Skip to content

Latest commit

 

History

History
180 lines (110 loc) · 11.3 KB

changelog.md

File metadata and controls

180 lines (110 loc) · 11.3 KB
Raw
title description
LangStream Changelog
Notes from releases

Releases{:target="_blank"}

0.6.0 - Jan 8, 2024

Improvements:

  • Runtime LangChain version updated: LangChain has been updated to version 0.0.353, this means you can use all their new features out of the box. Check them out in the LangChain release notes

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.6.0

0.5.0 - Nov 24, 2023

New features:

  • DataStax AstraDB vector support: it's now possible to leverage the AstraDB through the new JSON API to simplify the vector databases management directly in LangStream.
  • Enhanced Python custom agents: we added support for async function, message producing to a side topic and other stability improvements!

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.5.0

0.4.4 - Nov 17, 2023

New features:

  • Support for Ollama: you can now leverage OLLama to run the model locally to the LangStream executor.
  • App lifecycle improvements: we improved the status handling when the application fails the deployment. Also it's now possible to force-delete the application at any time!

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.4.4

0.4.1 - Nov 2, 2023

New features:

  • Full Support for Pravega: full-fledged support for Pravega, enabling seamless integration for the streaming functionality.

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.4.1

0.4.0 - Oct 31, 2023

Breaking change: Before upgrading the LangStream cluster to 0.4.x, you must update the Custom Resource, just run:

kubectl apply -f https://github.com/LangStream/langstream/releases/download/v0.4.4/agents.langstream.ai-v1.yml
kubectl apply -f https://github.com/LangStream/langstream/releases/download/v0.4.4/applications.langstream.ai-v1.yml

New features:

  • Full Support for Apache Pulsar: full-fledged support for Apache Pulsar, enabling seamless integration for the streaming functionality.
  • Stateful agents: agents can now provision persistent disks, offering improved data management and persistence capabilities. The webcrawler-source now utilizes a persistent disk to store status information, removing the need of a S3 bucket, without impacting the functionalities.
  • Apache Camel source agents: you can now run any Apache Camel Source as LangStream source, expanding the range of data sources that can be integrated into the system.
  • HTTP Gateway: websocket is great for long-runnnig communication but for simpler use-cases you can now interact with your applications using HTTP directly - and you don't need to redeploy your applications!
  • LangServe Integration: a new agent has been added to quickly build integrations with LangServe services (langserve-invoke), supporting the /invoke and /stream endpoints.
  • Python development enhancements: packaging the Python agent dependencies with a simple CLI command and the hot reload on source code changes, will accelerate the development cycle of custom Python agents.
  • Python service: deploy a general purpose http service using the new python-service agent. Your users will be able to access it leveraging through gateways, inheriting security and high-availability for free.

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.4.0

0.3.1 - Oct 20, 2023

Bugfixes:

  • Fix CLI docker run on MacOS and Linux

0.3.0 - Oct 19, 2023

New features:

  • Amazon Bedrock Support: LangStream now works seamlessly with Amazon Bedrock, allowing you to generate embeddings and perform text completions effortlessly.
  • Expanded Database Support: Now, you can seamlessly integrate Apache Solr and OpenSearch as Vector Databases. This means you can efficiently index and retrieve embeddings.
  • Local Hugging Face Model Execution: Running your favorite Hugging Face model on the LangStream executor is now a breeze. Just a few lines of YAML code and you're good to go.
  • Flexible SQL Queries for JDBC Databases: You can now perform both read and write operations on JDBC databases using the new query agent. This adds a new level of flexibility to your pipelines.
  • New Flow Control Agents: We've added three new agents - timer-source, trigger-event, and log-event - that help streamline data flow in your pipelines. These agents enhance the precision and control of your workflows.

