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 v2.6.3 adds WorkspaceContextProvider for project-aware repo context and expands SlackContextProvider with opt-in workspace search.
└──▷ GET THIS VERSION
$ git clone --branch v2.6.3 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:$ git checkout v2.6.3
└──▷ USE IT
Enable Slack workspace search in SlackContextProvider for broader channel discovery without using the removed factory methods.
python
from agno.context.slack import SlackContextProvider
provider = SlackContextProvider(enable_workspace_search=True)
context = provider.get_context()
›Adds WorkspaceContextProvider, a project-aware context provider for repository roots backed by the read-only Workspace toolkit instead of generic FileTools; centralizes local filesystem exclude patterns so both FileTools and Workspace skip .context, .venvs, and other agent/dependency/build noise by default.
›Adds exclude_patterns parameter to FilesystemContextProvider for explicit opt-out or customization of filesystem exclusions.
›Adds opt-in enable_workspace_search parameter to SlackContextProvider; tools are now self-documenting via SlackTools, removing runtime agent switching.
›Removes for_bot_read(), for_assistant_search(), and for_write() factory methods from SlackContextProvider in favor of explicit flags for direct construction.
└──▷ BREAKING ON UPGRADE
!The for_bot_read(), for_assistant_search(), and for_write() factory methods have been removed from SlackContextProvider; callers must switch to explicit flags on construction.
Adds tool_name and call_id convenience properties to tool items in the OpenAI Agents SDK.
└──▷ GET THIS VERSION
$ git clone --branch v0.14.7 https://github.com/openai/openai-agents-python.git
# already have the repo? check out this version:$ git checkout v0.14.7
└──▷ USE IT
Access the tool name and call ID directly from a tool item when handling tool call results in an agent run.
python
# Given a tool item from an agent run result
for item in result.tool_items:
print(item.tool_name) # e.g. 'search_web'
print(item.call_id) # e.g. 'call_abc123'
›Adds tool_name and call_id convenience properties to tool items, making it easier to inspect tool call context without manual attribute lookup.
›Adds openclaw matrix encryption setup to enable Matrix E2EE, bootstrap recovery, and print verification status in a single flow.
›Adds openclaw migrate with plan, dry-run, JSON output, pre-migration backup, onboarding detection, and a bundled Hermes importer for config, memory, plugins, model providers, MCP servers, skills, and credentials.
›Adds a bundled Claude importer to preview and apply Claude Code and Claude Desktop instructions, MCP servers, skills, command prompts, and archived state.
›Adds Cerebras as a bundled provider plugin with onboarding, static model catalog, docs, and manifest-owned endpoint metadata.
›Adds optional memorySearch.inputType, queryInputType, and documentInputType config for asymmetric embedding endpoints on OpenAI-compatible memory backends.
+7 moreshow less
›Adds model-specific retrieval query prefixes for nomic-embed-text, qwen3-embedding, and mxbai-embed-large in Ollama memory-search queries.
›Adds a raw config pending-changes diff panel in Control UI that parses JSON5 and redacts sensitive values until revealed.
›Adds opt-in agents.defaults.compaction.maxActiveTranscriptBytes preflight trigger to run local compaction when active JSONL grows too large.
›Adds openclaw nodes remove --node <id|name|ip> and node.pair.remove to clean stale gateway-owned node pairing records.
›Adds a generic browser realtime transport contract, Google Live browser Talk sessions with constrained ephemeral tokens, and a Gateway relay for backend-only realtime voice plugins.
›Moves model-id normalization, provider endpoint host metadata, and OpenAI-compatible request-family hints into plugin manifests, removing bundled-provider routing tables from core.
›Allows OPENCLAW_PLUGIN_STAGE_DIR to contain layered runtime-dependency roots, resolving read-only preinstalled deps before installing missing deps into the writable root.
└──▷ BREAKING ON UPGRADE
!Direct plugin config load/write helpers are deprecated in favor of passed runtime snapshots and transactional mutation helpers with explicit restart follow-up policy; existing callers must migrate to the new snapshot-based API.
HolmesGPT 0.26.0 introduces Skills extensibility, VictoriaLogs toolset, and new Grafana timeout controls
└──▷ GET THIS VERSION
$ git clone --branch 0.26.0 https://github.com/HolmesGPT/holmesgpt.git
# already have the repo? check out this version:$ git checkout 0.26.0
›Adds extraVolumes, extraVolumeMounts, and extraInitContainers configuration keys to the AWS MCP toolset for flexible volume and init-container injection.
›Adds max retries and timeout configuration for Grafana integrations.
›Introduces Skills, a new extensibility model enabling custom tool integrations and workflows via configuration.
›Adds a VictoriaLogs toolset for log search and analysis.
›Adds support for logging aggregated investigation token usage.
!The Skills functionality introduces a new extensibility model — existing custom tool integrations and workflows may need to be migrated to the new Skills configuration format.
