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 v0.7.0 adds Experts marketplace, per-expert threads, voice onboarding, Better Auth, and expanded LLM model catalog including Kimi K3.
└──▷ GET THIS VERSION
$ git clone --branch autogpt-platform-beta-v0.7.0 https://github.com/Significant-Gravitas/AutoGPT.git
# already have the repo? check out this version:$ git checkout autogpt-platform-beta-v0.7.0
›Replaces Supabase Auth with Better Auth for platform identity management.
›Introduces a single-source LLM model catalog with cutover support and adds Kimi K3 as a supported model.
›Adds expert-scoped sessions and identity context in the Copilot backend.
›Adds an Experts marketplace section, team page, and per-expert conversation threads.
›Adds a voice brain-dump onboarding step for new users.
+2 moreshow less
›Adds a compact wallet popover for the new platform layout.
›Replaces Phosphor icons with Hugeicons stroke-rounded across the UI.
Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.
CrewAI 1.15.12 adds URLReadTool, app metadata on platform action tools, and unifies scaffolding under crewai create
└──▷ GET THIS VERSION
$ git clone --branch 1.15.12 https://github.com/crewAIInc/crewAI.git
# already have the repo? check out this version:$ git checkout 1.15.12
└──▷ TRY IT
Scaffold a new CrewAI resource using the unified create command instead of legacy subcommands.
$ crewai create <resource>
›Unifies scaffolding commands under the crewai create <resource> subcommand.
langchain-anthropic 1.5.4 adds a user_profile_id convenience attribute to the Anthropic integration.
└──▷ GET THIS VERSION
$ git clone --branch langchain-anthropic==1.5.4 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout langchain-anthropic==1.5.4
›Adds user_profile_id convenience attribute to the Anthropic chat model class for passing user profile identifiers to the Anthropic API.
Autonomous coding agent as an SDK, IDE extension, or CLI assistant.
Cline v4.1.5 adds a dedicated expired-free-model notice and unifies reasoning settings across AI SDK providers including Ollama.
└──▷ GET THIS VERSION
$ git clone --branch v4.1.5 https://github.com/cline/cline.git
# already have the repo? check out this version:$ git checkout v4.1.5
›Adds a dedicated in-UI notice with a model-picker button when a request hits a retired free model (e.g. cline-free/<slug> returning 404), replacing the previous generic error with no recovery path.
›Maps reasoning settings (effort levels and enable/disable toggles) onto a shared resolution path across all AI SDK providers, including Ollama, so behaviour is consistent without per-provider overrides.
Cline v4.1.4 adds Chutes provider support, shows skills in slash command menu, and hard-blocks file-editing commands in plan mode.
└──▷ GET THIS VERSION
$ git clone --branch v4.1.4 https://github.com/cline/cline.git
# already have the repo? check out this version:$ git checkout v4.1.4
›Hard-blocks file-editing shell commands in plan mode — file manipulation, in-place editors, redirection to files, mutating git subcommands, and package installs are refused; read-only investigation still works.
›Adds Chutes as a supported provider.
›Shows skills alongside workflows in the slash command menu, and disambiguates commands that share a name instead of letting one shadow the other.
›Removes model-initiated plan-to-act switching — switching out of plan mode is now user-driven, not triggered by the model mid-turn.
Cline desktop v0.0.9 adds universal macOS binary, session-source filtering, context-window token ring, and Skills in slash commands.
└──▷ GET THIS VERSION
$ git clone --branch desktop-v0.0.9 https://github.com/cline/cline.git
# already have the repo? check out this version:$ git checkout desktop-v0.0.9
›Session history can now be filtered by source — Desktop, CLI, extension, or scheduled — via a new filter control in the sidebar.
›The composer displays a token usage ring showing the active model's context window utilization, cumulative cost, and a color change as the limit approaches.
›Skills now appear in the slash command menu alongside workflows, with name conflicts disambiguated instead of shadowed.
›Plan mode now hard-blocks file-editing shell commands rather than relying on prompting alone; read-only investigation remains available.
›Context exhaustion is now recovered from automatically — the run compacts and retries once instead of failing with a provider error.
+5 moreshow less
›Ships as a single universal macOS binary running natively on both Apple Silicon and Intel, with automatic migration from per-architecture installs.
›Adds Infomaniak and SCX.ai to the bundled provider and model catalog.
›Empty model responses are now retried on every provider — including OpenRouter, Cline, and OpenAI-compatible endpoints — not just Ollama.
