CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file. See standard-version for commit guidelines.

9.1.0 (2024-06-21)

Features

• add anthropic claude llms (1e09f67)
• add api rate-limiter support (e4a8863)
• add azure openai (4478ed1)
• add built-in function for embeddings (37dfa99)
• add busines information extraction prompt (cb2dca5)
• add caching proxy (ba57bf9)
• add caching proxy (335dcbd)
• add google gemini safety controls (7783bbe)
• add google palm models (cf6bf11)
• add gpt-4 support (405b87e)
• add ollama (c1181e0)
• add semantic router (2096158)
• add streaming support for gemini and cohere (a946d52)
• add streaming support to proxy (5edd33c)
• add support for embeddings to use with vector search (4bd4ba4)
• add support for json type and other fixes (3249746)
• add together compute llm api support (37fc9cf)
• add vector db and embeddings example (141ad9f)
• added a new llm alephalpha (bff5f51)
• added agent tracing (4bc9ae7)
• added claude3, gemini and fixed openai azure (1c902d1)
• added dsp (fc9c292)
• added in memory vector db with serialize to disk (5036933)
• added mistral support (67747a0)
• added more tests (fe33451)
• added multi-modal support to anthropic api and other fixes (95a0680)
• added openai chat-gpt api support (eb4b151)
• added prefix Ax across type namespace (3a9daf0)
• added proxy bin (65adabe)
• added the dsp style bootstrap few stop optimizer (eab69c8)
• agent framework, agents can use other agents (93cbfb3)
• agent framework, agents can use other agents (a7420e7)
• ai balancer to pick the cheapest llm with automatic fallback (a8f7b7b)
• auto-fix json syntax errors (dc27812)
• automatic long term memory (e94ffd5)
• automatic vectordb retrieval augmented generation in proxy (d081c18)
• aws bedrock support (acdc89b)
• big [breaking] refactor (c97395d)
• convert any document format to text and a full RAG engine builtin (7f3f28c)
• dbmanager handles chunking, embedding and search for text (5c125f8)
• deepseek and cohere function calling (5341700)
• enable remote logging with LLMC_APIKEY env var (e04b257)
• gpt-4o added (a388219)
• huggingface data loader (47c6c0e)
• improve trace logging (12776be)
• improved debug logs (38e869e)
• include reasoning in tracing (e11f665)
• iterate to completion if max token length reached (f9d0f50)
• JS code interpreter function (b3de309)
• library is now down to 1 single dependency (efaaa03)
• llm converts meeting notes to trello tasks example (633cb95)
• lots of api improvements and many bug fixes (036cd06)
• major refactor to enable traceing and usage tracking (a7c980c)
• make it easier to customize prompts (1870ee7)
• migrated from commonjs to ES2022 (c8ad44b)
• multi-modal dsp, use images in input fields (a170b55)
• new agent framework (1e7040c)
• new agent prompt (6c4df2c)
• new llm proxy for tracing usage (ab36530)
• new llmclient command and crawler to embed and vectorize websites (47a61f2)
• new llmclient command and crawler to embed and vectorize websites (ede47c5)
• openapi wispher support (50e864f)
• parsing and processing output fields and functions while streaming (95a7a93)
• proxy support for all llms (1e1edc3)
• req tracing added to ai classes (dad8c3c)
• spider to embed website (56f4598)
• streaming support (65839a9)
• support for google vertex (49ee383)
• support for Hugging Face and other updates (8f64432)
• support for remote tracing (033a67a)
• track token usage (bd4f798)
• true realtime output validation while streaming (4308b99)
• updates to agent framework api (472efbf)
• vector db query rewriting (52fad9c)
• vector db query rewriting (bce6d19)
• vector db support (0ea1c7f)
• welcome llm-client (225ec5a)

