The AI Toolchain — issue -378, May 31, 2024

Aider v0.35.0 switches to GPT-4o as the default model and adds --restore-chat-history to resume prior sessions.

$ git clone --branch v0.35.0 https://github.com/Aider-AI/aider.git
# already have the repo? check out this version:
$ git checkout v0.35.0

$ aider --restore-chat-history

• ›Adds --restore-chat-history flag to restore prior chat history on launch, enabling continuation of the last conversation.
• ›Switches the default model to GPT-4o, which scores 72.9% on the aider LLM code editing leaderboard (up from 68.4% for Opus).
• ›Improves reflection feedback to LLMs when using the diff edit format.

Aider v0.34.0 adds --show-prompts debug switch and explicit model settings for Claude 3 Opus and GPT-3.5 Turbo.

$ git clone --branch v0.34.0 https://github.com/Aider-AI/aider.git
# already have the repo? check out this version:
$ git checkout v0.34.0

$ aider --show-prompts

• ›Adds --show-prompts debug switch to inspect the prompts sent to the model.
• ›Adds explicit model settings for openrouter/anthropic/claude-3-opus and gpt-3.5-turbo.

Aider v0.32.0 adds LLM code-editing leaderboards, a new diff-fenced edit format for Gemini 1.5 Pro, and Deepseek-V2 support.

$ git clone --branch v0.32.0 https://github.com/Aider-AI/aider.git
# already have the repo? check out this version:
$ git checkout v0.32.0

• ›Gemini 1.5 Pro now defaults to the diff-fenced edit format, improving performance on larger codebases.
• ›Adds support for Deepseek-V2 via more flexible system-message configuration in the diff edit format.
• ›Benchmark output now serializes results in YAML format, compatible with the new leaderboard.
• ›Publishes LLM code-editing leaderboards ranking GPT-3.5/4 Turbo, Opus, Sonnet, Gemini 1.5 Pro, Llama 3, Deepseek Coder, and Command-R+ by code-editing ability.

Aider v0.31.0 adds a browser UI mode and in-chat model switching.

$ git clone --branch v0.31.0 https://github.com/Aider-AI/aider.git
# already have the repo? check out this version:
$ git checkout v0.31.0

$ aider --browser

• ›Adds --browser flag to launch an experimental browser-based UI instead of the terminal interface.
• ›Adds /model <name> chat command to switch the active AI model mid-session.
• ›Adds /models <query> chat command to search the list of available models.

SWE-agent v0.4.0 launches a web UI for browser-based agent runs.

$ git clone --branch v0.4.0 https://github.com/SWE-agent/SWE-agent.git
# already have the repo? check out this version:
$ git checkout v0.4.0

• ›Adds a web UI that lets users specify a bug and watch SWE-agent work through it interactively in the browser.

SWE-agent v0.3.0 adds Amazon Bedrock (Claude), GPT-4 Turbo, and GitHub Codespaces cloud support.

$ git clone --branch v0.3.0 https://github.com/SWE-agent/SWE-agent.git
# already have the repo? check out this version:
$ git checkout v0.3.0

• ›Adds Amazon Bedrock support, enabling Claude models as a backend via the Bedrock integration.
• ›Adds GPT-4 Turbo as a supported model option.
• ›Enables running SWE-agent in the cloud using GitHub Codespaces.

Zed v0.136.2 adds Claude AI support, tab bar toggle, git subfolder awareness, Vim changelist, and GPT-4o as default model.

$ git clone --branch v0.136.2 https://github.com/zed-industries/zed.git
# already have the repo? check out this version:
$ git checkout v0.136.2

{
  "assistant": {
    "version": "1",
    "provider": {
      "name": "anthropic"
    }
  }
}

{
  "tab_bar": {
    "show": false
  }
}

$ zed --foreground

• ›Adds tab_bar.show key to settings.json to hide the tab bar entirely.
• ›Adds current_line_highlight setting to settings.json to control how the current line is highlighted in the editor.
• ›Adds --foreground CLI flag to run Zed on the current PTY.
• ›Adds Claude (Anthropic) as an AI provider in the assistant panel, configurable via assistant.provider.name: 'anthropic' in settings.json.
• ›Adds low_speed_timeout_in_seconds setting to the assistant's OpenAI provider configuration in settings.json.

Zed v0.135.2 adds git hunk diff browsing, stop_at_soft_wraps setting, TypeScript function call completion, and Rust test gutter buttons.

$ git clone --branch v0.135.2 https://github.com/zed-industries/zed.git
# already have the repo? check out this version:
$ git checkout v0.135.2

{
  "stop_at_soft_wraps": true
}

• ›Adds editor::ToggleHunkDiff (bound to cmd-') and editor::ExpandAllHunkDiffs (bound to cmd-") actions for browsing git hunk diffs inline.
• ›Adds stop_at_soft_wraps setting for Editor::move_to_{beginning|end}_of_line; when true, Home/End navigate to the logical line boundary instead of the nearest soft-wrap point.
• ›Adds editor: convert to opposite case command.
• ›Adds gutter buttons to run tests directly from Rust files.
• ›Adds shift-k in Vim mode to show the hover tooltip.

Zed v0.134.2 adds Codeberg blame avatars, cursor scrollbar markers, ESLint Vue.js support, and a new Trash action for the project panel.