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 v2.1.122 adds Bedrock service tier selection, PR-URL session lookup, and new OpenTelemetry events.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.122 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.122
└──▷ TRY IT
Route Bedrock API calls to a specific service tier (e.g., priority) for latency-sensitive workloads.
$ ANTHROPIC_BEDROCK_SERVICE_TIER=priority claude --model anthropic.claude-opus-4-5
›Adds ANTHROPIC_BEDROCK_SERVICE_TIER env var to select a Bedrock service tier (default, flex, or priority), sent as the X-Amzn-Bedrock-Service-Tier header.
›Pasting a GitHub, GitHub Enterprise, GitLab, or Bitbucket PR URL into the /resume search box now finds the session that created that PR.
›Adds claude_code.at_mention OpenTelemetry log event for @-mention resolution.
›OpenTelemetry numeric attributes on api_request/api_error log events are now emitted as numbers instead of strings.
›The /mcp view now surfaces claude.ai connectors hidden by a manually-added server with the same URL, with a hint to remove the duplicate.
└──▷ BREAKING ON UPGRADE
!OpenTelemetry consumers parsing numeric attributes on api_request/api_error log events as strings will receive numbers instead — pipelines that treat those fields as strings may need to be updated.
Claude Code v2.1.121 adds MCP alwaysLoad, plugin pruning, PostToolUse output replacement, mTLS Workload Identity, and expanded OpenTelemetry spans.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.121 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.121
└──▷ USE IT
Pin a high-priority MCP server's tools so they are always available without waiting for tool-search deferral — useful for servers you rely on in every session.
json
# In your MCP server config (e.g. .claude/mcp_settings.json)
{
"mcpServers": {
"my-server": {
"command": "my-mcp-server",
"alwaysLoad": true
}
}
}
Clean up orphaned plugin dependencies after removing plugins, or cascade-remove a plugin and all its unique dependencies in one step.
$ # Prune all orphaned dependencies
claude plugin prune
# Or remove a plugin and cascade its dependencies
claude plugin uninstall my-plugin --prune
›Adds alwaysLoad option to MCP server config — when true, all tools from that server bypass tool-search deferral and are always available.
›Adds claude plugin prune command to remove orphaned auto-installed plugin dependencies; plugin uninstall --prune cascades removals.
›PostToolUse hooks can now replace tool output for ALL tools via hookSpecificOutput.updatedToolOutput, not just MCP tools.
›Adds type-to-filter search box to /skills for quickly finding skills in long lists.
›OpenTelemetry spans now include stop_reason, gen_ai.response.finish_reasons, and user_system_prompt (gated behind OTEL_LOG_USER_PROMPTS) on LLM request spans.
›SDK: mcp_authenticate now supports redirectUri for custom scheme completion and claude.ai connectors.
›Enables CLAUDE_CODE_FORK_SUBAGENT=1 in non-interactive sessions for SDK and claude -p.
›--dangerously-skip-permissions no longer prompts for writes to .claude/skills/, .claude/agents/, and .claude/commands/.
›MCP servers that hit a transient error at startup now auto-retry up to 3 times instead of staying disconnected.
›Terminal tab session title is now generated in the configured language setting.
›LSP diagnostic summaries now expand on click or Ctrl+O and show an expand hint.
›Scrollable overflow dialogs: terminal dialogs that exceed terminal height are now navigable with arrow keys, PgUp/PgDn, Home/End, and mouse wheel in both fullscreen and non-fullscreen modes.
›Clicking any line of a long URL wrapped across rows in fullscreen mode now opens the full URL.
›[VSCode] Voice dictation now respects the accessibility.voice.speechLanguage setting when no Claude Code language is configured.
›[VSCode] /context now opens a native token usage dialog.
Pi v0.70.6 adds Cloudflare Workers AI as a built-in provider with API key, default model, and /login support.
└──▷ GET THIS VERSION
$ git clone --branch v0.70.6 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:$ git checkout v0.70.6
›Adds Cloudflare Workers AI as a built-in provider, configured via CLOUDFLARE_API_KEY and CLOUDFLARE_ACCOUNT_ID, with default model resolution and /login support.
Qwen Code SDK TypeScript v0.1.7 adds session resumption, multi-modal input, concurrent batch execution, and a new settings.env config field.
└──▷ GET THIS VERSION
$ git clone --branch sdk-typescript-v0.1.7 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout sdk-typescript-v0.1.7
└──▷ TRY IT
Resume a previous session by ID to continue work without losing context.
$ qwen --session-id <your-session-id>
Set environment variables for a provider via the settings file instead of the shell environment.
Triton v2.68.0 disables client shared memory by default, adds ensemble queue limits, and brings OpenAI-compatible model control mode.
└──▷ GET THIS VERSION
$ git clone --branch v2.68.0 https://github.com/triton-inference-server/server.git
# already have the repo? check out this version:$ git checkout v2.68.0
└──▷ TRY IT
Re-enable client shared memory on upgrade, since it is now off by default and will silently break workloads that depend on it.
