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 v2.9.0 adds StudioRunnerTools for identity-aware dispatch and hardens MCP tool security and cache isolation
└──▷ GET THIS VERSION
$ git clone --branch v2.9.0 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:$ git checkout v2.9.0
└──▷ USE IT
Mount StudioRunnerTools on a router agent so it can discover and invoke Studio-built agents and teams on behalf of the authenticated user, without exposing create/edit/delete operations.
›Adds agno.tools.studio_runner.StudioRunnerTools, an identity-aware dispatch toolkit that lets any component (team lead, router) discover and run Studio-built agents, teams, and workflows — without exposing Studio's create/edit/delete surface — threading the caller's user_id into sub-runs for correct per-user state.
›Adds a name filter parameter to list_components for narrowing component discovery by name.
›MCP tool entrypoints no longer accept a call-time tool_name override; the executed tool name is now closed over from tool.name, closing a bypass of allow-lists, requires_confirmation, HITL approval, and logging gates.
›Rehydration of persisted components with unresolvable references now raises ComponentRehydrationError (an AgnoError, status_code=422) on strict paths — AgentOS lookups and all dispatch paths (POST /runs, continue, MCP run tools, StudioRunner) default to strict=True and return a 422 naming the unresolvable piece instead of silently running a degraded component; public from_dict/load default to strict=False.
└──▷ BREAKING ON UPGRADE
!MCP tool call-time tool_name overrides are now ignored and forwarded as ordinary arguments instead of selecting the tool to execute; any integration that relied on passing tool_name at call time to route to a different tool will no longer work as before.
!Tool result cache keys now include user_id and session_id, so all prior cache entries composed without those fields will not produce hits after upgrading.
!AgentOS lookups and all dispatch paths (POST /runs, continue, MCP run tools, StudioRunner) now default strict=True for rehydration and return a 422 ComponentRehydrationError for unresolvable references instead of silently degrading and running the component.
AutoGPT Platform beta v0.7.1 adds Tavily search/crawl blocks, Claude Sonnet 5, expert scheduling, and single-container distribution.
└──▷ GET THIS VERSION
$ git clone --branch autogpt-platform-beta-v0.7.1 https://github.com/Significant-Gravitas/AutoGPT.git
# already have the repo? check out this version:$ git checkout autogpt-platform-beta-v0.7.1
›Adds a new home dashboard aggregation endpoint to power the briefing-first home view.
›Adds Tavily provider blocks for search, extract, crawl, and map operations.
›Adds Claude Sonnet 5 support with Claude 5 tokenizer awareness.
›Adds expert scheduling with attribution, triggers, thread posts, and a credit guardrail.
›Adds expert scheduling UI including an expert page, chat schedules drawer, and team card polish.
+10 moreshow less
›Adds collapsible expert chat groups in the sidebar with per-group 'load more' pagination.
›Adds editable expert Soul documents.
›Adds a morning briefing and needs-attention unification with a briefing-first home layout.
›Adds status counts and duration totals to the execution cost summary.
›Adds single-container distribution with published single-container images.
›Adds a ChatGPT/Codex subscription transport preview.
›Adds chat-platform account linking.
›Adds first-visit tab intro cards for Agents, Marketplace, and Build sections.
›Adds a brain dump speech-to-text quality gate and evaluation corpus.
›Retires the brain dump greeting once a user has an established session.
Hermes v2026.8.13 adds setup_mcp consent cards, profiles.list/profiles.create ws RPCs, per-route webhook toolset overrides, and an image.generate ws RPC for plugins.
└──▷ GET THIS VERSION
$ git clone --branch v2026.8.13 https://github.com/NousResearch/hermes-agent.git
# already have the repo? check out this version:$ git checkout v2026.8.13
└──▷ USE IT
Enumerate all profiles from a desktop plugin to build a roster UI, including a last-session preview per profile.
javascript
// Inside a desktop plugin, via host.request ws JSON-RPC
host.request('profiles.list', { include_sessions: true })
›Adds profiles.list and profiles.create ws JSON-RPC methods (in tui_gateway/methods_profiles.py), letting desktop plugins enumerate profiles (name, path, model/provider, description, skill_count, optional last_session preview via include_sessions) and create new profiles without going through the dashboard REST router.
›Adds image.generate ws RPC for plugin surfaces, enabling plugins to trigger image generation through the agent backend.
