- Add
k:
parameter to allask()
vector search methods - Bump Faraday to 2.x
- 🗣️ LLMs
- Introducing Anthropic support
- Support for OpenAI functions
- Streaming vectorsearch ask() responses
- Langchain::Chunker::RecursiveText
- Fixes
- 🗣️ LLMs
- Introducing Llama.cpp support
- Langchain::OutputParsers::OutputFixingParser to wrap a Langchain::OutputParser and handle invalid response
- Fix
Langchain::Vectorsearch::Qdrant#add_texts()
- Introduce
ConversationMemory
- Allow loading multiple files from a directory
- Add
get_default_schema()
,create_default_schema()
,destroy_default_schema()
missing methods toLangchain::Vectorsearch::*
classes
- Add #destroy_default_schema() to Langchain::Vectorsearch::* classes
- Qdrant, Chroma, and Pinecone are supported by ActiveRecord hooks
- Adding support to hook vectorsearch into ActiveRecord models
- [BREAKING] Rename
ChainOfThoughtAgent
toReActAgent
- Implement A21 token validator
- Add
Langchain::OutputParsers
- Developer can modify models used when initiliazing
Langchain::LLM::*
clients - Improvements to the
SQLQueryAgent
and the database tool
- If used with OpenAI, Langchain::Conversation responses can now be streamed.
- Improved logging
- Langchain::Tool::SerpApi has been renamed to Langchain::Tool::GoogleSearch
- JSON prompt templates have been converted to YAML
- Langchain::Chunker::Text is introduced to provide simple text chunking functionality
- Misc fixes and improvements
- [BREAKING] Rename
Langchain::Chat
toLangchain::Conversation
- 🛠️ Tools
- Introducing
Langchain::Tool::Weather
, a tool that calls Open Weather API to retrieve the current weather
- Introducing
- 🔍 Vectorsearch
- Introducing support for HNSWlib
- Improved and new
Langchain::Chat
interface that persists chat history in memory
- 🗣️ LLMs
- Chat message history support for Langchain::LLM::GooglePalm and Langchain::LLM::OpenAI
- 🗣️ LLMs
- Auto-calculate the max_tokens: setting to be passed on to OpenAI
- 🛠️ Tools
- Modified Tool usage. Agents now accept Tools instances instead of Tool strings.
- [BREAKING] LLMs are now passed as objects to Vectorsearch classes instead of
llm: :name, llm_api_key:
previously - 📋 Prompts
- YAML prompt templates are now supported
- 🚚 Loaders
- Introduce
Langchain::Processors::Xlsx
to parse .xlsx files
- Introduce
- 🗣️ LLMs
- Introducing support for AI21
- Better docs generation
- Refactors
- Beautiful colored log messages
- 🛠️ Tools
- Introducing
Langchain::Tool::RubyCodeInterpreter
, a tool executes sandboxed Ruby code
- Introducing
- [BREAKING] Everything is namespaced under
Langchain::
now - Pgvector similarity search uses the cosine distance by default now
- OpenAI token length validation using tiktoken_ruby
- Drop Ruby 2.7 support. It had reached EOD.
- Bump pgvector-ruby to 0.2
- 🚚 Loaders
- Support for options and block to be passed to CSV processor
- 🔍 Vectorsearch
- Not relying on Weaviate modules anymore
- Adding missing specs for Qdrant and Milvus classes
- 🚚 Loaders
- Add Langchain::Data result object for data loaders
- 🗣️ LLMs
- Add
summarize()
method to the LLMs
- Add
- 🔍 Vectorsearch
- Pgvector support
- 🚚 Loaders
- CSV loader
- JSON loader
- JSONL loader
- 🔍 Vectorsearch
- Introduce namespace support for Pinecone
- 🚚 Loaders
- Loaders overhaul
- 🗣️ LLMs
- Introducing support for Google PaLM (Pathways Language Model)
- Bug fixes and improvements
- 🗣️ LLMs
- Introducing support for Replicate.com
- 🚚 Loaders
- Introduce
Loaders::Docx
to parse .docx files
- Introduce
-
🔍 Vectorsearch
- Introduce support for Chroma DB
-
🚚 Loaders
- Bug fix
Loaders::Text
to only parse .txt files
- Bug fix
- 🚚 Loaders
- Introduce
Loaders::Text
to parse .txt files - Introduce
Loaders::PDF
to parse .pdf files
- Introduce
- 🗣️ LLMs
- Bump
hugging-face
gem version
- Bump
- Bug fixes
- 🗣️ LLMs
- Introducing support for HuggingFace
- Dependencies are now optionally loaded and required at runtime
- Start using
standardrb
for linting - Use the Ruby logger
- 🤖 Agents
- Fix Chain of Thought prompt loader
- 🛠️ Tools
- Introducing
Tool::Wikipedia
, a tool that looks up Wikipedia entries
- Introducing
- 🤖 Agents
- Introducing
Agent::ChainOfThoughtAgent
, a semi-autonomous bot that uses Tools to retrieve additional information in order to make best-effort informed replies to user's questions.
- Introducing
- 🛠️ Tools
- Introducing
Tool::Calculator
tool that solves mathematical expressions. - Introducing
Tool::Search
tool that executes Google Searches.
- Introducing
- 📋 Prompt Templating
- Ability to create prompt templates and save them to JSON files
- Default
Prompt::FewShotPromptTemplate
- New examples added to
examples/
- Backfilling missing specs
- Initial release