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.
Zed v0.99.1 adds text-transform commands, shell comment toggling, and syntax highlighting for Cargo.lock and .zprofile.
└──▷ GET THIS VERSION
$ git clone --branch v0.99.1 https://github.com/zed-industries/zed.git
# already have the repo? check out this version:$ git checkout v0.99.1
›Adds convert to kebab case, convert to snake case, convert to upper camel case, convert to lower camel case, convert to title case, convert to lower case, and convert to upper case commands to transform selected text.
›Adds cmd-/ keybinding to toggle line comments in shell files.
›Adds syntax highlighting support for Cargo.lock and .zprofile files.
LangChain v0.0.252 adds RSS/OPML loading, ScaNN vector store, a rephrasing retriever, spell correction for Google Enterprise Search, and more.
└──▷ GET THIS VERSION
$ git clone --branch v0.0.252 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:$ git checkout v0.0.252
└──▷ USE IT
Load and persist a TFIDFRetriever so a fitted vectorizer survives process restarts.
python
from langchain.retrievers import TFIDFRetriever
# Build and save
retriever = TFIDFRetriever.from_texts(["doc one", "doc two", "doc three"])
retriever.save_local("tfidf_index")
# Reload in a later session
loaded = TFIDFRetriever.load_local("tfidf_index")
Ingest an RSS or OPML feed as LangChain documents for downstream summarisation or RAG.
Use the ScaNN vector store for fast approximate nearest-neighbor retrieval over large embedding corpora.
python
from langchain.vectorstores import ScaNN
from langchain.embeddings import OpenAIEmbeddings
db = ScaNN.from_texts(texts, OpenAIEmbeddings())
results = db.similarity_search("lateral movement detection", k=5)
›Adds model_revision parameter to ModelScopeEmbeddings for pinning embedding model versions.
›Adds regex control over separators in the character text splitter.
›Adds save() and load() serializer methods to TFIDFRetriever, enabling persistence of the TF-IDF vectorizer and its documents.
›Adds load() deserializer function that bypasses the need for JSON serialization when rehydrating chains.
›Adds spell-correction spec support to the Google Cloud Enterprise Search connector.
+7 moreshow less
›Adds a page_content formatter to AmazonKendraRetriever for customizing how document content is surfaced.
›Adds support for arbitrary kwargs pass-through to the LlamaCpp LLM integration.
›Adds Azure Active Directory token-based authentication support for AzureChatOpenAI.
›New RSS Feed and OPML document loader for ingesting feed content into chains.
›New ScaNN vector store integration for approximate nearest-neighbor search.
›New rephrasing retriever that reformulates user inputs before retrieval.
›New deterministic fake embedding model for reproducible testing.
Semantic Kernel Python 0.3.10.dev adds token bias controls, HF model kwargs, chat template restore, and orchestration serialization.
└──▷ GET THIS VERSION
$ git clone --branch python-0.3.10.dev https://github.com/microsoft/semantic-kernel.git
# already have the repo? check out this version:$ git checkout python-0.3.10.dev
└──▷ USE IT
Bias the model toward or away from specific tokens when configuring a prompt template.
python
from semantic_kernel.connectors.ai.open_ai import OpenAITextPromptExecutionSettings
from semantic_kernel.prompt_template.prompt_template_config import PromptTemplateConfig
config = PromptTemplateConfig(
template="Classify the sentiment: {{$input}}",
token_selection_biases={1234: -100, 5678: 50}
)
›Adds token_selection_biases field to PromptTemplateConfig, enabling per-token logit bias control when building prompt templates.
›Adds HuggingFace model and pipeline kwargs to allow customization of model loading behavior in HF connectors.
›Adds restore method to chat_prompt_template for reconstructing chat prompt template state.
›Enables serialization support for the sk/orchestration module.
›Adds logger parameter (optional) to inheriting classes in the Semantic Kernel Python library.
1 more release in this issue
· 2023-08-23 → 2023-08-28
Semantic Kernel Python gains Google PaLM connectors, a stepwise planner, USearch memory, and single-function kernel registration.
└──▷ GET THIS VERSION
$ git clone --branch python-0.3.9.dev https://github.com/microsoft/semantic-kernel.git
# already have the repo? check out this version:$ git checkout python-0.3.9.dev
›Adds num_records parameter to the text memory skill, allowing callers to control how many memory records are retrieved.
›Adds Google PaLM connector supporting text completion, chat completion, and text embedding services.
›Adds stepwise planner to the Python SDK.
