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.
Extensions can now dynamically unregister providers and register them at runtime without requiring /reload.
└──▷ GET THIS VERSION
$ git clone --branch v0.55.2 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:$ git checkout v0.55.2
└──▷ TRY IT
Swap in a custom provider at runtime from a command handler and later cleanly remove it, restoring the original built-in models — all without issuing /reload.
$ pi.unregisterProvider("my-custom-provider")
›Adds pi.unregisterProvider(name) to dynamically remove a custom provider and its models from the registry, restoring any overridden built-in models.
›pi.registerProvider() now takes effect immediately when called outside the initial extension load phase (e.g. from a command handler), eliminating the need for /reload after late registrations.
Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.
CrewAI 1.10.0 adds user input handling in Flows, enhanced MCP tool resolution, and auto-updating tool specs.
└──▷ GET THIS VERSION
$ git clone --branch 1.10.0 https://github.com/crewAIInc/crewAI.git
# already have the repo? check out this version:$ git checkout 1.10.0
›Adds user input handling in Flows, enabling human-in-the-loop data collection mid-flow execution.
›Enhances MCP tool resolution and emits related events for better observability of MCP-based tool use.
›Adds started_event_id tracking on the event bus for correlating flow and agent lifecycle events.
›Introduces auto-update of tools.specs, keeping tool specifications current without manual intervention.
›Enhances HITL (Human-in-the-Loop) self-loop functionality in human feedback integration.
+4 moreshow less
›Improves JSON argument parsing and validation in CrewAgentExecutor and BaseTool for more robust tool invocation.
›Migrates CLI HTTP client from requests to httpx, enabling async-compatible CLI operations.
›Adds versioned documentation support and yanked-version detection in release notes.
›Updates LanceDB version and adds lance-namespace packages for expanded vector store integration.
Haystack v2.25.1 extends auto variadic sockets to support Optional[list[...]] input types.
└──▷ GET THIS VERSION
$ git clone --branch v2.25.1 https://github.com/deepset-ai/haystack.git
# already have the repo? check out this version:$ git checkout v2.25.1
›Auto variadic sockets now support Optional[list[...]] input types in addition to plain list[...], enabling nullable list inputs to participate in variadic connection fan-in.
Bind an agent to an account-scoped route for centralized routing management.
$ openclaw agents bind <agent> <channel>
›New openclaw secrets workflow with audit, configure, apply, and reload subcommands for External Secrets Management, including runtime snapshot activation and ref-only auth-profile support.
›ACP agents are now first-class runtimes for thread sessions with acp spawn/send dispatch integration, acpx backend bridging, lifecycle controls, and coalesced thread replies.
›New openclaw agents bindings, openclaw agents bind, and openclaw agents unbind commands for account-scoped route management, including channel-to-account binding upgrades and role-aware binding identity handling.
›Adds openai-codex WebSocket-first transport by default (transport: "auto" with SSE fallback), with explicit per-model/runtime transport overrides retained.
›Channel plugins can now own interactive onboarding flows via optional configureInteractive and configureWhenConfigured hooks.
+3 moreshow less
›Adds an explicit account-risk warning and confirmation gate before starting Gemini CLI OAuth.
›New Android device capability with device.status and device.info node commands, including protocol and registry coverage for device status/info payloads.
›Adds notifications.list support on Android nodes and exposes nodes notifications_list in agent tooling for listing active device notifications.
└──▷ BREAKING ON UPGRADE
!openclaw onboard --reset now defaults to config+creds+sessions scope; workspace deletion requires --reset-scope full.
llama.cpp b8179 adds CDNA3 MFMA tensor core flash attention for AMD MI300X, yielding up to +39% prompt-processing throughput.
└──▷ GET THIS VERSION
$ git clone --branch b8179 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b8179
›Adds CDNA3 MFMA tensor core support for the flash attention MMA kernel on AMD MI300X (gfx942), using v_mfma_f32_16x16x16_f16 (FP16 input, FP32 accumulate), with head sizes 64, 80, 96, 112, and 128 supported — delivering pp512 +7%, pp1024 +13%, pp2048 +23%, pp4096 +39% on Qwen2.5-1.5B Q4_K_M.