$ git clone --branch v0.134.2 https://github.com/zed-industries/zed.git
# already have the repo? check out this version:
$ git checkout v0.134.2

{
  "context": "ProjectPanel",
  "bindings": {
    "backspace": "project_panel::Delete",
    "delete": "project_panel::Delete"
  }
}

{
  "lsp": {
    "eslint": {
      "settings": {
        "rulesCustomizations": [
          { "rule": "*", "severity": "warn" }
        ]
      }
    }
  }
}

• ›Adds project_panel::Trash action; backspace and delete in the project panel now send files to the system trash instead of permanently deleting them — restore old behavior by binding both keys to project_panel::Delete in keybindings.
• ›Adds scrollbar.cursors setting to toggle cursor position markers in the scrollbar.
• ›Adds enable_preview_from_code_navigation setting to control replacing the current preview tab when using code navigation.
• ›Adds support for configuring ESLint's rulesCustomizations via lsp.eslint.settings.rulesCustomizations in the LSP settings (e.g. {"lsp": {"eslint": {"settings": {"rulesCustomizations": [{"rule": "*", "severity": "warn"}]}}}}).
• ›Adds ESLint language server support for .vue files.

Zed v0.133.5 adds expandable multi-buffer excerpts, per-language server config, and richer git blame tooltips.

$ git clone --branch v0.133.5 https://github.com/zed-industries/zed.git
# already have the repo? check out this version:
$ git checkout v0.133.5

// In your keybindings array (keymap.json)
{
  "context": "Editor && mode == full",
  "bindings": {
    "shift-enter": ["editor::ExpandExcerpts", { "lines": 5 }]
  }
}

{
  "git": {
    "inline_blame": {
      "min_column": 80
    }
  }
}

$ zed --stable /path/to/project

• ›Adds editor::ExpandExcerpts action (default binding shift-enter) that expands the multi-buffer excerpt under the cursor by 3 lines; rebind with { "lines": N } to control the count.
• ›Adds git.inline_blame.min_column setting to control the minimum column at which inline git blame information is displayed (e.g., {"git": {"inline_blame": {"min_column": 80}}}).
• ›Adds project_panel.button setting to show or hide the Project Panel button in the status bar (e.g., {"project_panel": {"button": false}}).
• ›Adds drop_target_size setting (fractional percent, e.g., 0.5) to control drop-target hit area size.
• ›Adds language_servers setting inside per-language configuration to customize which language server(s) run for a given language.

Haystack v2.1.0-rc2 adds 8 evaluator components, sparse embedding support, per-component output inspection, and new HuggingFace API generators.

$ git clone --branch v2.1.0-rc2 https://github.com/deepset-ai/haystack.git
# already have the repo? check out this version:
$ git checkout v2.1.0-rc2

pipe.run(data, include_outputs_from=["prompt_builder", "llm", "retriever"])

from haystack.components.evaluators import DocumentMAPEvaluator

evaluator = DocumentMAPEvaluator()
result = evaluator.run(
    ground_truth_documents=[[Document(content="France")], [Document(content="9th century")]],
    retrieved_documents=[[Document(content="France")], [Document(content="9th century"), Document(content="10th century")]],
)
print(result["score"])

from haystack import Pipeline
from haystack_integrations.components.retrievers.qdrant import QdrantSparseEmbeddingRetriever
from haystack_integrations.components.embedders.fastembed import FastembedSparseTextEmbedder

sparse_text_embedder = FastembedSparseTextEmbedder(model="prithvida/Splade_PP_en_v1")
sparse_retriever = QdrantSparseEmbeddingRetriever(document_store=document_store)

query_pipeline = Pipeline()
query_pipeline.add_component("sparse_text_embedder", sparse_text_embedder)
query_pipeline.add_component("sparse_retriever", sparse_retriever)
query_pipeline.connect("sparse_text_embedder.sparse_embedding", "sparse_retriever.query_sparse_embedding")

• ›Adds include_outputs_from parameter to pipeline.run() accepting a set of component names whose intermediate outputs are returned in the final pipeline output dictionary.
• ›Adds trust_remote_code parameter to SentenceTransformersDocumentEmbedder and SentenceTransformersTextEmbedder for allowing custom models and scripts.
• ›Adds truncate and normalize parameters to HuggingFaceTEITextEmbedder for truncation and normalization of embeddings.
• ›Adds streaming_callback parameter to HuggingFaceLocalGenerator for handling streaming responses.
• ›Adds dimensions parameter to AzureOpenAITextEmbedder and AzureOpenAIDocumentEmbedder to support new embedding models including text-embedding-3-small and text-embedding-3-large.

Haystack v2.1.0-rc1 adds diversity ranking, six new evaluators, four unified HuggingFace API components, sparse embeddings, and a zero-shot text router.

$ git clone --branch v2.1.0-rc1 https://github.com/deepset-ai/haystack.git
# already have the repo? check out this version:
$ git checkout v2.1.0-rc1

from haystack.components.routers import FileTypeRouter
from pathlib import Path

router = FileTypeRouter(mime_types=[r"text/.*", r"application/(pdf|json)"])
result = router.run(sources=[Path("report.pdf"), Path("notes.txt"), Path("data.json"), Path("image.png")])
for mime_type, files in result.items():
    print(f"MIME Type: {mime_type}, Files: {[str(f) for f in files]}")

from haystack.components.evaluators import FaithfulnessEvaluator

evaluator = FaithfulnessEvaluator()
result = evaluator.run(
    questions=["What is the capital of France?"],
    contexts=[["Paris is the capital and largest city of France."]],
    predicted_answers=["The capital of France is Paris."]
)
print(result["score"])  # float between 0 and 1

from haystack.components.generators import HuggingFaceLocalGenerator

def my_callback(token):
    print(token, end="", flush=True)

generator = HuggingFaceLocalGenerator(
    model="google/flan-t5-large",
    streaming_callback=my_callback
)
generator.warm_up()
generator.run(prompt="Summarize the OWASP Top 10 in three sentences.")

