Releases: danny-avila/LibreChat
v0.3.3
What's Changed
An openAI preset used with this project helped me write this! Check it out on our discord
New Functionality:
- Created Docker Hub repo, addressing #182 [f5ff91c]
- Changed docker-compose.yml to use pre-built image for api service instead of building from local file [f5ff91c]
- Use gen_title response for askChatGPTBrowser.js (official title). [017447b]
- Export conversation: csv, json, txt, markdown, screenshot. [6f0b559] [9691438]
- Set document title when conversation is switched. [e0b0b68]
- Add support for deleting individual presets. [83df28f]
- Add RQ tokenizer. [88aea81]
- Add support for saving messages to database. [a81bd27]
- Support user-provided token to bingAI and chatgptBrowser by @wtlyu in #177
Bug Fixes and Cleanup:
- Fix abort messages and continue conversation on abort , prevent page refresh on stop generating button by @danny-avila #176
- fix nginx container not containing the actual build output by @llk89 in #178
- fix: fix infinite query failure when conversationId is not found by @danorlando in #179
- Clean code with newConversationId in askbingai. [fc91ed4]
- Refactor Message.jsx to remove cancelled message bubble and improve wording of unfinished message. [0bd2409]
- Use new conversation Id [3b94a98], remove blank in screenshot [5664a0c], and remove use-screenshot [c271f04].
- Fix withAuthentication on post requests [94e0636], fix conversations in nav (put refreshConvoHint back) and pagination [9a0e380].
- Fix error and default tone for Sydney [e5e4ee2], and fix bug when clicking on title in search results [c983670]. by @HyunggyuJang and @danorlando
- Resolve JavaScript heap out of memory error from Vite [6e42d4f], and fix isNotAppendable condition [78ae220].
- Address infinite query failure when conversationId is not found [fb7542c], and correct typo in environment variable name [6aa540c].
What's coming
- User system is in progress
- Currently Implementing a reverse engineering of the chatgpt plugins functionality. Official plugins are proving tricky and error prone but a good agent solution that can use community created plugins is much more likely and functional. Possibly also autoGpt/agentGpt in the future
- Models supported by llama.cpp can be easily integrated and will get around to this optional feature. The idea is to have users install them as desired.
- LLaMA 🦙, Alpaca, GPT4All, Chinese LLaMA / Alpaca, Vigogne (French), Koala
- the official site UI has received a good update and will get around to matching it, possibly as a configurable setting
- Optimization of search function and database CRUD for multi-user system
New Contributors
Full Changelog: v0.3.0...v0.3.3
v0.3.0
What's Changed
Introducing more customization for both OpenAI & BingAI conversations!
See screenshots below
Credits to @wtlyu for the system design here, and a lot of the hard work refactoring #142
Also thanks to @danorlando for collaborating and making improvements to the Dev environment
.env
variable for your Bing token has been renamed to BINGAI_TOKEN
Model menu Redesign: now Conversation Menu, serving Endpoints & Presets:
- Models are now organized by Endpoints (openAI, BingAI, and chatGPTBrowser respectively)
- Introducing Presets: Save Settings as Presets, which you can share (import/export)
Customizable Endpoint Settings
- OpenAI API: select model and open advanced settings
- Advanced: Adjust model parameters such as Temperature, Top P, and Frequency and Presence Penalties
- Also select model, adjust custom name and prompt prefix as before
- Advanced: Adjust model parameters such as Temperature, Top P, and Frequency and Presence Penalties
- Bing: enable jailbreak mode, tone style, and open advanced settings
- Advanced: Input up to 7k tokens for context and/or write a custom system Message for Jailbreak Mode
- Full instructions on writing a system message here
- ChatGPT Browser: select model (useful for plus members)
Presets
- This is really exciting! Save your presets, which will show up in the conversation menu
- Save Presets in the advanced settings of OpenAI/BingAI
- Export JSON files of your Preset in the Edit view (edit from convo menu)
- Import presets from the convo menu
- Switch to a regular endpoint to reset to default settings
Do share your presets in our Discord Server in the #presets channel!
Upcoming Changes
- Export conversations through a variety of methods
- Built-in multi-user support
- Test environment for CI/CD
Full Changelog: v0.2.0...v0.3.0
v0.2.0
What's Changed
- Big refactor of client code: @wtlyu #140
- Chat routes, navigate right to your favorite conversations by URL
- state mgmt. switch from redux to recoil
- less code, more organized
- Distinction between Fast vs. Balanced Tone Style for Bing @HyunggyuJang #127 #128
- Resized Tone style selections fa0c56f
- CSS fixes, bug fixes, and better matching with official site #137
- You can now set your default model for Chat API completions in .env file, 45f4145
- this will soon be an option on the frontend
Upcoming Changes
- More customizable parameters for both ChatGPT API & Bing
- Built-in multi-user support
- Test environment for CI/CD
Full Changelog: v0.1.1...v0.2.0
v0.1.1
What's Changed
Adds Tone Styles to Bing
- Tone Styles: Select between 3 options
- Will persist across conversations
- Tabs move along with textarea height
- Only configurable at the start of a conversation
- A lot of people prefer "Creative" with all convos with Bing/Sydney so I recommend this
- Minor updates:
- Many bug squashing done by myself and @HyunggyuJang
- adds very in-depth local install instructions thanks to @fuegovic
- setup docker-compose environment variables for meilisearch
Immediate roadmap:
- more customGPT and Sydney parameters (response temps, changing openai models, editing system message for Sydney)
- Improve bing features (styling, params, fix formatting bugs)
- Cleanup code & writing a test suite for better CI/CD
Notable planned features:
- Prompt templates
- Keyboard commands
- Folders view in Nav
- Configurable settings per user for UI preferences
Full Changelog: v0.1.0...v0.1.1
v0.1.0
What's Changed
Adds SearchBar for Fast Message & Conversation searching
- Optional and configurable, powered by Meilisearch
- Basic usage: will search after short delay of last keystroke in searchbar
- Shows up to 20 messages matching the query. No messages will result in Landing Page.
