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.
Hermes v2026.3.23 adds an OpenAI-compatible API server, 6 messaging adapters, 4 inference providers, and MCP OAuth 2.1 management.
└──▷ GET THIS VERSION
$ git clone --branch v2026.3.23 https://github.com/NousResearch/hermes-agent.git
# already have the repo? check out this version:$ git checkout v2026.3.23
└──▷ TRY IT
Inject a local config file and a live URL into context before asking a question — useful for grounding the agent in current data without copy-pasting.
$ hermes chat '@file config/rules.yaml @url https://example.com/policy "Does our config comply with this policy?"'
›Adds OpenAI-compatible /v1/chat/completions API server with /api/jobs REST API for cron job management, SQLite-backed response persistence, input limits, field whitelists, and CORS origin protection.
›Adds 6 new messaging platform adapters: Signal, DingTalk, SMS (Twilio), Mattermost, Matrix, and Webhook — joining existing Telegram, Discord, and WhatsApp support, with exponential-backoff auto-reconnect.
›Adds @file and @url context injection with tab completions in the CLI for Claude Code-style context references.
›Adds 4 new inference providers: GitHub Copilot (OAuth + token validation, 400k context), Alibaba Cloud / DashScope, Kilo Code, and OpenCode Zen/Go.
›Adds hermes mcp command suite for installing, configuring, and authenticating MCP servers with full OAuth 2.1 PKCE flow.
+20 moreshow less
›Adds gateway prompt caching — caches AIAgent instances per session to preserve Anthropic prompt cache across turns, reducing cost on long conversations.
›Adds NeuTTS local TTS provider backend with built-in setup flow.
›Enables CLI streaming by default with spinner and tool progress display during streaming mode.
›Adds /statusbar command to toggle a persistent config bar showing model and provider info in the prompt.
›Adds /queue command to queue prompts for the agent without interrupting the current run.
›Adds /permission command to switch approval mode dynamically during a session.
›Adds /browser command for interactive browser sessions from the CLI.
›Adds /cost command for live pricing and usage tracking in gateway mode.
›Adds /approve and /deny commands replacing bare text approval in gateway mode.
›Adds show_reasoning config option to display reasoning/thinking blocks during inference.
›Adds context pressure warnings for CLI and gateway modes.
›Adds ${ENV_VAR} substitution support in config.yaml.
›Adds real-time config reload — config.yaml changes apply without restarting the agent.
›Adds custom_models.yaml for user-managed model additions.
›Adds priority-based context file selection with CLAUDE.md support.
›Adds merging of nested YAML sections instead of full replacement on config update.
›Adds overhaul of context compression with structured summaries, iterative updates, token-budget tail protection, configurable summary endpoint, and fallback model support.
›Adds context length detection overhaul with models.dev integration, provider-aware resolution, fuzzy matching for custom endpoints, and /v1/props support for llama.cpp.
›Adds eager fallback to a backup model on rate-limit errors.
›Adds endpoint metadata support for querying local servers for actual context window size and pricing.
└──▷ BREAKING ON UPGRADE
!The ANTHROPIC_BASE_URL environment variable is removed to avoid collisions — configurations relying on it will no longer apply.
›Adds Agent.from_file for loading agents from YAML/JSON files, with templated instructions that reference deps via TemplateStr.
›Adds AbstractToolset.for_run and for_run_step methods for per-run and per-step state isolation in toolsets.
›Adds Capabilities: composable, reusable units of agent behavior that bundle tools, lifecycle hooks, instructions, and model settings into a single class pluggable into any agent.
›Adds AgentSpec for declarative agent definitions loadable from YAML/JSON.
›Adds Hooks capability for defining lifecycle hooks using decorators.
+3 moreshow less
›Adds Thinking capability and a cross-provider thinking model setting.
›Adds provider-adaptive tool capabilities WebSearch, WebFetch, MCP, and ImageGeneration that automatically fall back from builtin (provider) tools to local tools.
›Adds openai:gpt-5.4-mini and openai:gpt-5.4-nano model identifiers.
llama.cpp b8498 adds standard Hugging Face cache support, migrating manifests and using the HF API to resolve cached model files.
└──▷ GET THIS VERSION
$ git clone --branch b8498 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b8498
›Adds standard Hugging Face cache support: uses the HF API to enumerate all files, migrates existing manifests to the Hugging Face cache layout at startup, and falls back to cached files when the HF API is unavailable.
oobabooga textgen v4.2 adds an Anthropic-compatible /v1/messages endpoint and expands --extra-flags to accept literal flags.
└──▷ GET THIS VERSION
$ git clone --branch v4.2 https://github.com/oobabooga/textgen.git
# already have the repo? check out this version:$ git checkout v4.2
└──▷ TRY IT
Point Claude Code at your local textgen instance to use local models via the new Anthropic-compatible endpoint.
$ ANTHROPIC_BASE_URL=http://127.0.0.1:5000 claude
›Adds /v1/messages endpoint for Anthropic API compatibility, enabling Claude Code, Cursor, and other Anthropic clients to connect; supports system messages, content blocks, tool use, tool results, image inputs, and thinking blocks.