• ›Adds truncate and normalize parameters to HuggingFaceTEITextEmbedder for controlling truncation and normalization of embeddings.
• ›Adds trust_remote_code parameter to SentenceTransformersDocumentEmbedder and SentenceTransformersTextEmbedder to allow custom models and scripts.
• ›Adds streaming_callback parameter to HuggingFaceLocalGenerator to handle streaming responses.
• ›Adds dimensions parameter to AzureOpenAITextEmbedder and AzureOpenAIDocumentEmbedder to support newer embedding models such as text-embedding-3-small and text-embedding-3-large.
• ›Adds try_others parameter to HTMLToDocument (default true) to fall back through multiple extractors in priority order on failure.

langchain-openai 0.1.8 adds token usage tracking on AIMessage and GPT-4o pricing/context metadata.

$ git clone --branch langchain-openai==0.1.8 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain-openai==0.1.8

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o")
response = llm.invoke("Summarize zero-trust architecture in one paragraph.")
print(response.usage_metadata)  # {'input_tokens': ..., 'output_tokens': ..., 'total_tokens': ...}

• ›Adds a usage_metadata token usage attribute to AIMessage, exposing prompt, completion, and total token counts directly on the message object.
• ›Adds pricing and max context window metadata for GPT-4o to the model registry.
• ›Enables reading of stream_options from the OpenAI streaming response, making per-chunk usage data accessible.

langchain-core 0.2.2 adds a token usage attribute to AIMessage and exposes RunnableWithFallbacks internals.

$ git clone --branch langchain-core==0.2.2 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain-core==0.2.2

message = model.invoke('Summarize this document')
print(message.usage_metadata)

• ›Adds usage_metadata token usage attribute to AIMessage, giving callers direct access to token counts from model responses.
• ›Exposes attributes of the inner runnable on RunnableWithFallbacks, allowing access to wrapped runnable properties without unwrapping.

langchain-anthropic 0.1.14rc2 adds token usage attribute to AIMessage and allows tool call mutation.

$ git clone --branch langchain-anthropic==0.1.14rc2 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain-anthropic==0.1.14rc2

• ›Adds usage_metadata token usage attribute to AIMessage, exposing prompt and completion token counts directly on the message object.
• ›Allows mutation of tool call objects on AIMessage, enabling post-hoc modification of tool call data in agent pipelines.

langchain-community 0.2.1 adds CloudBlobLoader, Cassandra ByteStore, Scrapfly/AskNews/Aerospike integrations, and async Cassandra chat history

$ git clone --branch langchain-community==0.2.1 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain-community==0.2.1

from langchain_community.chat_message_histories import CassandraChatMessageHistory
import asyncio

history = CassandraChatMessageHistory(session_id="user-42", session=cassandra_session, keyspace="langchain")
await history.aadd_messages(messages)
msgs = await history.aget_messages()

from langchain_community.retrievers import AskNewsRetriever

retriever = AskNewsRetriever(k=5)
docs = retriever.invoke("latest vulnerabilities in industrial control systems")

• ›Adds CloudBlobLoader for loading data from cloud buckets.
• ›Adds CassandraByteStore as a new ByteStore backend.
• ›Adds async methods to CassandraChatMessageHistory.
• ›Adds ScrapflyLoader community integration for web scraping.
• ›Adds AskNewsRetriever and AskNews tool integrations.

LangChain 0.2.1 adds OpenAI Assistants v2 API support and a new revision_example prompt template.

$ git clone --branch langchain==0.2.1 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout langchain==0.2.1

• ›Adds revision_example prompt template to LangChain's prompt template library.
• ›Adds OpenAI Assistants v2 API support via OpenAIAssistantRunnable, with OpenAIAssistantV2Runnable moved to the community package.
• ›MultiQueryRetriever now defaults to returning a Runnable instead of the previous default.

Letta 0.3.15 adds Llama 3 support and expanded tool functionality for the Python client.

$ git clone --branch 0.3.15 https://github.com/letta-ai/letta.git
# already have the repo? check out this version:
$ git checkout 0.3.15

• ›Adds Llama 3 model support for local LLM inference.
• ›Expands tool functionality available in the Python client.

Letta 0.3.13 ships an alpha MemGPT Dev Portal accessible at memgpt.localhost (Docker) or localhost:8283 (CLI).

$ git clone --branch 0.3.13 https://github.com/letta-ai/letta.git
# already have the repo? check out this version:
$ git checkout 0.3.13

$ memgpt quickstart --backend openai && memgpt server

$ git clone [email protected]:cpacker/MemGPT.git && cd MemGPT && docker compose up

• ›Adds an alpha MemGPT Dev Portal accessible at memgpt.localhost when running with Docker Compose, or localhost:8283 when running with memgpt server.
• ›Adds a memgpt server CLI command to launch the backend service and serve the dev portal locally.
• ›Adds memgpt quickstart [--backend openai] as an initialisation path before running the server.

LlamaIndex v0.10.41 adds Mistral code and fill-in-middle models, embedding propagation to property graph retrievers, and streaming completion events.

$ git clone --branch v0.10.41 https://github.com/run-llama/llama_index.git
# already have the repo? check out this version:
$ git checkout v0.10.41

• ›Propagates embeddings from the index to the property graph retriever, enabling embedding-based graph retrieval without manual re-configuration.
• ›Adds the Mistral code model (llama-index-llms-mistralai 0.1.15) as a supported LLM integration.
• ›Adds fill-in-the-middle endpoint support for Mistral Codestral in llama-index-llms-mistralai.
• ›Adds missing instrumentation events for completion streaming in llama-index-core, enabling complete observability over streamed LLM responses.
• ›Uses the model kwarg for model name in the Gemini LLM integration (llama-index-llms-gemini 0.1.10).