Bug Fixes

• add opentelemetry support and other fixes (685fe80)
• Add Polyfill for TextDecoderStream to Ensure Compatibility with Bun #21 (540348d)
• add tools support to anthropic (1cc96b7)
• added missing typescript docs (2459a28)
• added system prompt to trace (335bb40)
• anthropic header issue (40dbce0)
• anthropic proxy endpoint (#7) (cf7c793)
• anthropic, updated other models (6e83d34)
• Azure OpenAI chat/completion call failed (#19) (#20) (0fad4f9)
• banner fixes (fdf453b)
• blank response with stream (ab7c62d)
• bug in new open ai chat model (654282d)
• bug with exporting enums (4fc0a3b)
• bug with exporting enums (4ca8c5d)
• build error (e4d29c3)
• build fixes (d8d4a47)
• build issue (7733463)
• build issue (071f476)
• build issue with previous version (cfe92fd)
• build issues (dd346c4)
• build issues (33394df)
• change name in package (9697d19)
• cleaup gitignore (6627ff3)
• common llm model config to ensure more deterministic outputs (0dce599)
• examples (973eb51)
• extra comma in package.json (16b36f5)
• fix in proxy (f778c9d)
• fix in proxy (f007d26)
• fix in proxy (ab87118)
• google gemini function calling (a9214f3)
• google gemini now works great (a6d6528)
• import issues (ce87294)
• improve examples (6e656cb)
• improved the customer suppot example (103b072)
• issue with cspell (806cba1)
• issue with rate limiter (0648ad7)
• json5 build issue (fe6de9c)
• major fixes to function calling (d44a373)
• make mistral function calling work (6bd9cf7)
• make stop sequence optional (d18addc)
• migrate npm package to llmclient (97a4a07)
• minor build fix (32f04b3)
• minor fix (62fbf38)
• minor fixes (3792700)
• minor fixes (574f73f)
• minor fixes (5ebf283)
• minor fixes (5bb612d)
• minor fixes (0fa4687)
• minor fixes (47eef31)
• minor fixes (0da9daf)
• minor fixes (15bd72f)
• minor issue with incorrect error (65e7ea8)
• missing imports (9071c9e)
• new ai balancer to route based on token pricing in case of error (7ea79a9)
• openai function tracing fixes (d623f03)
• opentelemetry tracing added to ai and vectordb (1918410)
• prettier config (8f4ca77)
• proxy command (9eaf6e7)
• proxy port can be set using the env var PORT (c08af8d)
• refactored request, response tracing (1ac7023)
• refactored usage reporting and other fixes (84bf661)
• remove comments (73e3a30)
• remove index folder committed by mistake (4d327ab)
• removed crawler (ab4ac6e)
• removed extra packages (faae6d2)
• renamed emailprompt to messageprompt (3063a90)
• renamed responseSchema to resultSchema for clarity (b238b88)
• resolved all typescript strict mode errors and warnings (4f08fac)
• result error-correction fixes (eeaa12e)
• rewrite or error-correction code and other fixes (37b620e)
• seperated embed usage data from completion usage (feb5619)
• set default model for openai to gpt3-turbo (14ba73c)
• signature parser (a149361)
• streamlined the llm apis (0bbc8b0)
• system prompt fixes (b56201c)
• tests breaking (4c047ce)
• ts cleanup (be645d9)
• tuning and optimization fixes (d0b892c)
• type validation logic (88eec1e)
• update model defaults (1c70dd7)
• update model defaults (cc7e82f)
• update model defaults (1251372)
• update package name (ed5346c)
• Updated Cohere reqValue to include input_type (required since v3), Groq model to llama3-70b-8192 (llama2-70b-4096 was depreciated), and added a temporary fix to the marketing.py example (messageGuidelines expects a string, not a string array). (#22) (5885877)
• updates to debugging proxy (6476a23)
• valid healthcheck path for proxy (3adea76)
• various fixes (a0099a8)
• various fixes (7b75342)
• various fixes (440cb84)
• various fixes (516fbcb)
• various fixes (0978b29)
• various proxy fixes (61b693a)
• version update (c4794a3)
• zprompt type spec (db1f7e7)