›Adds per-route toolset overrides for webhook agent runs, allowing different tool sets to be configured per webhook route.
›Adds setup_mcp tool with an inline MCP consent card rendered as an interactive clarify-style blocking bridge in the desktop transcript, letting users approve MCP server connections without leaving the conversation.
›Suggests MCP servers from the composer draft as brand pills in the desktop UI.
+6 moreshow less
›Adds cron and blueprint recipe shortcuts to the sidebar nav rail in the desktop app.
›Surfaces model drift impact in the desktop cron view.
›Adds skill-match, connection-repair, and recurrence-to-cron suggestion providers in the desktop composer.
›Labels the agent's recommended choice on every clarify surface, rendered in tertiary text in the desktop UI.
›Quietly suppresses composer suggestions the user has repeatedly ignored.
›Adds marquee animation for clipped inline row titles on hover in the desktop app.
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 v2.1.232 adds subagent forking by default, cross-session @-mentions, GitLab secret redaction, and deeper Remote Control resilience.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.232 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.232
└──▷ USE IT
Configure a managed gateway policy that references GitLab-hosted plugin marketplaces using the new friendlier alias allowedMarketplaces.
Control how cross-session messages from your other Claude sessions are handled — accept, hold, or refuse — via the new /config dialog.
📍Open /config in an interactive Claude Code session, then find 'Messages from your other sessions' and set it to accept, hold, or refuse.
›Adds additionalMarketplaces and allowedMarketplaces as accepted aliases for extraKnownMarketplaces and strictKnownMarketplaces in settings.
›Enterprise policy: a url-typed blockedMarketplaces entry for a bare repo URL now blocks that URL when the CLI classifies it as a git clone.
›Gateway: sandbox.bwrapPath, sandbox.socatPath, and sandbox.ripgrep now require managed-settings approval when a server override is present.
›Changes sandbox.ripgrep to be honored only from user, managed, and --settings settings; project settings can no longer override the sandbox ripgrep binary.
›Gateway: the desktop: overlay now accepts every released Desktop setting (validated at boot against Desktop's own schema); unknown or invalid keys fail boot.
+16 moreshow less
›Gateway: empty managed.policies[].match.groups / admin.admin_groups entries and malformed email_domain values now fail at boot instead of silently matching no one or granting admin access.
›Adds /config rows for 'Dialog expiry' and 'Messages from your other sessions' (cross-session inbound accept/hold/refuse controls).
›Adds secret redaction for GitLab token families glrt-, gloas-, glptt-, glagent-, glimt-, glsoat-, glcbt-, glft-, glffct-, and full redaction of routable glpat-/gldt- tokens; the glab CLI config store gets the same sandbox and credential-path protection as gh.
›Adds GitLab support to plugin marketplaces: bare gitlab.com repo URLs (including nested subgroups) now clone like github.com URLs.
›Subagent forking is now on by default: a subagent_type: "fork" subagent inherits the full conversation and prompt cache, and non-teammate agent spawns in interactive sessions now run in the background by default.
›Type @ in the prompt to mention another Claude session by name; Claude uses SendMessage to reach that session directly.
›SendMessage now delivers to a bare name that exactly matches one live session without requiring a ref confirmation first.
›Interactive sessions now keep unique names: starting or renaming a session to a name already in use gives it a name-word-word variant automatically.
›Fable 5 is offered as an advisor in /advisor again for organizations with Fable access, with usage-credits consent configured through /model fable.
›/plugin install plugin@marketplace now refreshes the marketplace first so newly published plugins install without a manual marketplace update.
›/code-review at high, xhigh, and max effort now runs in a background agent like the other effort levels.
›/feedback and /bug now open immediately when invoked while Claude is responding, without waiting for the turn to finish.
›Remote Control now keeps reconnecting for about 30 minutes after a network blip.
›Bash input redirections (< file) are now permission-checked like their argument spellings on all platforms.
›Hardened the auto-generated cross-session messaging socket directory on shared /tmp against pre-planted symlinks or another user's directory.
›Hardened the Linux filesystem sandbox against a protected-path bypass.
└──▷ BREAKING ON UPGRADE
!sandbox.ripgrep set in project settings is now ignored; only user, managed, and --settings sources are honored.
!Gateway: unknown or invalid keys in the desktop: overlay now fail boot instead of being silently accepted.
!Gateway: empty managed.policies[].match.groups / admin.admin_groups entries and malformed email_domain values (empty, containing @, whitespace, or commas) now fail at boot.