›Claude 4.6+ and 5.x models are now accepted on providers that previously rejected them with 'thinking.type.enabled is not supported'.
›Application errors are now captured and reported in diagnostics; telemetry configuration is baked into the sidecar at build time.
Cline CLI v3.0.50 adds user-selectable TUI color themes, Plan mode file-edit blocking, and context-exhaustion auto-recovery.
└──▷ GET THIS VERSION
$ git clone --branch cli-v3.0.50 https://github.com/cline/cline.git
# already have the repo? check out this version:$ git checkout cli-v3.0.50
└──▷ TRY IT
Set a persistent TUI theme at startup without entering settings, useful in scripts or dotfiles that launch Cline in a specific terminal palette.
$ CLINE_THEME='Tokyo Night' cline
Switch themes interactively mid-session and preview each one live before committing.
$ /theme
›Adds /theme command, command palette, and Theme row in /settings to select a live-previewing TUI color theme; CLINE_THEME environment variable overrides the persisted choice at startup.
›Supports 11 built-in themes — Auto (terminal-adaptive default), Cline Dark, Cline Light, Tokyo Night, Gruvbox Dark, Nord, Dracula, Catppuccin Mocha, One Dark, Solarized Dark, and Solarized Light — painting background, foreground, accents, syntax highlighting, and diff colors.
›Plan mode now hard-blocks file-manipulation shell commands, in-place editors (sed -i, perl -i), file-redirection, mutating git subcommands, package installs, and nested command strings (sh -c, eval, sudo) on Linux, Windows, and PowerShell; run_commands remains available for read-only investigation.
›Context exhaustion now triggers automatic force-compact and a single retry instead of surfacing a raw provider error; unrecoverable cases report the specific reason.
›Empty model responses are now retried on all providers — OpenRouter, Cline, and OpenAI-compatible endpoints — rather than failing the task with 'Model returned empty response'.
+5 moreshow less
›Reasoning models on OpenAI-compatible endpoints now receive max_completion_tokens instead of the rejected max_tokens.
›Connector tools are enabled by default; doctor and connect skip connectors that are already starting, and the hub supervises connector processes to prevent race collisions.
›Telegram slash commands such as /clear now reach the connector command host instead of being silently dropped.
›Two new model catalog providers — Infomaniak and SCX.ai — are now available.
›Upgrades the model layer to AI SDK 7 and switches Ollama to the native AI SDK provider.
›Adds a mode field to StartSessionInput (values: user, automation, subagent, team) alongside source to record how each session began.
›Plan mode now hard-blocks file-manipulation shell commands at the tool layer — rejecting in-place editors (sed -i, perl -i), file redirections, mutating git subcommands, package installs, and nested command strings (sh -c, eval, sudo, xargs) with a tool error on Linux, Windows, and PowerShell; run_commands remains available for read-only investigation.
›Context-window overflow errors are now detected and recovered from: the runtime force-compacts with a deterministic strategy requiring no extra LLM call, retries once, and fails with an actionable message for terminal cases.
›Empty-turn retries now apply to all providers, not just Ollama — OpenRouter, Cline, and OpenAI-compatible endpoints previously failed the task outright on empty responses.
›Root-session persistence is now lazy: allocating a runtime reserves the session ID in memory without writing a database row, so abandoned sessions no longer leave empty history entries.
+6 moreshow less
›Bedrock foundation models are now routed through geo inference profiles.
›Reasoning models on OpenAI-compatible endpoints now receive max_completion_tokens instead of max_tokens.
›Adaptive-era Claude models (4.6+ and 5.x) are no longer sent the manual thinking wire shape — the model catalog now carries reasoning metadata, with fallback inference for unlisted or user-typed adaptive IDs.
›Requests to models without image support now substitute image content instead of failing.
›Upgraded the model layer to AI SDK 7 with a native Ollama provider, fixing wire-contract issues for empty think settings, mid-stream errors, and attachment-only turns.
›sdk.error telemetry is no longer emitted twice per provider failure, and repeated failures from unattended retry loops are now rate-limited.
Lightweight coding agent that runs in your terminal
Adds disabled-by-default image resize notices, local auth requirement enforcement, and symlinked skill discovery path tracking.
└──▷ GET THIS VERSION
$ git clone --branch rust-v0.147.0-alpha.12 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.147.0-alpha.12
›Adds the disabled-by-default image_resize_notice feature: when enabled, appends a developer message after resized images reporting each image's original and prepared dimensions, for both user messages and tool outputs.