LlamaIndex v0.10.40 adds PropertyGraphIndex, Neo4jPGStore, SecGPT integration, OCI Generative AI, and Hologres vector store support.

$ git clone --branch v0.10.40 https://github.com/run-llama/llama_index.git
# already have the repo? check out this version:
$ git checkout v0.10.40

• ›Adds PropertyGraphIndex to llama-index-core along with supporting abstractions for property graph-based indexing workflows.
• ›Adds Neo4jPGStore to llama-index-graph-stores-neo4j for property graph support backed by Neo4j.
• ›Adds llama-index-packs-secgpt [0.1.0] integrating SecGPT, a cybersecurity-focused LLM pack, into LlamaIndex.
• ›Adds llama-index-llms-oci-genai [0.1.0] and llama-index-embeddings-oci-genai [0.1.0] bringing Oracle Cloud Infrastructure (OCI) Generative AI support for both LLMs and embeddings.
• ›Adds llama-index-vector-stores-hologres [0.1.0] integrating the Hologres vector database as a new vector store backend.

LlamaIndex v0.10.35 adds NVIDIA NIM embeddings, LLM, and rerank support, plus new CRITIC/reflection agents and Vespa/Vertex AI vector stores.

$ git clone --branch v0.10.35 https://github.com/run-llama/llama_index.git
# already have the repo? check out this version:
$ git checkout v0.10.35

• ›Adds llama-index-llms-nvidia [0.1.0] with NVIDIA NIM LLM support via the new llama_index.llms.nvidia integration.
• ›Adds llama-index-embeddings-nvidia [0.1.0] with NVIDIA NIM embeddings support via the new llama_index.embeddings.nvidia integration.
• ›Adds llama-index-postprocessor-nvidia-rerank [0.1.0] with NVIDIA NIM rerank support.
• ›Adds llama-index-vector-stores-vespa [0.1.0] introducing a VectorStore integration for Vespa.
• ›Adds llama-index-vector-stores-vertexaivectorsearch [0.1.0] introducing Vertex AI Vector Search as a vector store backend.

LlamaIndex v0.10.34 adds structured planning agent, chat summary memory, hybrid retrieval, YouTube reader, and streaming expansions across multiple LLM integrations.

$ git clone --branch v0.10.34 https://github.com/run-llama/llama_index.git
# already have the repo? check out this version:
$ git checkout v0.10.34

• ›Adds ChatSummaryMemoryBuffer to llama-index-core for memory-efficient chat history management via summarization.
• ›Adds a structured planning agent to llama-index-core with an updated base class for planner agents.
• ›Updates HitRate and MRR retrieval metrics in llama-index-core to support Evaluation@K documents retrieved, and introduces RR (Reciprocal Rank) as a separate standalone metric.
• ›Adds hybrid retrieval mode to MilvusVectorStore in llama-index-vector-stores-milvus.
• ›Adds llama-index-vector-stores-firestore [0.1.0] — a new Firestore Vector Store integration.

Semantic Kernel 1.13.0 adds Azure Cosmos DB NoSQL and Azure SQL/SQL Server vector memory connectors, logprobs support, and streaming tool call diagnostics.

$ git clone --branch dotnet-1.13.0 https://github.com/microsoft/semantic-kernel.git
# already have the repo? check out this version:
$ git checkout dotnet-1.13.0

• ›Adds logprobs property to OpenAIPromptExecutionSettings for retrieving log-probability output from OpenAI models.
• ›New memory connector for Azure Cosmos DB for NoSQL (#6148).
• ›New memory store implementation using Azure SQL / SQL Server with vector search support.
• ›Enables CreateFromType / CreateFromObject to work with closed generic types.
• ›Includes streaming tool call information in model diagnostics.

Semantic Kernel Python 1.0.0rc1 introduces pre- and post-function filters for hooking into function execution.

$ git clone --branch python-1.0.0rc1 https://github.com/microsoft/semantic-kernel.git
# already have the repo? check out this version:
$ git checkout python-1.0.0rc1

• ›Adds a filters system that lets developers hook into pre- and post-function execution to inject logging, validation, authentication, or other custom behaviors around kernel function calls.

Semantic Kernel .NET 1.12.0 adds MistralAI connector, OTel model diagnostics for streaming, and MistralClient activity tracing.

$ git clone --branch dotnet-1.12.0 https://github.com/microsoft/semantic-kernel.git
# already have the repo? check out this version:
$ git checkout dotnet-1.12.0

public class MyPlugin
{
    [KernelFunction]
    internal string GetSecret(string key) => _vault.Get(key);
}

• ›Adds AllowDangerouslySetContent (renamed surface) for controlling content safety boundaries in kernel operations.
• ›Adds OTel model diagnostics support for streaming APIs, extending observability to streaming call paths.
• ›Adds model diagnostics to non-streaming APIs for OpenTelemetry-based tracing of LLM calls.
• ›Adds MistralAI connector, enabling Semantic Kernel to target MistralAI models as a first-class backend.
• ›Adds OpenTelemetry activities to MistralClient for distributed tracing of Mistral calls.

Semantic Kernel Python 0.9.9b1 adds Pydantic Settings for secrets, a new kernel function decorator, and enhanced OpenAPI plugin parameter handling.

