterns.
json
# In your Claude Code settings (e.g. settings.json)
{ "autoMode.classifyAllShell": true }
Capture full assistant response text in your OTEL pipeline for audit or compliance logging.
📍OTEL_LOG_ASSISTANT_RESPONSES=1 claude ...
Prevent idle background shell commands from being reaped under memory pressure if your workflow depends on long-lived idle shells.
📍CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1 claude ...
›New autoMode.classifyAllShell setting routes all Bash/PowerShell commands through the auto-mode classifier, not just arbitrary-code-execution patterns.
›Adds auto-mode denial reasons to the transcript, denial toast, and /permissions recent denials for better visibility into blocked actions.
›New claude_code.assistant_response OpenTelemetry log event captures the model's full response text (redacted by default; enable with OTEL_LOG_ASSISTANT_RESPONSES=1).
›Adds live file path autocomplete in bash mode (!) for faster command entry.
›Adds a startup notice when MCP servers need authentication, pointing users to /mcp.
+1 moreshow less
›Adds automatic memory-pressure reaping for idle background shell commands (disable with CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1).
└──▷ BREAKING ON UPGRADE
!The claude_code.assistant_response OTEL log event is emitted by default when OTEL_LOG_USER_PROMPTS is enabled — deployments that already log prompt content will also start receiving response content on upgrade. Set OTEL_LOG_ASSISTANT_RESPONSES=0 to keep prompts-only logging.
Lightweight coding agent that runs in your terminal
Codex CLI gains default MCP tool search, macOS system proxy support, and dark-mode plugin logos.
└──▷ GET THIS VERSION
$ git clone --branch rust-v0.142.2 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.142.2
└──▷ USE IT
Route Codex CLI traffic through a corporate PAC/WPAD proxy on macOS without manual proxy configuration.
yaml
# In your Codex config, enable system proxy resolution:
respect_system_proxy: true
›Enables tool search for MCP tools by default when the model/provider supports it, improving tool discovery while staying compatible with older models.
›Adds macOS system proxy, PAC, and WPAD resolution when respect_system_proxy is enabled, letting the CLI route through corporate proxies automatically.
›Supports dark-mode logos for plugins via local manifests and remote catalogs.
›Surfaces richer safety-buffering UI using server-provided visibility and faster-model metadata.
SYCL backend gains --split-mode tensor for dual-GPU tensor parallelism with BF16-compressed all-reduce
└──▷ GET THIS VERSION
$ git clone --branch b9788 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b9788
›Adds --split-mode tensor support to the SYCL backend, enabling tensor parallelism across dual Intel Arc GPUs without OneCCL or new build dependencies.
›Implements a two-path all-reduce: FP32 direct memcpy for small tensors (nelem < 32768) and BF16-compressed cross-device transfer for large tensors (nelem >= 32768), halving PCIe bandwidth for the common large-tensor case.
›Delivers up to +78.6% token-generation throughput on Llama-3.3-70B Q4_K_M versus layer-split mode on dual Intel Arc Pro B70 hardware.
Phoenix v17.12.0 adds a direct server agent endpoint, opt-in user ID on traces, CI eval testing API for JS, and sub-agent progress streaming.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v17.12.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v17.12.0
›Adds CI eval testing API to the JavaScript phoenix-client library for automated evaluation testing in CI pipelines.
›Adds a direct server agent endpoint for agent interactions.
›Attaches opt-in user.id to Phoenix Intelligence (pxi) traces to correlate agent activity with individual users.
›Streams call_subagent progress in real time so agent task status is visible as it executes.
›Adds a copyable trace ID in the trace details dialog header within the Experiments UI.
Composio @composio/[email protected] adds per-request AbortSignal cancellation across all SDK methods and cooperative cancellation for custom tools.
└──▷ 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
Cancel a slow tool search after 5 seconds so a stalled agent does not block indefinitely.
›Adds a ComposioRequestOptions trailing argument ({ signal?: AbortSignal }) to all public SDK methods — tools.get, tools.execute, toolkits.get, authConfigs.list/create/get/update/delete, connectedAccounts.*, triggers.*, mcp.*, toolRouter.*, and toolRouterSession.* — so callers can cancel long-running requests without blocking an agent indefinitely.
›Adds ComposioRequestCancelledError as a typed, instanceof-detectable error class; any APIUserAbortError, AbortError, or DOMException(name='AbortError') from the underlying fetch is normalized to it, and catch paths in tools.execute, tools.getRawComposioToolBySlug, and toolkits.get re-throw it rather than remapping to ComposioToolExecutionError / ComposioToolNotFoundError / ComposioToolkitFetchError.
›Exposes SessionContext.signal (the caller's AbortSignal) inside Tool Router custom tools via experimental_createTool, enabling cooperative mid-execution cancellation by wiring ctx.signal into any abortable I/O.
›Adds a pre-execute signal check for custom tools: if signal.aborted is true before user execute runs, the SDK throws ComposioRequestCancelledError and never invokes user code.
›Adds search and showDisabled filter parameters to authConfigs.list().
+3 moreshow less
›Adds provider-agnostic JSON-schema property-key sanitizer utilities: sanitizeSchemaPropertyKeys(schema, policy), restoreOriginalKeys(value, mapping), mappingHasRenames(mapping), and the KeyMapping / KeySanitizationPolicy types, consumed by the @composio/anthropic provider to handle per-provider key character/length constraints.
›Exposes jsonSchemaToZodShape via the @composio/core/utils/json-schema subpath export, and adds correct conversion of both zod/v3 and Zod v4 schemas to JSON Schema for custom tools.
›Preserves root $defs / definitions blocks on tool parameter schemas and dereferences them in the Node file modifier, enabling auto file upload/download detection when file_uploadable or file_downloadable is hidden behind an internal $ref.
└──▷ BREAKING ON UPGRADE
!The packages are now ESM-only; CommonJS entrypoints and .cjs artifacts are removed. Node.js 22.22.3 or newer is required; CommonJS callers can only rely on Node's native require(esm) interop.
!The uuid field is removed from AuthConfigRetrieveResponse and AuthConfigListResponse; consumers must use id instead.
THE AI TOOLCHAIN