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 v2.6.19 adds tool-batch checkpointing with a unified /continue endpoint and a new StudioTool for dynamic agent/team/workflow composition.
└──▷ GET THIS VERSION
$ git clone --branch v2.6.19 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:$ git checkout v2.6.19
└──▷ USE IT
Use raw regex strings in PII guardrail patterns without pre-compiling them.
python
from agno.guardrails import PIIDetectionGuardrail
guardrail = PIIDetectionGuardrail(
custom_patterns=[r'\b\d{3}-\d{2}-\d{4}\b', r'\bACCT-\d{8}\b']
)
›Adds StudioTool toolkit for dynamic composition of agents, teams, and workflows at runtime.
›Adds tool-batch-level checkpointing and a unified /continue endpoint supporting both regenerate and fork-a-run workflows, plus session forking support.
›Extends custom_patterns on PIIDetectionGuardrail to accept raw regex strings in addition to compiled patterns.
›ClickHouse and Pinecone vector DBs now expose their supported search types via get_supported_search_types().
langchain-openrouter 0.2.4 surfaces parallel_tool_calls on bind_tools for concurrent tool execution control.
└──▷ GET THIS VERSION
$ git clone --branch langchain-openrouter==0.2.4 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain-openrouter==0.2.4
›Adds parallel_tool_calls parameter to bind_tools on the OpenRouter integration, allowing callers to control whether the model may invoke multiple tools concurrently.
PydanticAI v2.0 stable: capabilities primitive, gemini-embedding-2, AG-UI deferred tools, and new model settings.
└──▷ GET THIS VERSION
$ git clone --branch v2.0.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:$ git checkout v2.0.0
›Adds xai_max_turns to XaiModelSettings to cap the number of turns for xAI model runs.
›Adds gemini-embedding-2 embedding model support.
›Adds google_task text-prefix conditioning for gemini-embedding-2 embeddings to tune retrieval, classification, and other task types.
›Maps AG-UI interrupts to DeferredTools in AGUIAdapter, enabling human-in-the-loop interrupt handling in AG-UI workflows.
›Introduces V2 stable with capabilities as a core composable primitive, bundling an agent's tools, hooks, instructions, and model settings into a single unit.
+1 moreshow less
›Adds cerebras_clear_thinking setting and emits reasoning_effort='none' for Cerebras to suppress chain-of-thought output.
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 v2.1.187 adds sandbox credential blocking, org model restrictions, and mouse support in fullscreen menus.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.187 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.187
›Adds sandbox.credentials setting to block sandboxed commands from reading credential files and secret environment variables.
›Adds org-configured model restrictions enforced across the model picker, --model, /model, and ANTHROPIC_MODEL, with a clear 'restricted by your organization's settings' message.
›Adds mouse click support for selecting menus (permission prompts, /model, /config, etc.) in fullscreen mode.
›Improves /install-github-app to make GitHub Actions workflow setup optional — install just the GitHub App and skip the workflow/secret steps.
›Improves /btw with ←/→ arrow navigation to step through earlier answers.
+1 moreshow less
›Improves /plugin to surface recently unused plugins for easier cleanup.
Copilot CLI v1.0.64 adds worktree sessions, /diagnose, /mcp registry, CSV MCP output, and a batch of new commands and aliases.
└──▷ GET THIS VERSION
$ git clone --branch v1.0.64 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:$ git checkout v1.0.64
└──▷ TRY IT
Pick up exactly where you left off in the most recent session for the current repo without hunting for a session ID.
$ copilot --resume
Run a security review on the current codebase — now available to all users without needing the experimental flag.
$ /security-review
›Adds experimental --worktree [name] (-w) flag (enable with /experimental) that creates or reuses a git worktree under <repo>.worktrees/ and starts the session inside it.
›Adds --continue and --resume flags that select the most recent session for the current repository.
›Shows --remote-export and --no-remote-export in --help output.
›OpenTelemetry: chat spans after a successful compaction now carry gen_ai.conversation.compacted=true, with the summary emitted as a CompactionPart in gen_ai.input.messages.
›OpenTelemetry GenAI spans now emit gen_ai.usage.cache_read.input_tokens, gen_ai.usage.cache_creation.input_tokens, and gen_ai.usage.reasoning.output_tokens per the GenAI semantic conventions spec.
+52 moreshow less
›OpenTelemetry GenAI chat spans emit gen_ai.request.reasoning.level for the configured reasoning effort.
›Adds /diagnose command to analyze session logs.
›Adds /mcp registry installation for browsing and installing MCP servers.
›Adds /branch as an alias for /fork, matching Claude Code's command naming.
›Adds /loop as an alias for the /every command.
›Adds CSV output support for MCP tools.
›Makes /security-review available to all users without --experimental.
›Adds tab completion for /agent names.
›Configures the rubber-duck subagent in /subagents, including a complementary model strategy that picks an opposite-family model.
›Adds model family aliases (opus, sonnet, haiku, gpt, gemini) in the model setting.
›Adds Ctrl+Backspace binding in /terminal-setup for Windows Terminal.