›Routes MI300X flash attention automatically: MMA kernel used for prompt processing (effective batch eff_nq >= 128), tile kernel for small batches — no manual dispatch required.
›Adds comma-separated value support to --alias in the llama.cpp server, allowing a single model to be registered under multiple aliases simultaneously (stored as a deduplicated std::set).
›Adds new --tags flag to the llama.cpp server for attaching informational metadata to a model (not used for routing).
›The /v1/models endpoint now exposes aliases and tags fields, reflecting all registered aliases and tags for each loaded model.
llama.cpp b8164 adds option to merge gate and up (exp) weights for MoE models during GGUF conversion.
└──▷ GET THIS VERSION
$ git clone --branch b8164 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b8164
›Adds create_tensor_gate_up helper and gate_up merged-weight support to convert_hf_to_gguf.py for all MoE models, enabling fused gate/exp weight tensors in converted GGUF files.
Phoenix v13.5.0 adds configurable password policy and GET endpoints for the sessions REST API.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v13.5.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v13.5.0
›Adds GET endpoints to the sessions REST API, enabling programmatic retrieval of session data.
›Adds configurable password policy support for tighter authentication control.
Phoenix Evals v2.11.0 adds conciseness and refusal evaluators plus a utility to convert Phoenix prompts to eval templates.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-evals-v2.11.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-evals-v2.11.0
›Adds a conciseness classification evaluator to assess whether LLM responses are appropriately brief.
›Adds a refusal evaluator to detect when an LLM declines to answer a query.
›Adds a utility to convert Phoenix prompts into evals templates, enabling reuse of existing prompt assets in evaluation pipelines.
Weave v0.52.29 adds unfinished-call metadata to usage APIs, a python-magic MIME backend, real-time thread tracking, tags/aliases schema, and a new eval_results/query API.
└──▷ GET THIS VERSION
$ git clone --branch v0.52.29 https://github.com/wandb/weave.git
# already have the repo? check out this version:$ git checkout v0.52.29
›Adds eval_results/query API endpoint for retrieving structured evaluation data.
›Adds metadata about unfinished calls to usage APIs, surfacing in-flight call state.
›Adds tags and aliases schema to the data model.
›Adds python-magic backend for MIME-type detection.
›Adds thread tracking and usage reporting for real-time (OpenAI Realtime API) calls.
+2 moreshow less
›Adds API for editing annotation queue metadata.
›Adds claude-sonnet 4.6 to the supported model providers list.
Chroma 1.5.2 adds FTS opt-out in schema, a Perplexity embedding function, Client context manager support, and quantized SPANN indexing.
└──▷ GET THIS VERSION
$ git clone --branch 1.5.2 https://github.com/chroma-core/chroma.git
# already have the repo? check out this version:$ git checkout 1.5.2
└──▷ USE IT
Ensure Chroma client connections are always released, even if an exception occurs — useful in scripts and CI pipelines.
python
import chromadb
with chromadb.HttpClient(host='localhost', port=8000) as client:
col = client.get_or_create_collection('my_collection')
results = col.query(query_texts=['threat actor TTPs'], n_results=5)
print(results)
›Adds close() method and context manager (with statement) support to the Python Client class for deterministic resource cleanup.
›Allows users to disable full-text search (FTS) per collection via a new field in the collection schema.
›Promotes the Advanced Search API out of beta (beta label removed).
›Adds a Perplexity embedding function (Pplx EF) as a new built-in integration.
›Introduces a quantized SPANN segment writer and reader, wired into the compaction and query orchestration pipelines for more memory-efficient ANN indexing.
+9 moreshow less
›Adds garbage collection for usearch index files to reclaim disk space.
›Adds balanced split logic for SPANN segments, with configurable recursion depth control.
›Enables delete_collection support in the multi-collection/multi-region (MCMR) architecture.
›Adds ignore_dirty column to the WAL3 manifests table for improved compaction control.
›Adds separate concurrency limit for manifest loads in the log-service.
›Adds configuration for the fetch_log semaphore to tune log-fetch concurrency.
›Allows rebuild operations to specify individual segments as targets.
›Batches sysdb queries during collection enrichment in the scheduler for improved throughput.
›Uses cluster average as SPANN center, improving index quality.
THE AI TOOLCHAIN