Appearance
elsai Core — Release Notes
Installation command updated
The pip index URL for all elsai Core packages has changed. Replace any existing install commands that use https://elsai-core-package.optisolbusiness.com/root/ with the new URL:
https://core-packages.elsai.ai/root/<package-name>/All install commands in this documentation already reflect the updated URL.
elsai-model v2.1.3
Release Date: Sep 2026
New features
TypeSafeClassifier— Jev (TypeSafe System One) classifier via thetypesafeextra.classifyanswers Noul, Choice, and Score questions about a state and returns probabilities
See TypeSafe Classifier.
Installation
bash
pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-model/ "elsai-model[typesafe]==2.1.3"elsai-chat-history v1.1.2
Release Date: Sep 2026
Improvements
- AgentKit-compatible dependency ranges — pydantic
>=2.11,<3(a clean install takes 2.13.x), packaging>=23.2,<27, typing_extensions>=4.14,<5, annotated-types>=0.7,<1, and typing-inspection>=0.4.1,<1.pydantic_corefollows pydantic and is no longer pinned. - Same public API as 1.1.1 — stores, strategies, and
Messageare unchanged asyncpg,aiofiles,python-dateutil, andsixstay exact- Python 3.10–3.13
Prefer this version over 1.1.0 and 1.1.1. See Memory & Chat History and AgentKit 0.3.5.
Installation
bash
pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-chat-history/ elsai-chat-history==1.1.2elsai-chat-history v1.1.1
Release Date: Sep 2026
Improvements
Messageships as plain Python —message.pyis excluded from Cython soMessage.to_dictstays a real method and Pydantic v2 can import the package- Same library as 1.1.0 — no API or strategy changes
- Pydantic v2 import error — importing 1.1.0 could raise
PydanticUserErrorbecause Pydantic treated CythonizedMessage.to_dictas a non-annotated attribute.message.pynow ships as plain Python so that error no longer occurs. Prefer 1.1.2 (or at least 1.1.1) - Dependency pins are unchanged from 1.1.0 —
pydantic==2.11.7,packaging>=23.2,<25,typing_extensions==4.14.0. Those exact pins can conflict with AgentKit
Installation
bash
pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-chat-history/ elsai-chat-history==1.1.1elsai-model v2.1.1
Release Date: Sep 2026
Improvements
- LiteLLM prompt caching — fixed prompt cache control propagation (
cache_control) when using LiteLLM model provider and agent paths - Bedrock cache metadata & usage — normalized InvokeModel and Converse cache usage into the Elsai
Usagecontract (cacheReadInputTokens,cacheWriteInputTokens,cacheDetails,requestedCacheTtl); addedstream_text_with_usage()/astream_text_with_usage();CacheConfig(strategy="auto", ttl="5m"|"1h")for automatic cache point injection
See Amazon Bedrock — Prompt caching for usage examples.
Installation
bash
pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-model/ elsai-model==2.1.1elsai-model v2.1.0
Release Date: Jul 2026
New features
LLM/Providerfactory — construct any provider viafrom elsai_model import LLM, Provider- Lean installs — provider SDKs moved to optional extras (
openai,gemini,bedrock, …,all) - Existing
*Modelclasses unchanged; the factory returns the same instances
Installation
bash
pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-model/ elsai-model==2.1.0Add provider extras as needed — see LLM Models — Optional package extras.
v3.0.0
Release Date: Jun 2026
elsai Model v2.0.0 — unified *Model provider API
New features
- Unified
Modelbase class — all providers expose.invoke(),.stream(), and.stream_text() - Agent integration — pass
*Modelinstances directly toAgent(model=…) - New providers — Mistral, Writer, Meta Llama API, llama.cpp, OpenAI Responses, SageMaker
- Optional extras —
openai-responses,llamaapi,sagemaker,langchain
Breaking changes
*Connectorclasses andGeminiServicereplaced by*Modelclasses- Constructor shape:
model_id,client_args,paramsinstead of provider-specific connector kwargs - Legacy connector API documented at v2.0 LLM Models
Migration
| Legacy (v1.4.x) | v3 |
|---|---|
OpenAIConnector(model_name=…) | OpenAIModel(model_id=…, client_args=…) |
BedrockConnector(aws_access_key=…) | BedrockModel(model_id=…, region_name=…) |
GeminiService.generate_text(…) | GeminiModel.invoke(messages) |
See LLM Models and Model Providers.
Installation
bash
pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-model/ elsai-model==2.0.0v2.0.0
Release Date: July 2025
New features
elsai Cloud Connectors
- Microsoft Outlook Integration — Send emails with attachments and rich HTML formatting using the Microsoft Graph API.
- Microsoft Delta API Support — Track incremental changes in Outlook, OneDrive, and SharePoint data.
- EmailService — Generic email sending service.
elsai OCR Extractors v2.0.0
- Mistral OCR — Support for the Mistral OCR engine for complex document layouts.
- Amazon Boto3 Connector — Native connectors for AWS services using the Boto3 SDK (sync and async modes).
- Amazon Textractor — AWS Textract integration for document analysis (sync and async workflows).
elsai Model v1.0.0
- Gemini Integration — Google Gemini with full multimodal generation capabilities.