$ git clone --branch python-0.9.9b1 https://github.com/microsoft/semantic-kernel.git
# already have the repo? check out this version:
$ git checkout python-0.9.9b1

• ›Introduces Pydantic Settings support for managing secrets, keys, and configurations — reads environment variables or falls back to a .env file path; key, deployment_name, endpoint, and api_version remain available as optional parameters on Text, Chat, and Embedding classes.
• ›Adds a new @kernel_function decorator for defining kernel functions, including lambda function support.
• ›Adds @experimental class and function decorator to mark APIs as experimental.
• ›Adds function_name and plugin_name properties to function call and function call result objects.
• ›Allows the OpenAPI runner to accept a custom HTTP client.

Semantic Kernel 1.11.1 adds a Sessions Code Interpreter Core Plugin and a dimensions property on the OpenAI embedding service constructor.

$ git clone --branch dotnet-1.11.1 https://github.com/microsoft/semantic-kernel.git
# already have the repo? check out this version:
$ git checkout dotnet-1.11.1

var embeddingService = new OpenAITextEmbeddingGenerationService(
    modelId: "text-embedding-3-small",
    apiKey: "<your-api-key>",
    dimensions: 256
);

• ›Adds dimensions property to the OpenAI embedding service constructor, allowing callers to specify embedding vector size at instantiation.
• ›Adds a Sessions (Code Interpreter) Core Plugin and accompanying demo project for executing code in sandboxed Azure Container Apps sessions.
• ›Improves the Azure Cosmos DB for MongoDB connector with additional capability enhancements.

Semantic Kernel Python adds FunctionCallBehavior API, ACA Code Interpreter plugin, and retires three legacy planners.

$ git clone --branch python-0.9.8b1 https://github.com/microsoft/semantic-kernel.git
# already have the repo? check out this version:
$ git checkout python-0.9.8b1

from semantic_kernel.connectors.ai.function_call_behavior import FunctionCallBehavior

filter = {"excluded_plugins": ["ChatBot"]}
req_settings.function_call_behavior = FunctionCallBehavior.EnableFunctions(auto_invoke=True, filters=filter)

from semantic_kernel.connectors.ai.function_call_behavior import FunctionCallBehavior

req_settings.function_call_behavior = FunctionCallBehavior.AutoInvokeKernelFunctions()

• ›Adds FunctionCallBehavior class to semantic_kernel.connectors.ai.function_call_behavior with methods FunctionCallBehavior.EnableFunctions(auto_invoke=True, filters=filter) and FunctionCallBehavior.AutoInvokeKernelFunctions(), settable via req_settings.function_call_behavior, replacing the need to manually specify tool_choice and tools in prompt execution settings.
• ›Adds filters parameter to FunctionCallBehavior.EnableFunctions() supporting dict keys such as excluded_plugins to control which plugins are exposed to the model.
• ›Adds the ACA Python Sessions (Code Interpreter) Core Plugin, enabling sandboxed remote code execution via Azure Container Apps sessions.

Semantic Kernel .NET 1.11.0 adds Prompty support, request/response metadata on REST calls, dimensions control for OpenAI embeddings, and a netstandard2.0 ONNX connector.

$ git clone --branch dotnet-1.11.0 https://github.com/microsoft/semantic-kernel.git
# already have the repo? check out this version:
$ git checkout dotnet-1.11.0

builder.AddOpenAITextEmbeddingGeneration(
    modelId: "text-embedding-3-small",
    apiKey: Environment.GetEnvironmentVariable("OPENAI_API_KEY"),
    dimensions: 512);

• ›Adds RequestUri and Payload properties to RestApiOperationResponse, exposing the outbound request URI and body for inspection after REST API plugin calls.
• ›Adds dimensions property to OpenAI embedding generation services, allowing callers to control output embedding size.
• ›Adds netstandard2.0 build target to Microsoft.SemanticKernel.Connectors.Onnx, enabling use in .NET Standard 2.0 projects.
• ›Merges Prompty feature branch to main, adding native support for the Prompty format in Semantic Kernel.
• ›Adds agent logging (Agent Logging) for structured observability of agent execution.

Semantic Kernel Python adds FunctionCallContent/FunctionResultContent types, embedding dimensions support, and drops Python <3.10

$ git clone --branch python-0.9.7b1 https://github.com/microsoft/semantic-kernel.git
# already have the repo? check out this version:
$ git checkout python-0.9.7b1

• ›Introduces FunctionCallContent and FunctionResultContent content types for structured function-calling support inside ChatMessageContent, replacing flat message representations.
• ›Extends ChatMessageContent to hold one or more content items simultaneously, enabling mixed TextContent and function-call content in a single message.
• ›Refactors OpenAI classes to parse and emit FunctionCallContent and related new content types directly, removing the now-redundant OpenAIChatMessageContent and AzureChatMessageContent classes.
• ›Adds caller identity as a user-agent header on HTTP requests to Astra DB's Data API.
• ›Reorganizes samples into samples/getting_started (notebooks), samples/concepts (kernel syntax examples by topic), and a new root-level prompt_template_samples folder.

KoboldCpp v1.65 adds a standalone Stable Diffusion UI, CUDA 12 binaries, a new bypass_eos API field, and replaces three deprecated flags with granular replacements.