›Adds SDK support for host-provided OAuth tokens for remote MCP servers.
›Supports static OAuth client overrides, including client secrets, for MCP server authentication.
›Adds websocket responses support for BYOK OpenAI-compatible providers.
›Adds an option to bypass the sandbox for shell commands.
›Adds mouse click and double-click selection to paginated lists.
›Adds inline image rendering in the CLI.
›Adds argument-hint frontmatter support for skills.
›Adds a Folder column to the resume picker to show each session's working directory.
›Adds a setting to hide the conversation scrollbar.
›Sets an HTTP(S) proxy via a user setting.
›Shows installed plugin MCP servers in mcp commands.
›Discovers MCP servers provided by installed plugins.
›Marketplace plugins that declare MCP servers in marketplace.json now authenticate correctly with OAuth.
›Adds Ctrl+Q to enqueue a prompt while a completion picker is open.
›Autopilot mode auto-handles elicitation, ask_user, sampling, and permission prompts (including on launch with --autopilot and during continuation turns) instead of surfacing dialogs.
›Autopilot mode returns to interactive mode after the agent calls task_complete.
›Applies MCP org policy when sessions create or reload MCP servers.
›Respects deferTools on MCP servers configured in custom agent frontmatter.
›Custom agents using a tools wildcard * now respect the deferredToolLoading opt-in switch.
›/diff shows a session diff of Copilot's changes in non-git folders.
›Syntax-highlights shell commands in the timeline.
›Shows per-model token totals in /usage and speeds up large history scans.
›Shows the pay-as-you-go additional usage budget at launch, refreshes it after a request is rejected for hitting the additional spend limit, and shows a friendly message when the additional usage limit is reached.
›Resumed sessions reproduce the original attached-file references even if those files later change on disk, avoiding prompt-cache resets.
›Attached images and PDFs persist across session resume even if the source file is later changed or deleted.
›Images returned by tools stay visible to the model across later turns and after resuming a session.
›Allows disabling task and explore built-in subagents.
›Forwards typed rejection feedback from preToolUse prompts to the model.
›Automatically follows system light and dark mode changes.
›Preserves open canvas instances across reconnects and restarts.
›Resume sessions by name even when the name contains spaces.
›Hides unsupported slash commands in remote-hosted sessions.
›Allows /allow-all to work in relay sessions.
›Allows /update to restart sessions started with copilot -r.
›Remote export keeps running after /clear and /session info keeps the task URL.
›Click filenames in /diff tree to jump to that file's first change.
›Experimental: in the compact timeline, click a tool-call or reasoning row to expand or collapse just that entry.
›Loads workspace MCP servers in trusted server-mode sessions.
›The /settings search field supports readline editing keys and cursor movement.
›Preserves Markdown blockquotes in /share exports.
›Keeps task companion tools available to custom agents that use the task or agent alias.
└──▷ BREAKING ON UPGRADE
!OpenTelemetry GenAI span attribute names for cache and reasoning token counts are corrected to gen_ai.usage.cache_read.input_tokens, gen_ai.usage.cache_creation.input_tokens, and gen_ai.usage.reasoning.output_tokens; pipelines consuming the previous underscore-separated names will stop receiving data.
!Terminal-reported color scheme support is removed.
!The legacy intent-reporting tool is removed from the CLI.
!The bogus Ctrl+Enter VS Code keybinding created by old /terminal-setup is removed.
Pi v0.80.0 adds Ctrl+J as a newline shortcut and restructures the pi-ai extension API around a new provider-factory model.
└──▷ GET THIS VERSION
$ git clone --branch v0.80.0 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:$ git checkout v0.80.0
›Adds Ctrl+J as a default newline keybinding alongside Shift+Enter.
›Renames the zai provider label to 'ZAI Coding Plan (Global)' for clarity in the provider list.
└──▷ BREAKING ON UPGRADE
!The @earendil-works/pi-ai/base and @earendil-works/pi-agent-core/base entrypoints have been removed; use the root packages with explicit Models provider factories instead.
!The pi-ai global API (stream/complete/completeSimple, getModel/getModels/getProviders, registerApiProvider, getEnvApiKey, …) has moved off the @earendil-works/pi-ai root entrypoint to @earendil-works/pi-ai/compat; extension sources that typecheck against pi-ai's published types must switch those imports to @earendil-works/pi-ai/compat or migrate to the new createModels()/provider-factory API (the compat entrypoint and loader alias will be removed in a future release).
Qwen Code v0.19.0 adds voice dictation, revivable sub-agents, an Artifact HTML publisher, Dynamic Workflows, and MCP resource browsing.
└──▷ GET THIS VERSION
$ git clone --branch v0.19.0 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout v0.19.0
›Adds fastOnly and voiceOnly flags to model definitions to hide models from the main model picker — keeping fast and voice-specific models out of the primary list.
›Enables preserve_thinking by default for the DashScope provider so chain-of-thought tokens are retained automatically.
›Adds an [HH:MM:SS] timestamp display before each assistant turn in the CLI (optional).