›Adds local requirements.toml allowlists for login methods and ChatGPT workspaces, enforced before stored or environment-provided credentials are used — including during bootstrap before cloud requirements are fetched.
›Makes Windows PathUri equality and hashing ASCII-case-insensitive for drive and UNC paths while preserving case-sensitive POSIX behavior, including starts_with and relative_path_from comparisons.
Codex CLI alpha.6.5 adds Amazon Bedrock cached web search, tightened Agent Plugin runtime boundaries, and fuller MCP namespace descriptions.
└──▷ GET THIS VERSION
$ git clone --branch rust-v0.147.0-alpha.6.5 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.147.0-alpha.6.5
›Raises the MCP namespace tool-spec description limit from 1,000 bytes to 512 KiB, truncating only at a UTF-8 character boundary, so complete server instructions are preserved in tool-search metadata.
›Enables cached web search for Amazon Bedrock: advertises hosted web search while marking external live/indexed access unsupported, and resolves unsupported modes to cached search or disables the tool when managed requirements prohibit it.
›Skips symbolic links and other non-file, non-directory entries when installing a plugin into the store, allowing installations that include symlinked skill files or executables to succeed.
Codex CLI alpha.11 adds per-session code-mode execution limits and switches API-key Guardian reviews to gpt-5.6-luna.
└──▷ GET THIS VERSION
$ git clone --branch rust-v0.147.0-alpha.11 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.147.0-alpha.11
›Adds create_session_with_limits and session-scoped cell execution limits via CodeModeSessionCellExecutionLimits, clamping execute and wait yield times to max_yield_time_ms without terminating the running cell; limits are negotiated with remote code-mode hosts and included in session/open only when non-default.
›Switches Guardian approval reviews for API-key authenticated sessions to the gpt-5.6-luna model (retaining codex-auto-review for ChatGPT authentication), routing API-key reviews through Responses Lite with fallback to the bundled Guardian policy.
›Caches McpBinding per published MCP runtime once the tool catalog revision is stable, reusing the binding across sampling steps and rebuilding only after a runtime publish or catalog change.
›Bounds interactive TUI telemetry shutdown to a 500 ms budget via OtelProvider::shutdown_with_timeout, preventing a stalled OTLP exporter from blocking exit indefinitely.
Codex CLI alpha adds concurrent exec-server dispatch, tool registry collision policy, paginated TUI history, and leaf model support for multi-agent v2.
└──▷ GET THIS VERSION
$ git clone --branch rust-v0.147.0-alpha.10 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.147.0-alpha.10
└──▷ TRY IT
Run the exec-server with up to 4 concurrent requests per connection to increase throughput in high-load environments.
Start Codex with process-scoped PSP routing enabled for first-party ChatGPT requests.
$ codex --psp
›Adds --concurrent-requests flag (alias --concurrent-requests COUNT, default 1) to the exec-server subcommand for opt-in concurrent request dispatch across connections.
›Adds --psp global flag to enable process-scoped PSP routing for first-party ChatGPT requests.
›Adds tool registry collision policy configuration with strict tool name collision errors enforced under a new canonical functions namespace.
›Adds paginated transcript history loading in the TUI, enabling navigation of long session histories.
›Enables cached web search for Amazon Bedrock.
+11 moreshow less
›Enables remote compaction for Amazon Bedrock.
›Prompts before trusting local project directories, and rejects implicitly discovered bare Git repositories.
›Adds durable per-thread user submission queues.
›Links Codex attribution in pull request bodies.
›Allows disabling the built-in image viewer.
›Skips symlinks when installing plugins.
›Limits RMCP logs persisted to SQLite.
›Negotiates MCP extensions per app-server session.
Qwen Code desktop v0.1.0 adds session-scoped MCP, git branch/PR UI, streaming charts, channel management, and voice hold mode.
└──▷ GET THIS VERSION
$ git clone --branch desktop-v0.1.0 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout desktop-v0.1.0
›Adds session-scoped runtime MCP via acp so each session can carry its own Model Context Protocol server without affecting others.
›Adds a Channel management page and Channel configuration flows in the web shell UI for creating and editing channel integrations.
›Adds Channel pairing request management in the web shell, letting users accept or reject incoming channel pairings.
›Adds git branch picker, commit dialog, and create PR flow directly in the web shell composer.
›Adds a native workspace folder picker in the web shell for selecting working directories.
+18 moreshow less
›Adds streaming chart rendering via markdown-chart in the web shell for live data visualization.