›Expands --extra-flags to accept literal flags directly (e.g. --extra-flags "--rpc 192.168.1.100:50052 --jinja"), in addition to the existing key=value format.
›Moves the OpenAI-compatible API from extensions/openai to modules/api; the old --extensions openai flag remains accepted as an alias for --api.
›Sets top_p=0.95 as the new default sampling parameter for API requests.
›Enables gradient_checkpointing by default during training to reduce VRAM usage.
+4 moreshow less
›Removes the higher_rank_limit training parameter.
›Updated UI theme with new colors, borders, and button styles across light and dark modes.
›Removes 52 obsolete instruction templates from 2023 (Airoboros, Baichuan, Guanaco, Koala, Vicuna v0, MOSS, and others).
›Reduces portable build sizes by switching to a stripped Python distribution.
└──▷ BREAKING ON UPGRADE
!The default top_p for API requests is now 0.95; any setup relying on the previous default will receive different sampling behavior without explicit configuration.
!The higher_rank_limit training parameter has been removed; configs or scripts referencing it will need to be updated.
!52 instruction templates (Airoboros, Baichuan, Guanaco, Koala, Vicuna v0, MOSS, and others) have been removed; any saved configurations referencing these templates will lose them.
Phoenix Evals v2.12.0 adds support for structured data as eval inputs.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-evals-v2.12.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-evals-v2.12.0
›Accepts structured data (not just plain text) as inputs to evaluators, enabling richer eval pipelines.
Phoenix 13.18.0 adds a prompt version diff view and replaces markdown rendering with streamdown.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v13.18.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v13.18.0
›Adds a prompt version diff view in the UI, enabling side-by-side comparison of prompt versions.
›Replaces all markdown rendering with the streamdown renderer across the application.
Langfuse v3.162.0 adds error-level filtering in dashboard widgets, paginated experiments, SSE streaming for query progress, and gzip blob exports.
└──▷ GET THIS VERSION
$ git clone --branch v3.162.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:$ git checkout v3.162.0
›Adds an environment variable to enable single-level query optimization for v1 ClickHouse queries, allowing operators to tune query performance without code changes.
›Adds a Server-Sent Events (SSE) streaming endpoint for ClickHouse query progress, enabling dashboards to surface live query status.
›Adds gzip compression support for blob storage exports, reducing export payload sizes.
›Enables error-level filtering in custom dashboard widgets so practitioners can scope charts to specific log severity levels.
›Adds pagination to the experiments list for navigating large experiment sets.
Adds workbench.enable session config option to exclude code execution tools from a session entirely.
└──▷ 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]
└──▷ USE IT
Disable the workbench in a session to prevent any code execution tools from being available — useful for read-only or restricted agent environments.
json
{
"workbench": {
"enable": false
}
}
›Adds workbench.enable to session config (default true); setting it to false excludes COMPOSIO_REMOTE_WORKBENCH and COMPOSIO_REMOTE_BASH_TOOL from the session, disabling the workbench entirely.
Composio CLI 0.2.8 scopes commands by role, adds proxy execute, and introduces a workbench disable option.
└──▷ 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]
›Adds composio manage ... as the developer-scoped namespace, moving developer-only operations under it and removing those flags from the root help.
›Moves composio search, composio link, and composio execute to consumer-only top-level commands, with short related-command hints added to root help.
›Adds proxy execute capability via composio execute, defaulting to an empty object {} when no -d/--data flag or piped stdin is provided.
›Adds workbench.enable config option to disable the workbench in sessions.
›Search CTA now uses -d '{}' (shell-safe) for tools with no schema properties.
└──▷ BREAKING ON UPGRADE
!Developer-only flags previously available at the root CLI level are removed from root help; developer workflows must now use composio manage ....
Milvus Go SDK v2.6.3 adds TruncateCollection, replication config API, nullable pointer fields, and per-cluster mTLS for CDC.
└──▷ GET THIS VERSION
$ git clone --branch client/v2.6.3 https://github.com/milvus-io/milvus.git
# already have the repo? check out this version:$ git checkout client/v2.6.3
└──▷ USE IT
Quickly wipe all data from a collection during a test or data-refresh workflow without dropping and recreating the schema.
›Adds TruncateCollection method to the Go SDK client, invoked via NewTruncateCollectionOption(collectionName), to clear all data from a collection without dropping and recreating it.
›Adds GetReplicateConfiguration API for viewing replication topology (with tokens redacted) and a force_promote field to UpdateReplicateConfigurationRequest; supports strong consistency via the WithFreshRead option.
›Adds TLSConfig field to ClientConfig and a BuildTLSConfig helper for mTLS support in CDC; NewMilvusClient can now read per-cluster TLS config by target cluster ID via GetClusterTLSConfig(clusterID).
›Go pointer struct fields (*string, *int32, etc.) now represent nullable columns in the row-based data path — a nil pointer maps to NULL — across ParseSchema, AnyToColumns, SetField, fillData, and fillPKEntry.
THE AI TOOLCHAIN