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.3.0

0.2.0 - Oct 12, 2023

New features:

  • Support for FLARE pattern: FLARE is an emerging pattern, similar to RAG, that iteratively uses a prediction of the upcoming sentence to anticipate future content, which is then utilized as a query to retrieve relevant documents to regenerate the sentence if it contains low-confidence tokens. Read more in the paper.
  • New agent 'dispatch': you can implement message routing easily by defining condition on the record.
  • New agent 'http-request': run http requests to enrich your data model from any public service or call webhooks when an event happens on the pipeline.
  • New agent 'azure-blob-storage-source': read and process files from Azure Blob Storage containers.
  • New UI for apps monitoring and testing: you can now spins up a local UI webserver to monitor and test your application in a friendly way.
  • CLI support for Windows: CLI now is now installable on Windows.
  • Export your application as Mermaid graph: run langstream apps get myapp -o mermaid to generate a MermaidJS graph that shows you the entire application workflow.

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.2.0

0.1.0 - Oct 4, 2023

New features:

  • MMR based re-ranking agent: read more details in the documentation about the new re-rank agent.
  • API Reference: all the agents, resources and assets have a new API Reference page.
  • Default global values: you can now specify application default values for the global variables.
  • Embedded Vector Database HerdDB in mini-langstream: you can run vector-based applications in mini-langstream without the need for an external vector database.

Breaking changes:

  • Refactor Python agent API: the Python API is now more handy and structured. Checkout the documentation for getting started.
  • New templating system for globals and secrets: for referring to a secret you should now use use ${secrets.my-secret.key} instead of {% raw %}{{{ secrets.my-secret.key }}}{% endraw %} . Note that the old syntax will continue to work but it's highly recommended to switch to the new one.

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.1.0

0.0.22 - Sep 27, 2023

Langstream 0.0.22 new features:

  • Topics replication factor: when declaring a topic, you can now specify the replication factor with the property topics.<TOPIC>.options.replication-factor: 3.
  • Batched AI Embeddings: the compute-embeddings is now fully async and uses batches to improve throughtput.
  • mini-langstream: mini-langstream is a brand new command that creates a local LangStream cluster, based on Minikube. Try it with brew (MacOS): brew install LangStream/langstream/mini-langstream && mini-langstream start.
  • Milvus vector database support: open source project Milvus and his cloud version - Zillis - is now pluggable as vector database in your pipeline.
  • New agent ai-text-completions: for autonomus agents you can now use the OpenAI and VertexAI text models. Have you heard about gpt-3.5-turbo-instruct or text-bison? Here you go.
  • JDBC and vectors: you can now write to vector databases that are JDBC compliants. Ready to store vectors on PostGRE?
  • Multiple AI services in the same application: now you can mix different AI service types/configuration to use different models in the same application.
  • JDBC table assets: leverage assets to let LangStream handling the JDBC tables for you.
  • Embedded vector database in langstream docker run: when you test the application on Docker, you can now startup an in-memory vector databases to verify your vector-based pipelines in a couple of seconds.
  • CLI: allow updates to the default profile.
  • GRPC Python agents: the python runtime has been rewritten to be faster, more stable and composable with Java agents.

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.0.22

0.0.21 - Sep 19, 2023

Langstream 0.0.21 is a maintenance release and contains only core bugfixes about OpenAI client and Cassandra/Astra sink.

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.0.21

0.0.20 - Sep 19, 2023

Langstream 0.0.20 new features:

  • Web crawler agent has been improved and it can now works better with sitemaps, robots.txt and other nuances
  • Introduce deletion-mode for assets. Assets can be deleted when the application is deleted.
  • Core improvements like CLI chat output for streaming messages, usage of the OpenAI client

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.0.20

0.0.19 - Sep 15, 2023

Langstream 0.0.19 new features:

  • Support local run without passing secrets
  • Fix CLI dependencies download and enable Cassandra Sink e2e test
  • Introduce deletion-mode for topics. Topics can be deleted when the application is deleted.

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.0.19

0.0.18 - Sep 14, 2023

Langstream 0.0.18 new features:

  • langstream run docker command improvements
  • Support writing messages with different schemas on Kafka topics

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.0.18

0.0.17 - Sep 14, 2023

LangStream 0.0.17 new features:

  • New chat gateway which simplifies the usage of the gateways
  • Introduced support for Azure Blob Storage as code storage
  • New tool for testing the application locally without Kubernetes

Full release notes: https://github.com/LangStream/langstream/releases/tag/v0.0.17