›Adds a thinking block viewer in the TUI, toggled with Alt+T to expand or collapse model reasoning blocks.
›New Artifact tool lets the agent publish interactive HTML pages directly from a session.
+10 moreshow less
›Supports an artifact auto-open setting so published HTML artifacts open automatically.
›Adds voice dictation with native audio capture, streaming transcription, and biasing support.
›Introduces revivable background sub-agents with a configurable subagent transcript TTL.
›Completes the Dynamic Workflows port — supporting resume, saved workflows, keyword triggers, and notifications.
›Adds daemon session branching in the web-shell interface.
›Adds browsing of MCP server resources inside the /mcp dialog.
›Supports installing extensions from archive sources.
›Respects configurable agent ignore files so the agent skips paths defined in project-level ignore configs.
›Desktop file preview now opens in a resizable side panel instead of fullscreen.
›TUI removes tool group borders and auto-collapses completed tool results for a cleaner conversation view.
LocalAI 4.5.0 adds four new backends (depth-anything, ced, supertonic, privacy-filter.cpp), speaker-aware realtime voice with compaction, and concurrent serving on by default.
└──▷ GET THIS VERSION
$ git clone --branch v4.5.0 https://github.com/mudler/LocalAI.git
# already have the repo? check out this version:$ git checkout v4.5.0
└──▷ USE IT
Enable conversation compaction on a long-running realtime session to keep LLM context costs bounded while preserving a rolling summary of older turns.
›New depth-anything backend exposes POST /v1/depth returning depth map, stats, and camera extrinsics (3×4) / intrinsics (3×3); also reachable via a typed Depth gRPC RPC and GenerateImage/Predict — no Python at inference.
›New ced backend exposes POST /v1/audio/classification for 527-class AudioSet sound tagging (baby cry, glass breaking, alarms); opt-in pipeline.sound_detection emits conversation.item.sound_detection events in realtime, decoupled from VAD via turn_detection: none.
›New supertonic ONNX TTS backend supports voice, language, steps, speed, and silence request knobs across 10 voice styles (F1–F5 / M1–M5) — no Python, no espeak-ng/G2P.
›New privacy-filter.cpp backend adds NER token-classification PII filtering (encoder scans whole conversations as a single document) alongside a restricted-regex secret detector; detections labelled by source (ner vs pattern) with confidence and analyze/redact API; request filtering extended to completions, embeddings, edits, and Ollama.
›Realtime voice sessions gain pipeline.voice_recognition config keys including enforce: false to surface speaker identity to the client via a new conversation.item.speaker event and feed it to the LLM via per-message name field.
+10 moreshow less
›Conversation compaction via pipeline.compaction config block — keys enabled, trigger_items, summary_model, and max_summary_tokens — rolls aged-out turns into a rolling summary instead of dropping them; pipeline.max_history_items controls the live verbatim window.
›Adds OpenAI-parity realtime history events conversation.item.delete, conversation.item.truncate, and input_audio_buffer.clear.
›Prefix caching (n_cache_reuse) now defaults to 256 (was 0), so system prompts, RAG context, and multi-turn chat prefixes are reused across requests without configuration.
›Default n_batch/n_ubatch raised to 2048 on NVIDIA Blackwell consumer GPUs (sm_120/121, including GB10 / DGX Spark) when batch: is unset.
›New alias: field in model config transparently routes all traffic for one model name to another configured model, swappable live without client reconfiguration; alias→alias chains are rejected at load.
›Word-level ASR timestamps supported for NeMo and CrispASR, plumbed through the gRPC transcription path.
›Eight depth-anything GGUFs ship at mudler/depth-anything.cpp-gguf (base/small/large/giant + mono-large, q4_k/q8_0/f16/f32); eight ced GGUFs at mudler/ced-gguf (6 MB–86 MB); Depth Anything V2 gallery models also added.
›Distributed staging detached from the request context so large models finish cold-loading; staging progress broadcast across replicas; resumable downloader added.
llama.cpp b9768 adds conversion and inference support for Granite Speech Plus with multi-layer audio concatenation.
└──▷ GET THIS VERSION
$ git clone --branch b9768 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b9768
›Adds conversion support for IBM Granite Speech Plus models, extending the granite_speech architecture to handle multi-layer (feature_layers) audio concatenation for richer speech understanding.
LanceDB python-v0.34.0-beta.2 adds blob v2 schema declaration, write path, and blob read/materialization APIs in Rust.
└──▷ GET THIS VERSION
$ git clone --branch python-v0.34.0-beta.2 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:$ git checkout python-v0.34.0-beta.2
›Adds blob v2 schema declaration and write path in the Rust backend.
›Adds blob read and materialization APIs in the Rust backend.
LanceDB v0.31.0-beta.2 adds blob v2 schema declaration, write path, and blob read/materialization APIs in Rust.
└──▷ GET THIS VERSION
$ git clone --branch v0.31.0-beta.2 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:$ git checkout v0.31.0-beta.2
›Adds blob v2 schema declaration and write path in the Rust API.
›Adds blob read and materialization APIs in the Rust API.
THE AI TOOLCHAIN