$ git clone --branch v1.65 https://github.com/LostRuins/koboldcpp.git
# already have the repo? check out this version:
$ git checkout v1.65

$ koboldcpp.exe --model mymodel.gguf --sdmodel sd_v15.safetensors --sdthreads 4 --sdquant --sdclamped

$ koboldcpp.exe --model mymodel.gguf --hordekey YOUR_API_KEY --hordeworkername MyWorker --hordemodelname mistral-7b --hordemaxctx 4096 --hordegenlen 256

$ curl -X POST http://localhost:5001/api/v1/generate -H 'Content-Type: application/json' -d '{"prompt": "Once upon a time", "max_length": 200, "bypass_eos": true}'

• ›Adds --sdmodel, --sdthreads, --sdquant, and --sdclamped flags to replace the deprecated --sdconfig, enabling per-parameter validation and easier extension of Stable Diffusion options.
• ›Adds --hordemodelname, --hordeworkername, --hordekey, --hordemaxctx, and --hordegenlen flags to replace the deprecated --hordeconfig, giving each AI Horde setting its own named flag.
• ›Adds bypass_eos field to the API, allowing EOS tokens to be skipped during generation while still permitting them to appear in output.
• ›Adds official CUDA 12 binary (koboldcpp_cuda12.exe / koboldcpp_cu12.exe) for newer NVIDIA GPUs, providing increased inference speeds at the cost of a larger download.
• ›Adds a standalone browser-based image generation UI (StableUI port, A1111-compatible) accessible at http://localhost:5001/sdui/ when a Stable Diffusion model is loaded.

KoboldCpp v1.64.1 adds --flashattention, dynamic banned_tokens, render_special, and removes --bantokens.

$ git clone --branch v1.64.1 https://github.com/LostRuins/koboldcpp.git
# already have the repo? check out this version:
$ git checkout v1.64.1

$ koboldcpp.exe --flashattention mymodel.gguf

$ curl -X POST http://localhost:5001/api/v1/generate -H 'Content-Type: application/json' -d '{"prompt": "Once upon a time", "banned_tokens": [1234, 5678]}'

$ curl -X POST http://localhost:5001/api/v1/generate -H 'Content-Type: application/json' -d '{"prompt": "Hello", "render_special": true}'

• ›Adds --flashattention experimental flag to enable Flash Attention for compatible models.
• ›Adds banned_tokens field to the generate API, allowing per-generation token banning dynamically (replaces the removed --bantokens flag).
• ›Adds render_special to the generate API, enabling rendering of special tokens such as <|start_header_id|> or <|eot_id|>.
• ›Adds trim_stop support in SSE streaming modes so stop sequences are hidden during streaming when enabled; Chat Completions endpoint automatically applies trim_stop to instruct tag format for better compatibility with third-party clients like LibreChat.
• ›Adds finish_reason communication in both sync and SSE streamed mode responses when generation stops due to EOS/EOT tokens.

LocalAI v2.15.0 adds Vision API in Chat WebUI, single binary releases, --debug CPU/GPU info, and trust_remote_code UI flag.

$ git clone --branch v2.15.0 https://github.com/mudler/LocalAI.git
# already have the repo? check out this version:
$ git checkout v2.15.0

• ›Adds --debug flag output to display CPU/GPU information at startup.
• ›Exposes trust_remote_code as a configurable flag in the WebUI for model loading.
• ›Adds llama.cpp backend autoloading without requiring explicit backend specification, plus llama.cpp variant support.
• ›Integrates Vision API into the Chat WebUI, enabling image processing model testing directly in the browser.
• ›Adds system prompt configuration in the WebUI chat interface.

LocalAI v2.14.0 adds OpenVINO acceleration, user-defined inference devices, model deletion, llama3 AIO, and a new WebUI with chat/TTS/image-gen pages.

$ git clone --branch v2.14.0 https://github.com/mudler/LocalAI.git
# already have the repo? check out this version:
$ git checkout v2.14.0

• ›Adds user-defined inference device selection for CUDA and OpenVINO backends, letting practitioners pin workloads to specific hardware.
• ›Adds OpenVINO acceleration for embeddings in the transformer backend, enabling fast inference on Intel CPUs and GPUs.
• ›Adds model deletion support to the gallery UI, allowing installed models to be removed directly from the interface.
• ›Adds gallery job status display during navigation so model install progress is visible while browsing.
• ›Adds Chat, TTS, and image-generation pages to the WebUI for quick interactive debugging and model assessment.

Ollama v0.1.39 adds Llama 3 Safetensors import, flash-attention flag, new models, and OLLAMA_NOHISTORY support

$ git clone --branch v0.1.39 https://github.com/ollama/ollama.git
# already have the repo? check out this version:
$ git checkout v0.1.39

$ ollama create --quantize q4_0 -f Modelfile my-llama3

• ›Adds OLLAMA_NOHISTORY=1 environment variable to disable shell history when using ollama run.
• ›Adds experimental OLLAMA_FLASH_ATTENTION=1 environment variable flag for ollama serve to improve token generation speed on Apple Silicon Macs and NVIDIA GPUs.
• ›Adds --quantize flag to ollama create (e.g. --quantize q4_0) enabling import and quantization of Llama 3 and its finetunes from Safetensors format.
• ›ollama create now supports creating models from I-Quant GGUF files.
• ›Adds Ctrl+W keyboard shortcut to ollama run.

Ollama v0.1.38 adds ollama ps to inspect loaded models and /clear to reset chat session history.

$ git clone --branch v0.1.38 https://github.com/ollama/ollama.git
# already have the repo? check out this version:
$ git checkout v0.1.38

$ ollama ps

$ ollama run llama3
>>> /clear

• ›Adds ollama ps command to display currently loaded models, their memory footprint (in GB/MB), and processor usage (GPU vs CPU percentages).
• ›Adds /clear command inside ollama run sessions to reset chat history without ending the session.
• ›Adds Falcon 2, an 11B-parameter causal decoder-only model trained on 5T tokens, available via ollama pull falcon2.
• ›Adds Yi 1.5 (Apache 2.0) in 6B (yi:6b), 9B (yi:9b), and 34B (yi:34b) sizes.