Autonomous coding agent as an SDK, IDE extension, or CLI assistant.
Cline CLI v3.0.54 improves Hub daemon upgrades and Claude Code provider reliability for agentic sessions.
└──▷ GET THIS VERSION
$ git clone --branch cli-v3.0.54 https://github.com/cline/cline.git
# already have the repo? check out this version:$ git checkout cli-v3.0.54
›Managed Hub daemons now upgrade directionally: when another Cline install ships a newer Hub build, the CLI attaches to the newer daemon and prompts you to update and restart, instead of the two installs repeatedly retiring each other's daemons; Yolo and sandbox sessions are not interrupted by that prompt.
›Upgrading the CLI now retires an already-running Hub daemon and respawns it on the new code, instead of the upgraded CLI continuing to talk to a daemon executing the previous release.
Cline SDK v0.0.74 adds directional Hub daemon upgrades, improved run.started event correlation, and accurate per-request token telemetry.
└──▷ GET THIS VERSION
$ git clone --branch sdk/sdk/v0.0.74 https://github.com/cline/cline.git
# already have the repo? check out this version:$ git checkout sdk/sdk/v0.0.74
›Token telemetry now reports disjoint per-request buckets — uncached input, cache reads, and cache writes — instead of re-counting the full cached conversation on every event, eliminating ~5x inflation on cache-heavy sessions.
›Managed Hub daemons now upgrade directionally: when a newer Hub build is detected, hosts attach to the newer daemon and prompt to update and restart rather than two installs repeatedly retiring each other's daemons.
›Involuntary Cline logouts caused by a rejected refresh token are now surfaced explicitly instead of credentials being cleared silently; transient network failures on startup no longer register as logouts.
›Per-token stream deltas are no longer mirrored into telemetry, removing ~97% of agent event volume that had no analytical value.
Cline desktop v0.0.12 gives every tool call its own transcript row, fixes Claude Code provider for agentic sessions, and adds Vertex AI Fable 5 support.
└──▷ GET THIS VERSION
$ git clone --branch desktop-v0.0.12 https://github.com/cline/cline.git
# already have the repo? check out this version:$ git checkout desktop-v0.0.12
›Adds Fable 5 and custom model IDs to the Vertex AI provider, and fixes the global-region picker so it no longer hides models from the live catalog.
›Compaction now respects the Max Output Tokens setting instead of a hardcoded 1024-token cap, preventing reasoning models from exhausting the budget on thinking and skipping the summary.
›Every tool call now gets its own row in the session transcript with its own icon, status indicator, and expandable detail — commands render as terminal output ($ bun test) and edits show inline per-hunk diffs.
›Running tool rows are highlighted in brand violet and settle to gray on completion; errors remain red.
›File diffs in both chat rows and the diff panel now render through a shared syntax-highlighted renderer that follows the app theme.
+2 moreshow less
›The thinking indicator stays visible during quiet stretches of a turn (such as while tool arguments are streaming), preventing turns from appearing frozen.
›Message actions (copy, fork, timestamp) are repositioned so they no longer crowd message text, and expanded panels render at full opacity.
$ git clone --branch rust-v0.148.0-alpha.12 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.148.0-alpha.12
›Adds RecoverTurnRequest and CodexThread::recover_turn_if_idle to resume an interrupted regular turn using its existing turn ID and updated thread settings, including Plan-mode recovery without injecting an empty user message.
›Routes blocked network requests through the shared approval pipeline so permission hooks, automatic review, and user review all use the common approval flow — including for background terminals started by an earlier turn.
›Materializes inline visualization viewer documents in a dedicated cache under CODEX_HOME, keyed by source and artifact thread IDs, so sandboxed sessions cannot overwrite viewer files before they are opened in a browser.
›Persists deny amendments for blocked network requests and keeps denied requests denied across turns.
›Records the final applied network decision in tool telemetry without exposing the destination or assigning an approval source.
+1 moreshow less
›Disables visualization links for full-disk-write sessions and rejects viewer cache paths containing symbolic links; reuses materialized viewer documents in memory when the source is unchanged.
Codex CLI alpha.11 adds thread usage displays, thread-credits and estimated-thread-cost terminal title items, and improved MCP tool-call approval routing.