›Adds submitted-prompt auto recall in external-context, surfacing previously submitted prompts automatically.
›Adds submitted-prompt provenance tracking in hooks, recording where each prompt originated.
›Gates session writer lease behind an opt-in flag, giving operators control over write-lock acquisition.
›Adds voice hold mode support in the web shell, scoped to the composer workspace.
›Adds BTW side-question suggestions in the web shell composer footer.
›Adds a composer footer renderer in the web shell for extensible footer content.
›Adds monitor task details view in the web shell for inspecting running task status.
›Exposes loop tools in daemon sessions via the channels layer.
›Dispatches GitHub notifications by reason in the channels integration, enabling reason-based routing.
›Adds ARMS session user ID field to core session tracking.
›Adds full-resolution image zoom tool in core with bounded image reads for reliability.
›Persists and replays Goal v3 state in core, and connects Goal v3 to the turn engine.
›Adds early Anthropic cache breakpoint on the stable system prefix for reduced latency on repeated calls.
›Adds autofix deferred suggestions after five change rounds, reducing noise during iterative edits.
›Adds autofix retry on deterministic rejection, improving success rate on first-pass linting failures.
›Adds revert-pattern high-risk path detection in triage.
›Adds agent.compaction_model setting to specify the model used for context compaction.
›Adds agent_ui_font_family and agent_buffer_font_family settings for configuring the Agent Panel's UI and buffer fonts.
›Adds a 'Skip Hooks' toggle (also available as git: skip hooks action) in the Git Panel Commit button menu and commit modal, bypassing pre-commit and commit-msg hooks for the next commit.
›Adds the _ keybinding in Helix mode to trim whitespace from selections.
›Adds sandboxing for the Agent's terminal and fetch tools.
+7 moreshow less
›Adds a reasoning effort selector for Anthropic-compatible providers when the selected model supports adaptive thinking.
›Adds support for undoing and redoing file operations in the Project Panel.
›Improves OpenAI Responses API requests by sending system messages through the instructions field.
›Improves Markdown preview links so links to Markdown files open a preview scrolled to the linked heading or position, with alt-click opening the raw source instead.
›Improves call diagnostics with per-participant audio health history and exportable reports.
›Improves proxy support by respecting NO_PROXY and percent-encoded proxy credentials.
›Improves support for language servers that dynamically register inlay hints, code lenses, or document symbols more than once.
LocalAI 4.8.0 adds vllm.cpp alpha backend, 3D generation, audio.cpp, VRAM budgets, two new TTS engines, and more
└──▷ GET THIS VERSION
$ git clone --branch v4.8.0 https://github.com/mudler/LocalAI.git
# already have the repo? check out this version:$ git checkout v4.8.0
└──▷ TRY IT
Cap GPU memory to 12 GB on a shared node so LocalAI does not crowd out other workloads.
$ local-ai --vram-budget 12GB
›Adds POST /v1/3d/generations endpoint, Generate3D RPC, and FLAG_3D capability for 3D generation as a new modality, backed by the trellis2cpp image-to-3D backend and a UI page with a native GLB viewer and print remeshing.
›Adds LOCALAI_VRAM_BUDGET environment variable and --vram-budget CLI flag (accepts 80% or 12GB) to cap GPU memory use per node, with admin API endpoints PUT /api/nodes/:id/vram-budget and DELETE /api/nodes/:id/vram-budget, and an set_node_vram_budget MCP tool.
›Adds --disable-http-compression and --http-compression-min-length flags; /api/traces and /api/backend-traces now accept limit, offset, and full query parameters with per-record access via GET /api/traces/{id}.
›Adds GET /api/models?has_variants=true to filter gallery models with variant ladders, with variant selection via variant on POST /models/apply, local-ai models install --variant, the install_model MCP tool, or the UI split-button.
›Ships vllm-cpp as an alpha backend: a C++20 engine (no Python, PyTorch, or ggml at inference) implementing vLLM V1 architecture (paged KV cache, continuous batching, prefix caching), loading safetensors and GGUF, with in-engine structured output (JSON schema, regex, GBNF), running on CPU, CUDA 12/13, Metal, and Vulkan.
+10 moreshow less
›Ships audio-cpp backend serving six endpoints — /v1/audio/speech, /v1/audio/transcriptions, /v1/audio/vad, /v1/audio/diarize, /audio/transform, and /v1/sound-generation — with family auto-detected from the GGUF audiocpp.model_spec.family metadata key.
