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 v2.5.15 adds Team skills, nested workflows, post-execution HITL output review, and new SessionSummaryManager controls.
└──▷ GET THIS VERSION
$ git clone --branch v2.5.15 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:$ git checkout v2.5.15
└──▷ TRY IT
Enable full tracebacks in error logs during development to diagnose agent failures without changing code.
$ AGNO_LOG_TRACEBACKS=true python my_agent.py
Limit how much history is fed into session summaries to keep token costs predictable for long-running sessions.
python
from agno.memory import SessionSummaryManager
summary_manager = SessionSummaryManager(
last_n_runs=10,
conversation_limit=4000
)
›Adds requires_output_review on Step, Router, and Loop to pause a workflow after a step runs and allow human review, approval, rejection with feedback, retry, or output editing before execution continues.
›Consolidates HITL parameters into a HumanReview config class — pass human_review=HumanReview(...) on Step, Loop, and Router instead of flat params; fully backward compatible.
›Adds last_n_runs and conversation_limit parameters to SessionSummaryManager to control how much conversation history is included when generating session summaries.
›Adds AGNO_LOG_TRACEBACKS environment variable (opt-in, off by default) to enable full tracebacks in log_error and log_warning.
›Adds skills support to Team, enabling teams to use shared skill sets.
+1 moreshow less
›Supports nested workflows — a Workflow can now be used as a step inside another Workflow.
OpenClaw v2026.4.9 adds REM backfill for memory/dreaming, diary timeline UI, character-vibes QA reports, and provider auth aliasing.
└──▷ GET THIS VERSION
$ git clone --branch v2026.4.9 https://github.com/openclaw/openclaw.git
# already have the repo? check out this version:$ git checkout v2026.4.9
└──▷ TRY IT
Pin an iOS release to a specific CalVer gateway version for a release train.
$ pnpm ios:version:pin ----from-gateway
›Adds grounded REM backfill lane via rem-harness --path so old daily notes can replay into Dreams and durable memory without a second memory stack.
›Adds structured diary view in Control UI with timeline navigation, backfill/reset controls, traceable dreaming summaries, and a grounded Scene lane with promotion hints.
›Adds character-vibes evaluation reports in QA/lab with model selection and parallel runs for faster candidate behavior comparison.
›Adds providerAuthAliases declaration in provider manifests so variants can share env vars, auth profiles, config-backed auth, and API-key onboarding choices.
›Adds pnpm ios:version:pin -- --from-gateway workflow for iOS release train versioning pinned to explicit CalVer.
+1 moreshow less
›Adds openclaw/plugin-sdk/command-status subpath to the Plugin SDK, splitting command status builders out of auth-only imports.
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 v2.1.98 adds Vertex AI setup wizard, Perforce mode, subprocess sandboxing, and a new Monitor tool for background scripts.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.98 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.98
└──▷ TRY IT
Protect read-only Perforce-managed files so writes fail loudly with a p4 edit hint instead of silently overwriting them.
$ CLAUDE_CODE_PERFORCE_MODE=1 claude
Sandbox background script subprocesses with PID namespace isolation and cap the number of script invocations per session.
$ CLAUDE_CODE_SUBPROCESS_ENV_SCRUB=1 CLAUDE_CODE_SCRIPT_CAPS=10 claude
Print the current prompt in a cache-friendly form that strips dynamic sections, useful for sharing stable prompt hashes across users in CI.
$ claude --print--exclude-dynamic-system-prompt-sections
›Adds interactive Google Vertex AI setup wizard from the login screen under '3rd-party platform', guiding through GCP auth, project/region config, credential verification, and model pinning.
›New CLAUDE_CODE_PERFORCE_MODE env var: Edit/Write/NotebookEdit fail on read-only files with a p4 edit hint instead of silently overwriting them.
›New Monitor tool for streaming events from background scripts.
›Adds subprocess sandboxing with PID namespace isolation on Linux when CLAUDE_CODE_SUBPROCESS_ENV_SCRUB is set; new CLAUDE_CODE_SCRIPT_CAPS env var limits per-session script invocations.
›New --exclude-dynamic-system-prompt-sections flag for print mode to improve cross-user prompt caching.
+8 moreshow less
›Adds workspace.git_worktree to status line JSON input, set when the current directory is inside a linked git worktree.