›Adds methods to register a single native function directly to the kernel without wrapping it in a skill/plugin.
›Adds USearch memory connector for vector memory storage.
+1 moreshow less
›Azure Cognitive Search memory connector now uses HNSW for vector indexing.
KoboldCpp v1.40.1a adds an idle field to /api/extra/perf and an aesthetic instruct UI option in Lite.
└──▷ GET THIS VERSION
$ git clone --branch v1.40.1a https://github.com/LostRuins/koboldcpp.git
# already have the repo? check out this version:$ git checkout v1.40.1a
└──▷ TRY IT
Poll whether KoboldCpp is currently busy generating before submitting a new request, avoiding queuing conflicts.
$ curl http://localhost:5001/api/extra/perf
›Adds idle field to the GET /api/extra/perf endpoint, allowing callers to check whether a generation is currently in progress without submitting a generation request.
›Adds an aesthetic instruct UI option (early beta) in the Lite interface.
›Converts the embedded horde worker to daemon mode to improve stability.
›Adds --usecublas mmq flag (also togglable in GUI) to enable Quantized MatMul (MMQ) prompt processing on CUDA, reducing memory usage and improving speed for Q4_0 models.
›Displays a terminal warning when received context exceeds the max launcher-allocated context.
›Adds a Help button in the GUI.
└──▷ BREAKING ON UPGRADE
!--noavx2 mode no longer uses OpenBLAS, now behaving the same as Failsafe mode.
›Adds GGUF model format support via bumped llama.cpp (see breaking changes).
+7 moreshow less
›Adds img2img and depthimg2img pipeline support to the diffusers backend, along with clip_skip and additional kernel schedulers.
›Adds DPMSolverMultistepScheduler++, DPMSolverMultistepSchedulerSDE++, and guidance_scale options to the diffusers backend.
›Overcomes the prompt length limit in the diffusers backend.
›Adds customizable RWKV tokenizer support.
›Adds usage tracking features (API-level usage collection).
›Allows restricting which backends are built via Makefile targets.
›Makes the gRPC backend initializer accept configurable delay times.
└──▷ BREAKING ON UPGRADE
!llama.cpp has been bumped with GGUF support added; existing GGML-format models may no longer be compatible and will require conversion or replacement with GGUF variants.
3 more releases in this issue
· 2023-08-02 → 2023-08-26
Triton v2.37.0 adds parallel model instance loading, PyTorch serving via torch.compile(), OpenTelemetry resource settings, and BLS trace capture.
└──▷ GET THIS VERSION
$ git clone --branch v2.37.0 https://github.com/triton-inference-server/server.git
# already have the repo? check out this version:$ git checkout v2.37.0
└──▷ USE IT
Explicitly restore pre-v2.37.0 response ordering behavior for an oldest-strategy sequence batcher after the default changes to false.
›Exposes preserve_ordering field to the oldest-strategy sequence batcher, letting operators control whether response order is preserved across independent requests belonging to different sequences.
›Triton OpenTelemetry Tracing exposes resource settings to configure service name and version via OpenTelemetry trace API settings.
›Supports loading model instances in parallel for backends that opt in via TRITONBACKEND_BackendAttributeSetParallelModelInstanceLoading; Python and ONNXRuntime backends enabled as of this release.
›Python backend models can now capture traces for composing child models when executing BLS (Business Logic Scripting) requests.
›Python backend supports directly loading and serving PyTorch models using torch.compile() (experimental PyTorch platform).
└──▷ BREAKING ON UPGRADE
!The default value of preserve_ordering in the oldest-strategy sequence batcher changes from true to false, meaning response order across independent requests from different sequences is no longer preserved by default on upgrade.
LanceDB python-v0.2.0 adds iterator-based ingestion, pydantic auto-conversion, drop_database, and renames the distance column.
└──▷ GET THIS VERSION
$ git clone --branch python-v0.2.0 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:$ git checkout python-v0.2.0
└──▷ USE IT
Ingest a large dataset from a generator without loading it all into memory at once.
python
import lancedb
def record_generator():
for i in range(100_000):
yield {"id": i, "vector": [float(i), float(i)], "text": f"item {i}"}
db = lancedb.connect("./mydb")
table = db.open_table("items")
table.add(record_generator())
Use the renamed _distance field to filter and rank vector search results after upgrading from v0.1.x.
›Adds drop_database method to programmatically delete an entire database.
›Supports adding records via Python iterators with table.add(), enabling streaming or lazily-generated data ingestion without materializing the full dataset in memory.