›Adds --allow-client-shm=true flag required to re-enable client shared memory, which is now disabled by default.
›Enforces max_inflight_requests as a shared limit across ensemble requests, preventing unbounded queue growth in ensemble pipelines.
›Adds support for explicit model control mode and model management in the OpenAI-compatible frontend.
›Adds model name validation at load time following the POSIX standard.
›Introduces safe GetElementCount and GetByteSize APIs with overflow protection across core and common libraries.
+1 moreshow less
›PyTorch backend adds Meta OpTree support.
└──▷ BREAKING ON UPGRADE
!Client shared memory is now disabled by default; servers that relied on the previous default must add --allow-client-shm=true to their startup command or shared-memory functionality will be unavailable.
llama.cpp b8966 adds CUDA flash-attention support for Mistral Small 4's non-standard head sizes (DKQ=320/DV=256) with GQA=32.
└──▷ GET THIS VERSION
$ git clone --branch b8966 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b8966
›Adds CUDA flash-attention (MMA-f16) kernel support for head sizes DKQ=320/DV=256 with ncols2=32, enabling GPU-accelerated inference for Mistral Small 4's GQA ratio of 32.
llama.cpp b8956 adds seven new CANN ops and fused GLU/cross-entropy kernels for Ascend NPU backends.
└──▷ GET THIS VERSION
$ git clone --branch b8956 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b8956
›Adds GGML_OP_SET to the CANN backend, implemented via aclnnInplaceCopy on the target region.
›Adds GGML_OP_CUMSUM to the CANN backend, implemented via aclnnCumsum.
›Adds GGML_OP_FILL to the CANN backend, implemented via aclnnInplaceFillScalar.
›Adds GGML_OP_DIAG to the CANN backend, implemented via aclnnInplaceCopy on diagonal strides.
›Adds GGML_OP_TRI (lower/lower_diag/upper_diag/upper modes) to the CANN backend via aclnnTril and aclnnTriu with appropriate diagonal offsets.
+5 moreshow less
›Adds GGML_OP_SOLVE_TRI to the CANN backend via aclnnTriangularSolve.
›Adds GGML_UNARY_OP_SOFTPLUS to the CANN backend via aclnnSoftplus.
›Fuses GLU variants (SwiGLU, GeGLU, GeGLU_ERF, GeGLU_QUICK) into single aclnnSwiGlu/aclnnGeGluV3 calls on the CANN backend when applicable.
›Replaces the five-kernel cross-entropy sequence (LogSoftmax→Mul→ReduceSum×2→Muls) with a single aclnnSoftmaxCrossEntropyWithLogits call on the CANN backend.
›Implements CANN backend memset_tensor via aclrtMemset, replacing a NULL stub.
Phoenix 14.16.0 adds page-context-aware agent chat, consolidated assistant config, and GraphQL field redaction.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v14.16.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v14.16.0
›Introduces RedactedString scalar to automatically redact sensitive fields returned by the GraphQL API.
›Agent assistant now advertises the current Phoenix page context to the chat, enabling context-aware responses.
›Moves advanced PXI toggles into the Assistant config panel, consolidating agent configuration in one place.
Composio [email protected] makes file upload/download opt-in, adding new constructor flags and a per-upload source hint for hooks.
└──▷ GET THIS VERSION
$ git clone --branch [email protected] https://github.com/ComposioHQ/composio.git
# already have the repo? check out this version:$ git checkout [email protected]
└──▷ USE IT
Restore the old auto file-staging behaviour after upgrading, while restricting uploads to a specific allowed directory.
›Adds dangerously_allow_auto_upload_download_files: bool = False to Composio(...) to opt back in to automatic file staging — when True, collapses file_uploadable schemas to {"type": "string", "format": "path"} for the model and stages local paths/URLs at execute time.
›Adds file_upload_dirs: Sequence[str] | Literal[False] | None = None to Composio(...) as a fail-closed allowlist for local upload paths — None defaults to [<home>/.composio/temp], False rejects all local paths, and an explicit list replaces the default.
›Adds file_download_dir: str | None to Composio(...) to set the directory used to stage downloads on file_downloadable results.
›The before_file_upload modifier hook now receives source: Literal['path', 'url', 'file'] so hooks can branch on the original input type.
›When auto-upload is off and a tool with file_uploadable inputs is executed, the SDK emits a one-shot warning per tool slug pointing at composio.files.upload() for manual staging.
└──▷ BREAKING ON UPGRADE
!Automatic file upload/download is now off by default — existing code that relied on auto-staging local paths/URLs or auto-downloading file_downloadable results will stop working unless dangerously_allow_auto_upload_download_files=True is set on Composio(...).
!The auto_upload_download_files constructor option is removed; code that sets it must migrate to dangerously_allow_auto_upload_download_files.
THE AI TOOLCHAIN