└──▷ GET THIS VERSION
$ git clone --branch rust-v0.148.0-alpha.11 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.148.0-alpha.11
└──▷ USE IT
Surface per-thread credit consumption and estimated cost directly in your terminal title bar while running Codex.
yaml
thread-credits, estimated-thread-cost
›Adds thread-credits and estimated-thread-cost as configurable terminal title items, letting practitioners track per-thread quota consumption at a glance.
›Shows estimated thread usage in the /status command and TUI status surfaces.
›Adds dynamic HTTP header helpers for MCP servers.
›Integrates workload identity with Codex authentication.
›Supports gRPC code-mode hosts in the app server via --code-mode-host.
+8 moreshow less
›Enables executor skill packages to be passed directly to skills.read without a preceding skills.list lookup, including support for skill root aliases.
›Enforces non-interactive approval policy for Codex delegate sessions.
›Reconnects gRPC code-mode sessions automatically after host restarts.
›Adds per-thread usage queries to the backend client.
›Stamps conversation history items with creation times.
›Adds durable reverts for paginated threads.
›Adds manifest-defined metrics and bounded measurement analytics for trusted plugin scripts.
›Attaches connector ID, action name, and model to file creation requests for hosted app tool calls.
Qwen Code v0.21.11 adds Agent Plugins v1, the /coordinate multi-agent command, OpenTelemetry session lifecycle events, and a Channels sidebar view.
└──▷ GET THIS VERSION
$ git clone --branch v0.21.11 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout v0.21.11
└──▷ TRY IT
Spin up a multi-agent session where specialist read-only teammates contribute to a task and results are forwarded automatically to the leader agent.
$ /coordinate
Check memory health in the current session, including tool result sizes and warnings for any results over 30k characters that may degrade context.
$ /doctor memory
›Adds the /coordinate command to enable native multi-agent workflows with read-only teammates and automated result forwarding to the leader agent.
›Adds /doctor memory command reporting that now includes tool result retention stats, character counts, and warnings for results exceeding 30k characters.
›Adds support for Agent Plugins v1 to extend agent capabilities.
›Adds session.start and session.end OpenTelemetry lifecycle events to improve observability of session creation and shutdown.
›ACP sessions now use the unified Goal v3 runtime, supporting create, edit, pause, resume, and clear actions with improved turn scheduling.
+8 moreshow less
›Web Shell sidebar gains a Channels view to track integration sessions from DingTalk, Feishu, and WeCom alongside standard tasks.
›Web Shell now supports Qwen 3.8 reasoning controls, letting users toggle Thinking mode and select effort levels directly from the model chip.
›Terminal window titles now display status symbols (◐, ✳) to indicate task state in multiplexers where color cues are unavailable.
›Adds an Aliyun OSS release mirror for desktop builds.
›Project memory isolation now defaults to workspace scope for qwen serve runtimes, while standalone CLI behavior remains unchanged.
›Improved text selection with word-wise drag on double-click and line-wise extension on triple-click in the UI.
›Web Shell subagent rows now display persistent chevrons, improved hover states, and inline agent types and metrics.
›Background task notifications in Web Shell are now localizable and display structured metadata in chat-style bubbles.
An MCP that lets AI tools securely connect to your infrastructure, write IaaS code, debug issues, and assist during incidents - without risking production stability. Built for security teams to approve and infrastructure teams to experience like magic.
emisar v0.40.0 adds self-updating runners, paginated MCP responses, and 5 new catalog integrations
└──▷ GET THIS VERSION
$ git clone --branch v0.40.0 https://github.com/AndrewDryga/emisar.git
# already have the repo? check out this version:$ git checkout v0.40.0
└──▷ TRY IT
Keep a runner up to date in a verified, attestation-checked way from the host where it is installed.
$ emisar update
Capture machine-readable pack update output in CI even when post-update validation fails, so downstream automation can inspect the partial report.
›Adds emisar update subcommand for installer-managed runners to self-update, verifying release checksum and GitHub build attestation before executing stop, swap, restart, and rollback.
›emisar pack update --json now emits its partial report before returning a post-update validation error, preserving machine-readable output for automation pipelines.
›wait_for_run now pages large terminal runbook results as ordered 64 KiB chunks via an opaque continuation token when a result exceeds one MCP response, with small executions still completing in a single call.
›Expands the catalog to 100 packs and 1,682 actions, adding JFrog Artifactory, Databricks, Sentry, Symbolicator, and NTPsec, with expanded Cassandra and Cloudflare coverage.
