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 v2.8.6 adds Smallest AI TTS tools, OpenSearch vector DB, and a new AgentOS metrics-refresh status endpoint.
└──▷ GET THIS VERSION
$ git clone --branch v2.8.6 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:$ git checkout v2.8.6
└──▷ USE IT
Add Smallest AI text-to-speech capability to an agent and save audio output to disk.
python
from agno.tools.smallest import SmallestTools
from agno.agent import Agent
agent = Agent(tools=[SmallestTools(voice_id='<voice_id>', model='lightning_v3.1', output_file='output.wav')])
agent.run('Convert this text to speech: Hello from Agno!')
Poll AgentOS for completion of a background metrics refresh instead of waiting for a blocking response.
$ curl -X POST 'http://localhost:8000/metrics/refresh?background=true'
# Then poll until completed:
curl 'http://localhost:8000/metrics/refresh/status'
Use OpenSearch as a vector database backend for hybrid search in a retrieval workflow.
›Adds SmallestTools toolkit in agno for Smallest AI text-to-speech, exposing text_to_speech (returns audio as a ToolResult artifact, optionally saved to disk) and get_voices; supports lightning_v3.1 and lightning_v3.1_pro models.
›Adds OpenSearch vector database support at agno.vectordb.opensearch, installable via the agno[opensearch] extra, with vector, keyword, and hybrid search in both sync and async variants; includes a run_opensearch.sh script for local setup.
›Adds GET /metrics/refresh/status endpoint to AgentOS to poll the state of a background metrics refresh, returning idle, running, completed, or failed with started_at, finished_at, and error fields.
›Exposes AgentOSClient.get_metrics_refresh_status() as the client-side counterpart to GET /metrics/refresh/status.
›Adds ?background=true query parameter to POST /metrics/refresh, returning HTTP 202 immediately and running the refresh as a single-flight background task per database.
+1 moreshow less
›Caches the Pydantic version lookup during tool wrapping, cutting repeated-wrap overhead from 65.9 ms to 11.0 ms per 100 wraps.
Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.
CrewAI 1.15.9 surfaces tool failures accurately, adds FlowFailedEvent, and introduces progressive disclosure for skills.
└──▷ GET THIS VERSION
$ git clone --branch 1.15.9 https://github.com/crewAIInc/crewAI.git
# already have the repo? check out this version:$ git checkout 1.15.9
›Emits FlowFailedEvent when a flow execution fails, enabling downstream handlers to react to flow-level errors.
›Surfaces tool failures as actual failures instead of silently reporting them as success, improving error visibility.
›Implements progressive disclosure for skills, controlling how skill details are revealed during execution.
checkpoint-postgres 3.1.1 adds opt-in omit_expired flag to skip expired checkpoint rows on read.
└──▷ GET THIS VERSION
$ git clone --branch checkpointpostgres==3.1.1 https://github.com/langchain-ai/langgraph.git
# already have the repo? check out this version:$ git checkout checkpointpostgres==3.1.1
›Adds opt-in omit_expired parameter to checkpoint read operations, allowing callers to skip expired rows and avoid processing stale state.
LangGraph checkpoint-postgres 3.1.1 adds opt-in omit_expired flag to skip expired checkpoint rows on read.
└──▷ GET THIS VERSION
$ git clone --branch checkpointpostgres==3.1.1 https://github.com/langchain-ai/langgraph.git
# already have the repo? check out this version:$ git checkout checkpointpostgres==3.1.1
›Adds omit_expired opt-in parameter to checkpoint reads, allowing callers to skip expired rows instead of returning them.
PydanticAI v2.21.0 adds per_request_input_tokens_limit to UsageLimits for per-call token budgets.
└──▷ GET THIS VERSION
$ git clone --branch v2.21.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:$ git checkout v2.21.0
└──▷ USE IT
Prevent any single LLM call from consuming more than a set number of input tokens, useful for guarding against unexpectedly large context windows in multi-turn agents.
python
from pydantic_ai import Agent
from pydantic_ai.usage import UsageLimits
agent = Agent('openai:gpt-4o')
result = agent.run_sync(
'Summarize this document.',
usage_limits=UsageLimits(per_request_input_tokens_limit=4000),
)
›Adds per_request_input_tokens_limit field to UsageLimits to cap input tokens on a per-request basis, independently of aggregate limits.
$ git clone --branch v1.0.77 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:$ git checkout v1.0.77
└──▷ TRY IT
Force browser-based OAuth login on a machine where you want to avoid the device-code flow.
$ copilot login --web-flow
Force device-code login on a local terminal when you prefer not to open a browser.
$ copilot login --device-code
›Adds --web-flow and --device-code flags to copilot login to force browser-based or device-code OAuth mode; browser OAuth is now the default on local interactive terminals, device code remains default on remote/headless terminals; mode is also selectable via the interactive /login command.