Ollama v0.1.35 adds on-the-fly quantization via --quantize flag and a new done_reason field in API responses.

$ git clone --branch v0.1.35 https://github.com/ollama/ollama.git
# already have the repo? check out this version:
$ git checkout v0.1.35

$ ollama create -f Modelfile --quantize q4_0 mymodel

• ›Adds --quantize / -q flag to ollama create to quantize float16 or float32 models (from GGUF files or library models) at import time, supporting formats such as q4_0.
• ›Adds done_reason field to API responses describing why generation stopped.
• ›Adds Llama 3 ChatQA model (llama3-chatqa), an NVIDIA model optimised for conversational QA and retrieval-augmented generation.

Ollama v0.1.34 adds five new models including multimodal Llava Llama 3 and Llava Phi 3.

$ git clone --branch v0.1.34 https://github.com/ollama/ollama.git
# already have the repo? check out this version:
$ git checkout v0.1.34

$ ollama run llava-llama3

$ ollama run starcoder2:15b-instruct

• ›Adds Llava Llama 3 (llava-llama3), a high-performing multimodal LLaVA model fine-tuned from Llama 3 Instruct.
• ›Adds Llava Phi 3 (llava-phi3), a small multimodal LLaVA model fine-tuned from Phi 3.
• ›Adds StarCoder2 15B Instruct (starcoder2:15b-instruct), an instruction-tuned variant of the StarCoder2 code model.
• ›Adds CodeGemma 1.1 (codegemma), an updated release of Google's CodeGemma model.
• ›Adds StableLM2 12B (stablelm2:12b), a new 12B parameter version of Stability AI's StableLM 2 model.

Phoenix Evals 0.10.0 adds Mistral model support for LLM-based evaluations.

$ git clone --branch arize-phoenix-evals-v0.10.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-evals-v0.10.0

• ›Supports Mistral as an LLM backend for running evaluations via the arize-phoenix-evals library.
• ›Docker image now runs as root by default, with additional tags available for nonroot and debug image variants.

Phoenix Docker image now runs as root by default, with new nonroot and debug image tags available.

$ git clone --branch arize-phoenix-v4.2.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v4.2.0

• ›Docker image now runs as root by default, with dedicated image tags for nonroot and debug variants.

Arize Phoenix 4.1.0 adds an ASGI root path parameter to the Phoenix server.

$ git clone --branch arize-phoenix-v4.1.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v4.1.0

• ›Adds ASGI root path parameter support to the Phoenix server, enabling deployment behind a reverse proxy or sub-path prefix.

Phoenix Evals 0.9.0 adds default_headers support for Azure OpenAI models.

$ git clone --branch arize-phoenix-evals-v0.9.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-evals-v0.9.0

• ›Adds default_headers parameter support to the Azure OpenAI integration, enabling custom HTTP headers on all requests.

Phoenix v4.0.0 adds gRPC ingestion, PostgreSQL support, basic auth, OpenAPI UI, new eval templates, and a log_traces method.

$ git clone --branch arize-phoenix-v4.0.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v4.0.0

import phoenix as px

px.log_traces(trace_dataset=trace_dataset)

• ›Adds log_traces method to send TraceDataset traces directly to Phoenix from Python.
• ›Adds gRPC endpoint for trace ingestion, with a Prometheus interceptor for gRPC metrics.
• ›Adds default limit to GET /v1/spans and corresponding client methods to prevent unbounded responses.
• ›Adds trace and document evaluations to GET /v1/evaluations, including span evaluations.
• ›Adds support for basic auth on the Phoenix server.

Langfuse v2.47.0 adds a JSON linter for prompt/trace inputs and client-side error notifications in the UI.

$ git clone --branch v2.47.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.47.0

• ›Adds a linter to JSON inputs in the UI, catching malformed JSON inline before submission.
• ›Adds client-side error notifications in the UI to surface runtime errors directly to the user.

Langfuse v2.46.0 adds custom API key support in the playground.

$ git clone --branch v2.46.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.46.0

• ›Enables use of custom keys in the playground for testing LLM calls with user-supplied credentials.

Langfuse v2.45.1 adds AWS Cognito as an authentication provider and introduces a score configs table.

$ git clone --branch v2.45.1 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.45.1

• ›Adds AWS Cognito as an OAuth/SSO authentication provider for self-hosted deployments.
• ›Adds a new score_configs database table (via migration) to support configurable scoring schemas.

Langfuse v2.45.0 adds dark mode, playground deep links, a dedicated scores tab, and a new authorUserId field on scores.

$ git clone --branch v2.45.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.45.0

• ›Adds authorUserId (nullable) field to scores, laying groundwork for upcoming annotation releases.
• ›Adds dark mode across the UI.
• ›Adds bidirectional deep links between the playground and both prompt management and tracing views.
• ›Moves the scores table on trace and observation previews into its own dedicated tab.

Langfuse v2.43.2 adds a data region selector and info modal for cloud deployments.

$ git clone --branch v2.43.2 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.43.2

• ›Adds a data region select UI and info modal for cloud users to view and choose their data residency region.

Langfuse v2.43.0 lets you disable model parameters in the playground and evals.

$ git clone --branch v2.43.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.43.0

• ›Adds the ability to disable model parameters in the playground and evals configurations.

Langfuse v2.41.0 adds a prompt-version metrics comparison table and API-editable dataset item status.