›Runner, pack, action, approval, and audit discovery now enforces the member's current runner and pack scope across both the console and MCP.
Skip requant + convert + compile on subsequent runs by caching the compiled OpenVINO model to disk, reducing startup from ~15 s to ~6 s for an 8B model.
›Adds GGML_OPENVINO_RELEASE_WEIGHTS env var to madvise(MADV_DONTNEED) host weight buffers after GPU compile, dropping steady-state RSS from ~1555 MB to ~710 MB for a 1B Q4_K_M model on Arc iGPU.
›Adds GGML_OPENVINO_REDUCE_COMPILE_MEM env var to enable streaming weight requantization and cross-compile weight-node caching, cutting compile-time peak RSS by ~1 GB (1B model) to ~2 GB (8B model).
›Adds GGML_OPENVINO_MODEL_CACHE_DIR env var to enable a frontend model cache that imports a previously compiled CompiledModel, reducing compile phase from 15.3 s to 6.3 s (~2.4x) on an 8B Q4_K_M GPU run.
›Adds cache-key coverage for GGML_OPENVINO_DISABLE_KV_SLICE and GGML_OPENVINO_MANUAL_GQA_ATTN to prevent the frontend model cache from reusing blobs compiled under different attention-graph configurations.
›Enables Qwen3.5 (including dense multi-sequence and MoE variants) on the OpenVINO backend.
+7 moreshow less
›Enables OpenVINO fallback to CPU backend when GPU ops are unsupported.
›Adds GGML_OPENVINO_RELEASE_WEIGHTS hard-abort (GGML_ABORT) on cache-miss recompile or second model load to prevent silent wrong-model or zeroed-weight inference.
›Optimizes MoE expert aggregation in the OpenVINO backend using ReduceSum.
›Enables GGML_UNARY_OP_SIGMOID, SQR, SQRT, FILL, and GET_ROWS with multi-dimensional support in the OpenVINO backend.
›Adds gpt-oss MoE and mxfp4 support to the OpenVINO backend.
›Enables zero-size copy for view ops in the OpenVINO backend.
›Adds per-node runtime offsets (rs_slot_begin_*, rs_src_begin_*) to the OpenVINO backend for correct recurrent state rollback CPY writeback across any kv head, sequence count, and snapshot slot.
llama.cpp b10418 adds SYCL host pinned memory support via ggml_backend_sycl_host_buffer_type_get_max_size to improve Host-to-Device memory access.
└──▷ GET THIS VERSION
$ git clone --branch b10418 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b10418
›Adds ggml_backend_sycl_host_buffer_type_get_max_size to support host pinned memory on SYCL backends, improving Host-to-Device memory access performance.
llama.cpp b10414 adds TQ2_0 ternary quantization support to the Metal backend for Apple Silicon.
└──▷ GET THIS VERSION
$ git clone --branch b10414 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b10414
›Adds GGML_TYPE_TQ2_0 (ternary, 2 bits per element) quantization support to the Metal backend, enabling GPU-accelerated inference with TQ2_0 models on Apple Silicon.
llama.cpp b10413 auto-detects speculative decoding type from local draft GGUF metadata, fixing silent no-op when using -md without --spec-type.
└──▷ GET THIS VERSION
$ git clone --branch b10413 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b10413
›Adds common_speculative_types_from_gguf() in speculative.cpp/.h to auto-detect --spec-type from a local draft model's general.architecture GGUF header when -md is used without --spec-type, mapping dflash + markov_w1.weight tensor to draft-dspark and dflash alone to draft-dflash.
›Emits a LOG_INF message when spec-type auto-detection fires, giving users visibility into why speculative decoding was enabled.
llama.cpp b10408 adds SYCL ESIMD kernels for Q3_K, Q4_K, and Q6_K quantized inference on Intel GPUs
└──▷ GET THIS VERSION
$ git clone --branch b10408 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b10408
›Adds DMMV ESIMD kernels for Q3_K, Q4_K, and Q6_K quantization formats under the SYCL backend, enabled at build time with -DGGML_SYCL_ESIMD=ON and used by default at runtime when available on the device.
Phoenix agents now emit machine-readable approval decisions on gated PXI tools.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v20.2.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v20.2.0
›Agents emit a machine-readable approval decision on gated PXI tools, enabling automated handling of human-in-the-loop approval gates.
THE AI TOOLCHAIN