›Adds support for enforcing managed sandbox policy via macOS and Windows native MDM settings.
›Adds Ctrl+G keyboard shortcut to open your editor for freeform ask_user answers without closing the prompt.
›Unconditional autopilot approval now disables the sandbox for the current session when bypass is allowed.
›Allows reasoning effort to be omitted so the server can select the default.
Enable the multi-session Sessions sidebar to manage concurrent Copilot sessions during a complex investigation.
$ /experimental on
›Adds sidebar.hoverFocus and sidebar.accentActiveSession config keys to the split-view sidebar: hover-to-focus is off by default (opt in with sidebar.hoverFocus), and the active session card is accented by default (opt out with sidebar.accentActiveSession).
›Adds stayInAutopilot setting to control whether autopilot remains selected after task_complete; set to false to return to interactive mode after each task.
›Adds /limits predict subcommand to suggest a session AI-credit limit based on similar past sessions.
›Adds enable/disable controls in /plugins for plugins, instructions, agents, LSP servers, and hooks.
›Adds support for the grok-4.5 model.
+14 moreshow less
›Adds a new Sessions sidebar for managing multiple concurrent sessions — switch between them, spawn new ones, and see their status; enable with /experimental on.
›Adds a directable queue manager to reorder, edit, remove, repeat, and immediately send queued messages; Ctrl+C removes your own newest queued message.
›Enterprise administrators can enforce a restrictive sandbox floor via managed settings that tighten (but never loosen) the user's sandbox policy; the /sandbox dialog surfaces org-configured managed values with locked fields and managed filesystem paths.
›Supports configurable timed refreshes for custom status-line commands.
›Changing the mouse setting mid-session now takes effect immediately via /settings mouse on|off or the /settings dialog, without requiring a restart.
›web_fetch now follows HTTP redirects, requesting permission for redirect targets on different origins and showing redirect origin context.
›web_fetch routes through the configured sandbox proxy when outbound is allowed, and denies egress when network.allowOutbound is false.
›Sandbox denied paths are enforced for relative and symlinked entries on macOS and Linux.
›MCP tools load faster from definition-scoped snapshots, with process-wide and per-server cache opt-outs.
›Renders inline images in Rio terminals that support Kitty graphics.
›Voice mode pauses playing media before recording and resumes it afterward on macOS and Windows.
›Shows the number of active scheduled prompts in the footer.
›Queued mid-turn /model changes are now applied after the current response finishes.
›The /instructions picker now respects --no-custom-instructions.
$ git clone --branch rust-v0.147.0-alpha.2 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.147.0-alpha.2
›Adds MCP 2026-07-28 discovery support and completes full MCP 2026 client support.
WebGPU EP v0.2.1 adds 2-bit GatherBlockQuantized, Gemma 4 support, Qwen3 fusions, and expanded FlashAttention capabilities.
└──▷ GET THIS VERSION
$ git clone --branch plugin-ep-webgpu/v0.2.1 https://github.com/microsoft/onnxruntime.git
# already have the repo? check out this version:$ git checkout plugin-ep-webgpu/v0.2.1
›Adds 2-bit support to GatherBlockQuantized for more aggressive quantized-path compression on WebGPU.
›Adds Opset 24 and KV-shared decoder layer support for Gemma 4 model paths.
›Adds QKV and MLP fusions for Qwen3-style models, plus Q/K RMSNorm fusion into GroupQueryAttention.
›Extends FlashAttention decode kernels to handle any sequence length, removing previous length constraints.
›Adds QKV bias support for FlashAttention in MultiHeadAttention.
+6 moreshow less
›Enables dynamic max_k_step for NVIDIA hardware in the FlashAttention path.
llama.cpp server gains support for input embeddings to drive next-token sampling
└──▷ GET THIS VERSION
$ git clone --branch b10199 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b10199
›Adds inp embd (input embedding) support to the server, enabling callers to supply raw embedding vectors directly to drive sampled token generation instead of text prompts.
Phoenix v19.11.0 adds project evaluation metric charts, subagent tool call counting, and a pinned note-taking bar for span details.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v19.11.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v19.11.0
›Adds project evaluation metrics charts to visualize eval results at the project level.
›Counts subagent tool calls online during evaluations, enabling real-time tracking of tool usage in agentic workflows.
›Defers metric chart loading and removes the chart selection cap, allowing more charts to be displayed simultaneously without blocking.
›Displays tool and tool call counts in LLM span card headers for at-a-glance agentic span inspection.
›Adds a pinned note-taking bar to the span details view for persistent inline annotations.
+2 moreshow less
›Moves root-span scoping into the filter condition UI, integrating trace-root filtering with the standard filter builder.
›Updates the Monty provider icon to use the Pydantic logo in the sandboxes UI.
THE AI TOOLCHAIN