›Ships magpie-tts-cpp backend for NVIDIA Magpie TTS Multilingual (5 voices, 9+ languages, 22.05 kHz mono, single self-contained GGUF).
›Ships moss-tts-cpp backend for MOSS-TTS-Local v1.5 (48 kHz stereo, optional reference-audio voice cloning, no Python at inference).
›Ships bonsai backend for Q1_0 and Q2_0 quant formats, covering Bonsai 8B/27B and Ternary-Bonsai 8B/27B models via eight gallery entries.
›Ships valkey-store backend adding Valkey Search as a vector store option.
›Adds a new admin /app/activity page tracking install, failure, and cancellation events with a 50-entry bounded history, per-node in-progress breakdown, and a 'needs attention' lane with Cancel and Retry actions.
›Gallery entries can now declare variants: so LocalAI automatically selects the largest build that fits the host's VRAM and RAM at install time.
›Adds Hugging Face artifact materialization with immutable snapshot resolution, authenticated downloads with progress reporting, and staged artifact binding for remote workers.
›Adds systemd socket activation support.
›Inline GRPO reward code in POST /api/fine-tuning/jobs is now blocked by default; operators must set LOCALAI_TRL_ALLOW_INLINE_REWARD=true to re-enable it.
llama.cpp b10282 adds speculative-decoding counters to the /metrics endpoint, aligned with vLLM naming.
└──▷ GET THIS VERSION
$ git clone --branch b10282 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b10282
└──▷ TRY IT
Scrape speculative-decoding acceptance stats from a running llama.cpp server to feed into a Prometheus/Grafana monitoring stack.
$ curl http://localhost:8080/metrics | grep spec
›Adds spec-decode counters to the /metrics endpoint on the llama.cpp server, with parameter names aligned to vLLM conventions for drop-in compatibility.
Phoenix v19.18.0 adds a REST endpoint for prompt metadata updates and an expression filter DSL for sessions.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v19.18.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v19.18.0
›Adds a REST endpoint for updating prompt metadata.
›Adds an expression filter DSL for filtering sessions.
Arize Phoenix ships arize-phoenix-sqlean v0.1.0, vendoring nalgeon/sqlean.py as a first-party library.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-sqlean-v0.1.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-sqlean-v0.1.0
›Introduces the arize-phoenix-sqlean package, a vendored distribution of nalgeon/sqlean.py under the Arize Phoenix project.
Composio CLI 0.3.2 makes automatic shell setup the default during installation.
└──▷ 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]
›Automatic shell setup is now the installer default, removing the need for manual shell configuration after install.
Qdrant v1.19.0 adds TurboQuant 4-bit primary storage, prefix match filters, per-query IDF, slice filtering, a global quota API, and routing tokens for read affinity.
└──▷ GET THIS VERSION
$ git clone --branch v1.19.0 https://github.com/qdrant/qdrant.git
# already have the repo? check out this version:$ git checkout v1.19.0
└──▷ USE IT
Filter results to only points whose 'category' keyword field starts with a given prefix, useful for hierarchical tag lookups.
›Adds "memory": "cold" / "cached" / "pinned" configuration per individual collection component to enable fine-grained control over memory behavior and performance.
›Adds "match": {"prefix": "..."} in filter to match keywords by prefix (must be enabled in the keyword index).
›Introduces a Global Quota API for managing resource quotas across the cluster.
›Adds a routing token for deterministic read routes, enabling read-affinity consistency guarantees.
›Introduces a Slice filtering condition supporting sliced scroll and deterministic sampling.
+8 moreshow less
›Adds per-query IDF corpus for sparse vector search, enabling per-tenant IDF statistics in full-text search.
›Adds TurboQuant 4-bit as a datatype for primary vector storage, storing only 4-bit quantized vectors and eliminating the need to retain full-precision originals.
›Web UI gains a management interface for payload indexes and their configuration.
›Web UI gains a management interface for global resource quotas.
›Web UI adds a multi-delete feature for collections.
›Web UI adds display and editing for collection metadata.
›Web UI adds a display for resharding progress.
›Significantly more performant vector visualization and dimensionality reduction in the Web UI using a WASM-based UMAP implementation and WebGL rendering.
└──▷ BREAKING ON UPGRADE
!The default update queue length is reduced from 1,000,000 to 200; deployments relying on the previous high-water mark will need to reconfigure.
!max_resident_memory_percent in strict mode is deprecated in favor of the new global quota API; existing configurations using this field should migrate.
THE AI TOOLCHAIN