$ git clone --branch v2.41.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.41.0

• ›Adds status field on dataset items as writable via the API, enabling programmatic updates to item status.
• ›Adds a [promptName]/metrics table view to compare metrics across prompt versions in the UI.

Langfuse v2.40.0 adds GPT-4o support to the playground and evaluation pipelines.

$ git clone --branch v2.40.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.40.0

• ›Adds GPT-4o models as selectable options in the Langfuse playground.
• ›Adds GPT-4o as an available model for running LLM-based evaluations.

Langfuse v2.39.0 adds GPT-4o pricing/tokenization, a new default datetime offset env var, and score project scoping.

$ git clone --branch v2.39.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.39.0

$ LANGFUSE_DEFAULT_TABLE_DATETIME_OFFSET=-7d

• ›Adds LANGFUSE_DEFAULT_TABLE_DATETIME_OFFSET environment variable to control the default datetime offset applied to table views.
• ›Adds pricing information and tokenization support for GPT-4o models.
• ›Adds nullable projectId to scores, enabling unordered ingestion of scores (foreign key constraint to be dropped in the next version).

Adds LANGFUSE_DISABLE_EXPENSIVE_POSTGRES_QUERIES env var to skip costly DB queries on large deployments.

$ git clone --branch v2.38.2 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.38.2

$ LANGFUSE_DISABLE_EXPENSIVE_POSTGRES_QUERIES=true

• ›Adds optional LANGFUSE_DISABLE_EXPENSIVE_POSTGRES_QUERIES environment variable to disable expensive PostgreSQL queries, useful for large-scale deployments where certain queries cause performance issues.

Langfuse v2.38.1 adds PostHog integration for product analytics.

$ git clone --branch v2.38.1 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.38.1

• ›Adds PostHog integration for product analytics instrumentation.

Langfuse v2.38.0 adds per-prompt-version generations table and observationCount field to the traces table.

$ git clone --branch v2.38.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.38.0

• ›Adds observationCount field to the traces table, surfacing generation counts directly in the traces view.
• ›Adds a generations table scoped per prompt version, with an accordion to show or hide generations inline on prompt version pages.

Langfuse v2.37.3 adds post-login redirect to the originally requested path.

$ git clone --branch v2.37.3 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v2.37.3

• ›Redirects users to the original targetPath after login, preserving deep links when authentication is required.

LanceDB v0.5.1 adds IVF_HNSW_PQ index support, a JS embedding registry, and a Java table-names API.

$ git clone --branch v0.5.1 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.5.1

• ›Adds IVF_HNSW_PQ index type, combining IVF, HNSW, and product quantization for approximate nearest-neighbor search.
• ›Adds tableNames Java API for listing tables in a LanceDB connection from the Java client.
• ›Introduces a JavaScript embedding registry, enabling registration and lookup of embedding functions in the Node.js SDK.
• ›Adds Arrow version compatibility across the Node.js SDK, supporting multiple Arrow versions interoperably.

LanceDB python-v0.7.0 adds IVF_HNSW_SQ index support, an async optimize function, and Ollama embeddings integration.

$ git clone --branch python-v0.7.0 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout python-v0.7.0

await table.optimize()

• ›Adds optimize function to async Python and Node.js APIs for index and storage optimization.
• ›Supports new IVF_HNSW_SQ index type, combining IVF, HNSW, and scalar quantization for approximate nearest-neighbor search.
• ›Adds Ollama embeddings function, enabling local LLM-backed embedding generation within LanceDB pipelines.
• ›Upgrades underlying Lance to version 0.11.0, bringing its new storage and indexing capabilities.

LanceDB v0.5.0 adds IVF_HNSW_SQ index support, an optimize function for Node.js and async Python, and Ollama embeddings integration.

$ git clone --branch v0.5.0 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.5.0

• ›Adds optimize function to the Node.js and async Python APIs for index and storage optimization.
• ›Adds support for the IVF_HNSW_SQ index type, combining IVF, HNSW, and scalar quantization for ANN search.
• ›Adds Ollama embeddings function, enabling local LLM-backed embeddings via Ollama.

LanceDB v0.4.19 adds Polars DataFrame interop and an embedding registry to the Rust SDK.

$ git clone --branch v0.4.19 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.4.19

• ›Adds an embedding registry to the Rust SDK, enabling model registration and lookup for vector embedding workflows.
• ›Implements Polars DataFrame converters (to and from) in the Rust SDK via C FFI, enabling direct interop between LanceDB tables and Polars DataFrames in Rust.

Milvus v2.3.15 adds channel checkpoint info in flush responses and a config to validate IDs on autoID insert.

$ git clone --branch v2.3.15 https://github.com/milvus-io/milvus.git
# already have the repo? check out this version:
$ git checkout v2.3.15

• ›Adds a config option to check whether an ID is provided during data insertion when autoID is enabled, optimizing data migration workflows with Milvus-CDC.
• ›Returns channel checkpoint info in flush responses, giving callers visibility into replication progress at flush time.

Milvus 2.4.1 adds Float16/BFloat16 bulk insert, sparse vector iterator search, and a declarative resource group API.

$ git clone --branch v2.4.1 https://github.com/milvus-io/milvus.git
# already have the repo? check out this version:
$ git checkout v2.4.1

• ›Adds a declarative resource group API for programmatic resource management.
• ›Adds a configuration option to control the maximum amount of data that can be inserted in a single request.
• ›Adds a configuration option to control whether to enforce activation of the partitionKey feature.
• ›Adds Float16 and BFloat16 vector data type support in bulk insert.
• ›Adds db label to metrics for delete and bulk insert operations, enabling per-database observability.