- LiteLLM Integration — Unified interface across all LLM providers through LiteLLM.
- Async OpenAI & Azure OpenAI — Asynchronous connectors for high-throughput inference.
elsai VectorDB v2.0.0
as_retriever()method — All vector stores now expose a retriever interface for seamless integration withelsai-retrieversin RAG workflows.
Changes
elsai Model
- Native Bedrock implementation — Migrated from LangChain to a native boto3 implementation, reducing latency and dependency overhead.
elsai OCR Extractors
- VisionAI enhancements — Sync and async processing, configurable batch sizes.
- Azure Document Intelligence — Accepts additional keyword arguments for customization.
Upgrade guide
bash
pip install --upgrade \
--extra-index-url https://core-packages.elsai.ai/root/elsai-model/ elsai-model==1.4.1 \
--extra-index-url https://core-packages.elsai.ai/root/elsai-ocr-extractors/ elsai-ocr-extractors==2.0.1 \
--extra-index-url https://core-packages.elsai.ai/root/elsai-cloud-connectors/ elsai-cloud==2.1.0 \
--extra-index-url https://core-packages.elsai.ai/root/elsai-vectordb/ elsai-vectordb==2.2.1v1.0.0
Initial release
| Package | Version |
|---|---|
| elsai-model | 1.0.0 |
| elsai-embeddings | 0.1.0 |
| elsai-vectordb | 1.0.0 |
| elsai-retrievers | 0.1.0 |
| elsai-chat-history | 0.1.0 |
| elsai-utilities | 0.1.0 |
| elsai-text-extractors | 0.1.0 |
| elsai-ocr-extractors | 1.0.0 |
| elsai-cloud-connectors | 0.1.0 |
| elsai-db | 0.1.0 |
| elsai-stt | 0.1.0 |
| elsai-mcp | 0.1.0 |
Package version history
elsai-model
| Version | Changes |
|---|---|
| 2.1.3 | Sep 2026 — TypeSafeClassifier (Jev) via the typesafe extra; not accepted by LLM(...) |
| 2.1.1 | Sep 2026 — LiteLLM prompt caching fix; Bedrock cache TTL, usage reporting, and stream_text_with_usage |
| 2.1.0 | Jul 2026 — LLM / Provider factory; lean per-provider extras (openai, gemini, bedrock, …, all) |
| 2.0.0 | Jun 2026 — unified *Model provider API; see v3.0.0 |
| 1.4.1 | Legacy connector API (v1.4.x) |
| 1.3.0 | implementation param (native/langchain) for OpenAI and Azure |
| 1.2.2 | All connectors return full response object |
| 1.2.1 | Gemini logprobs support |
| 1.2.0 | Bedrock invoke_with_image for OCR |
| 1.1.2 | Gemini response format change |
| 1.1.1 | Bedrock Messages API (Claude Sonnet 4 support) |
| 1.1.0 | AnthropicBedrockConnector; Bedrock streaming |
| 1.0.1 | Streaming for OpenAI and Azure OpenAI |
| 1.0.0 | Gemini; LiteLLM; native Bedrock boto3 |
elsai-vectordb
| Version | Changes |
|---|---|
| 2.2.1 | Current stable |
| 2.1.0 | Filter-based retrieve_document for ChromaDB and Pinecone |
| 2.0.0 | as_retriever() method added |
| 1.1.0 | update_document, delete_document, list_collections, delete_collection |
| 1.0.0 | ChromaDB, Pinecone, Weaviate |
elsai-ocr-extractors
| Version | Changes |
|---|---|
| 2.0.1 | Current stable |
| 2.0.0 | Mistral OCR; Amazon Boto3 Connector; Amazon Textractor |
| 1.0.0 | VisionAI sync/async; Azure Document Intelligence keywords |
elsai-embeddings
| Version | Changes |
|---|---|
| 0.2.0 | Amazon Bedrock embeddings |
| 0.1.0 | Azure OpenAI embeddings |
elsai-chat-history
| Version | Changes |
|---|---|
| 1.1.2 | Relaxed pydantic / packaging / typing pins for AgentKit 0.3.5 |
| 1.1.1 | Message ships as plain Python; fixes Pydantic v2 import error |
| 1.1.0 | SemanticMemoryStrategyImpl, SimilaritySearchStrategy |
| 1.0.0 | Initial release |
elsai-cloud-connectors
| Version | Changes |
|---|---|
| 2.1.0 | Current stable |
| 1.0.0 | Outlook; Delta API; EmailService |
Other packages
| Package | Version | Notes |
|---|---|---|
| elsai-graph-constructor | 0.1.0 | Initial release |
| elsai-graph-generator | 0.1.0 | Initial release |
| elsai-graph-query-rag | 0.1.0 | Initial release |
| elsai-retrievers | 0.1.0 | Initial release |
| elsai-text-extractors | 0.1.0 | Initial release |
| elsai-parsers | 0.1.0 | Initial release |
| elsai-nli | 0.1.0 | Initial release |
| elsai-stt | 0.1.0 | Initial release |
| elsai-utilities | 0.2.0 | Current stable |
| elsai-db | 0.2.0 | LangChain upgrade |
| elsai-mcp | 0.1.1 | Current stable |