›Propagates W3C TRACEPARENT env var to Bash tool subprocesses when OTEL tracing is enabled, correctly parenting child-process spans in the trace tree.
›LSP: Claude Code now identifies itself to language servers via clientInfo in the initialize request.
›Improved /agents with a tabbed layout: Running tab shows live subagents; Library tab adds Run agent and View running instance actions.
›Improved /reload-plugins to pick up plugin-provided skills without requiring a restart.
›Improved Vim mode: j/k in NORMAL mode now navigate history and select the footer pill at the input boundary.
›Improved OTEL tracing: interaction spans now correctly wrap full turns under concurrent SDK calls, and headless turns end spans per-turn.
›Updated /claude-api skill to cover Managed Agents alongside Claude API.
GitHub Copilot CLI v1.0.22 adds bearer token BYOK auth, a skills field for custom agents, plugin post-install messages, and tighter sub-agent limits.
└──▷ GET THIS VERSION
$ git clone --branch v1.0.22 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:$ git checkout v1.0.22
›Custom agents can now declare a skills field in their configuration to eagerly load skill content into agent context at startup.
›Adds support for bearer token authentication for BYOK Anthropic models.
›CLI now reads MCP server configuration exclusively from .mcp.json; .vscode/mcp.json and .devcontainer/devcontainer.json are no longer recognized as MCP server config sources.
›Adds sub-agent depth and concurrency limits to prevent runaway agent spawning.
›Plugin agents now respect the model specified in their frontmatter.
+5 moreshow less
›Plugins can now display a post-install message with setup instructions after installation.
›Plugins remain enabled across sessions and auto-install on startup based on user config.
›sessionStart and sessionEnd hooks now fire once per session in interactive mode instead of once per prompt.
›Warns when resuming a session that is already in use by another CLI or application.
›Shows a clear message to contact your organization administrator when remote sessions are blocked by policy.
└──▷ BREAKING ON UPGRADE
!.vscode/mcp.json and .devcontainer/devcontainer.json are removed as MCP server config sources; the CLI now only reads .mcp.json. Existing setups relying on .vscode/mcp.json will stop loading MCP servers until config is migrated to .mcp.json (a migration hint appears when .vscode/mcp.json is detected without .mcp.json).
!sessionStart and sessionEnd hooks now fire once per session rather than once per prompt in interactive mode; setups that relied on per-prompt hook execution will behave differently.
llama.cpp b8739 adds AMD Instinct MI350X/MI355X (gfx950, CDNA4) GPU support via HIP backend
└──▷ GET THIS VERSION
$ git clone --branch b8739 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b8739
›Adds GGML_CUDA_CC_CDNA4 and GGML_CUDA_CC_IS_CDNA4 macros in common.cuh to identify AMD CDNA4 (gfx950) architecture at build time.
›Supports AMD Instinct MI350X/MI355X (gfx950, CDNA4) in the HIP backend, enabling GPU-accelerated inference on these accelerators when building with -DAMDGPU_TARGETS=gfx950.
›Routes CDNA4 matrix operations to compatible MFMA instructions: mfma_f32_16x16x4f32 for f32, mfma_f32_16x16x16bf16_1k for bf16, and mfma_i32_16x16x32_i8/mfma_i32_32x32x16_i8 for int8.
›Includes CDNA4 (gfx950) in stream-k kernel dispatch in mmq.cuh, enabling MMQ inference on MI355X.
llama.cpp b8738 adds experimental backend-agnostic tensor parallelism with NCCL/RCCL support across up to 8 GPUs.
└──▷ GET THIS VERSION
$ git clone --branch b8738 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b8738
›Introduces experimental backend-agnostic tensor parallelism via ggml-backend-meta.cpp and the new ggml-ext.h staging API header, supporting arbitrary numbers of GPUs and uneven tensor splits.
›Adds NCCL support for CUDA multi-GPU tensor parallelism, enabling AllReduce operations across devices using pinned (vs. pageable) memory when all backends expose the same host buffer type.
›Adds RCCL support for HIP/ROCm multi-GPU tensor parallelism, extending the same backend-agnostic tensor parallel path to AMD hardware.
›Extends tensor parallelism support to GPT-OSS, Qwen 3 MoE, Qwen 3.5, Qwen 3.5 MoE, and Gemma 4 MoE model architectures.