›Automatically converts Pydantic models to the appropriate schema when adding records, removing manual Arrow conversion steps.
›Makes schema a property on table objects for direct attribute-style access.
└──▷ BREAKING ON UPGRADE
!The score column returned by vector search is renamed to _distance; any code reading result['score'] must be updated to result['_distance'].
!schema is now a property instead of a method; any code calling .schema() must be updated to .schema.
Milvus 2.3.0 adds GPU support, upsert API, range search, cosine metrics, ScaNN index, iterators, JSON_CONTAINS, CDC, and NATS message queue.
└──▷ GET THIS VERSION
$ git clone --branch v2.3.0 https://github.com/milvus-io/milvus.git
# already have the repo? check out this version:$ git checkout v2.3.0
└──▷ USE IT
Retrieve only vectors whose distance from the query falls between 10 and 20 (L2), scoping results to a meaningful similarity band rather than a fixed top-K.
›Adds radius and range_filter parameters to search_params in the search API to retrieve vectors within a specific distance range (max 16,384 results).
›Adds JSON_CONTAINS filter expression to query entities whose JSON field values contain one or more specified elements.
›Adds native support for the upsert API, combining search, delete, and insert operations for ambiguous insert/update scenarios.
›Adds the count statement as an alternative to num_entities for counting entities in a collection without triggering flushes.
›Introduces iterator support in PyMilvus for retrieving more than 16,384 entities in search or range search operations, similar to Elasticsearch's scroll API.
+11 moreshow less
›Adds support for the ScaNN index (based on FAISS FastScan), delivering ~20% higher QPS than HNSW and ~7x improvement over IVF-FLAT in benchmarks.
›Adds native Cosine Metrics support, eliminating the need to quantize vectors for IP (Inner Product) comparisons.
›Enables raw vectors to be included in search results for supported metrics (HNSW, IVF_FLAT); not supported for IVF_PQ or IVF_SQ8.
›Introduces MMap-based capacity expansion, mapping disk space to memory to increase single-machine data capacity with ~20% performance trade-off.
›Adds Change Data Capture (CDC) support for active/standby synchronization, incremental backup, and data migration scenarios.
›Adds GPU-accelerated query support via NVIDIA RAFT algorithm integration, achieving up to 10x QPS improvement on certain datasets.
›Adds Arm64 (aarch64) Docker images, enabling native support on Arm-based cloud instances and Apple Silicon (MacOS) systems.
›Introduces an experimental NATS-based built-in message queue as an alternative to Pulsar and Kafka.
›Merges IndexCoord and DataCoord into a single component, simplifying Milvus deployment topology.
›Adds a growing index for streaming (unindexed) data segments to avoid brute-force search degradation during ingestion.
›Improves scalar query performance in hybrid searches by optimizing HNSW data-filtering policies.
└──▷ BREAKING ON UPGRADE
!The time-travel feature has been removed; queries relying on time-travel will no longer work.
!CentOS-based images are no longer provided; deployments using CentOS images must migrate to Amazonlinux or Ubuntu images.
!The ANNOY and RHNSW index types for float vectors have been removed; collections using these indexes must be rebuilt with a supported index type.
!The TANIMOTO index type for binary vectors has been removed; collections using this index must be rebuilt.
!The Superstructure and Substructure metric types have been removed; queries using these metrics will break.
2 more releases in this issue
· 2023-08-09 → 2023-08-23
›Adds ContainsAny and ContainsAll filter operators for easier filtering on array types and tokenized text fields.
›Introduces the text2vec-gpt4all module for local GPT4All-based text vectorization.
›Introduces the multi2vec-bind module for multi-modal vectorization via ImageBind.
›Adds opt-in pread as an alternative to mmap for LSM store access, improving performance and stability on disk-bound setups.
›Backup compression support: backups can now be compressed into pre-configurable chunks, reducing file operations and lowering S3/GCS storage costs.
+8 moreshow less
›Adds ability to deactivate tenants (experimental) so inactive tenants consume no resources, enabling denser multi-tenant deployments on the same node.
›Enforces a minimum replication factor according to system-wide configuration.
›Adds a configurable nested cross-reference query limit.
›Adds batch queue congestion info to node status.
›Adds gRPC batching support.
›Adds batch support in the reranker-transformers module.
›Enables creating object references without specifying ToClass.
›Adds NEON SIMD acceleration for L2 and dot-product distance calculations on ARM, improving HNSW vector search performance.
THE AI TOOLCHAIN