›Moves most meta backend public API into ggml-backend-impl.h, hiding previously exposed internals and introducing llama_device as the new abstraction for device selection.
+1 moreshow less
›Uses BF16 precision for AllReduce operations in the tensor parallel path.
└──▷ BREAKING ON UPGRADE
!ggml_backend_dev_is_meta() has been removed; callers must migrate to the new llama_device abstraction.
SYCL Flash Attention now supports head size 512 (DKQ/DV), up from the previous 256 limit.
└──▷ GET THIS VERSION
$ git clone --branch b8724 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b8724
›Extends SYCL Flash Attention (both tile and vector kernels) to support head sizes (DKQ/DV) of 512, previously capped at 256, enabling larger-model attention workloads on SYCL backends.
vMLX v1.3.34 adds deferred SSM re-derive for hybrid thinking models, cutting multi-turn cache misses on Nemotron and Qwen3.5-VL.
└──▷ GET THIS VERSION
$ git clone --branch v1.3.34 https://github.com/jjang-ai/vmlx.git
# already have the repo? check out this version:$ git checkout v1.3.34
›Adds deferred SSM re-derive: after generation completes on hybrid SSM + thinking models (Nemotron, Qwen3.5-VL), the scheduler queues an idle-time prefill pass on prompt tokens only, storing clean SSM state so the next conversation with the same prompt prefix gets KV + SSM cache hits instead of a full re-prefill.
›Adds _ensure_mx_array() for dtype normalization: input_ids is cast to mx.int32 and pixel_values to mx.array, enabling batched VLM inference on Mistral 3 / Pixtral.
›mx.compile() now targets language_model.model only for VLM engines, preserving the wrapper's .config attribute and fixing JIT + MLLM compatibility.
›_get_n_kv_heads() now scans cfg.text_config as a nested fallback, resolving 100% prefix-cache miss on Gemma 4 VLM where num_key_value_heads lives inside model.config.text_config.
›Adds --prefix-cache-max-bytes CLI flag to enforce a byte-budget eviction limit on the prefix cache.
›Exposes size and max_entries fields on the /v1/cache stats endpoint for live cache introspection.
›Adds cache_type column and per-type bucket eviction to the disk SQLite cache, with priority eviction and last-fetch type round-trip in L1 backfill.
›Extracts SSMCompanionCache to vmlx_engine/utils/ssm_companion_cache.py with a per-entry is_complete flag for SSM companion cache lifecycle tracking.
›Adds compute_model_cache_key() for content-derived cache identity that survives JIT reload and prevents cross-config pollution.
+13 moreshow less
›Adds _compute_segment_boundaries helper to engine/batched.py, firing segment boundaries on memory-aware and paged schedulers (not just legacy).
›Adds is_hybrid_ssm_cache, is_hybrid_ssm_model, and is_hybrid_ssm_config polymorphic helpers for hybrid SSM model detection.
›Adds is_mla_model centralized helper that walks top → text_config → language_config → llm_config for MLA model detection.
›Adds Gemma 4 native text MoE local-port shim with forward-compat that defers to upstream when available.
›Sets trust_remote_code=True as the default in utils/tokenizer.py, unblocking custom tokenizers without manual configuration.
›Enables LRU+Trie cross-session prefix sharing on MemoryAwarePrefixCache (production default) with system → user → assistant priority eviction.
›Adds cache_type LRU and per-type bucket eviction on BlockAwarePrefixCache for paged and hybrid auto-switch scenarios.
›Surfaces SSM companion cache stats (nbytes_mb) to the CachePanel UI.
›Backports BatchMambaCache.lengths, advance, prepare, and finalize from mlx-lm 0.31.2.
›Caches is_mllm_model() result in api/utils.py to eliminate repeated INFO log spam.
›Adds scheduler.deep_reset wiring of reset_factory_cache() on shutdown.
›Bumps mlx-lm dependency from 0.30.2 to 0.31.2.
›Removes orphan disk-streaming artifacts (ssd_generate, streaming_wrapper, weight_index) superseded by the memory_cache and disk_cache architecture.
└──▷ BREAKING ON UPGRADE
!Orphan disk-streaming artifacts ssd_generate, streaming_wrapper, and weight_index have been removed; setups relying on them must migrate to the memory_cache + disk_cache architecture.
THE AI TOOLCHAIN