- Navbar shows conversations with at least 1 hit in messages or title, sorted by most number of hits within a convo, to least
- No conversation matches will show empty nav bar, and will show as many hits with criteria above
- Easily clear the searchbar by hitting 'new chat'
- Click a conversation title over a message to open a conversation and use as normal.
- Easy setup with docker, necessary environment variables shown and explained in
.env.example
Click here for Full Setup Instructions
Enabling Search:
- Requires installation of free self-hosted Meilisearch or Paid Remote Plan (Remote not tested)
- The easiest setup for this is through docker-compose, which takes care of it for you.
- for local instructions, see the official meilisearch site
- Follow .env.example for proper env variable setup.
- Note: environment variables are DIFFERENT for Local vs. Docker use.
Local batch file instructions (windows)
@fuegovic (thank you for this): In case someone is wondering, here's the content of the batch file I used to get the search working without docker. The search function also needs to be enabled in the .env file.
@echo off
REM place the meilisearch executable at the root of the chatgpt-clone directory
start "MeiliSearch" cmd /k "meilisearch --master-key WjRvKvpelIYHa4dlcDGsdT9c0CKRbY-4VcuCQ1iZPcc
REM meilisearch is the name of the meilisearch executable ↑↑↑
REM meilisearch will generate a master key when you start meilisearch without providing one
start "ChatGPT-Clone" cmd /k "cd api && npm start"
REM this batch file goes at the root of the chatgpt-clone directory (edited)
I think that's the bare minimum to have it working, I just downloaded the meilisearch executable from github and placed it in the root folder after pulling the search-final branch (I also did the npm ci and npm build steps once after updating)
Other changes
- Clearing all conversations now requires confirmation from prompt
- Edit user messages in conversations with Sydney for branching the conversation #115
- Supports changing ChatGPT Browser model through .env file, see .env.example for proper env variable setup.
- This enables use of GPT-4 for ChatGPT Plus members through the ChatGPT Browser model #114
Immediate roadmap:
- more customGPT and Sydney parameters (response temps, changing openai models, editing system message for Sydney)
- Improve bing features (styling, params, fix formatting bugs), thank you @HyunggyuJang
- Cleanup code & writing a test suite for better CI/CD
Notable planned features:
- Prompt templates
- Keyboard commands
- Folders view in Nav
- Configurable settings per user for UI preferences
Tagging for major contributions up to this point @wtlyu
Full Changelog: v0.0.6...v0.1.0
v0.0.6
What Changed
This is the latest stable release before message/conversation search goes live!
By request, there is now also a community discord server
- Significant improvements across the board to existing features, the biggest of which is better Markdown handling
- Cursors follows text better and correctly blinks as expected
- If there is ever any choppiness, this is due to the AI stream which is not easily controlled and random.
- Code blocks more stable
- Cursors follows text better and correctly blinks as expected
- Sydney is once again available through new jailbreaks (though formatting still needs improvement in edge cases)
- Uses latest api client, allowing usage of GPT-4 (must be manually set at the moment)
- Better docker config with options #108
Following soon after this release (tomorrow):
- Search all messages and conversations feature (optional and requires meilisearch, provided through docker)
Immediate roadmap:
- more customgpt/bing parameterss (response temps and changing openai models)
- improve bing features (smoother streaming, style params, fix formatting bugs)
- cleanup code & writing a test suite for better CI/CD
Notable planned features:
- Prompt templates
- Keyboard commands
As always open to more integrations, contributions, suggestions as they come up and time allows
v0.0.5
Changes
- Adds Stop generating & Regenerate Message options #88
- Skeleton for Multi-User support (bare bones integration, currently DIY) 2ad9b33 8029ed7
- Models without credentials are omitted from model menu 29ee765
- Better titling (and titling as a separate request from messages) f1e3bbd
- Various UI fixes/additions
o Current model highlighted at the top of the conversation
o Better textarea and button styling for mobile
On my immediate to-dos for this weekend:
- search messages feature (75% done already)
- more customgpt params (including changing openai models)
- rewrite code/markdown handling (I have many things in mind to significantly clean this up) #67 #69
Stretch Goals if time allows:
- improve bing features (smoother streaming, style params, fix formatting bugs)
- cleanup code & writing a test suite for better CI/CD
v0.0.4
Many thanks to @wtlyu as they initiated and coded a lot of the features in this release, see a lot of the work done here: #55
Changes
-
Resubmit messages & Branching messages
- Mirrors official ChatGPT feature: edit sent message to branch messages (works only with ChatGPT)
v004demo.mp4
-
Paginated conversation list
-
Model menu icons
Immediate Roadmap
- Message search
- Multi-user system
- Bing AI Styling (for conversation style, suggested responses, etc.) - In progress
- Address issues & feature requests
- See complete roadmap in readme.md
v0.0.3
Changes
This is to mark a new tag before a major PR merge, to properly document the latest changes before.
- Mobile Styling added #43
- Build docker container with one command now from root dir #56
- Auto focus support added to Chat Input #58
- Model menu key issue fixed #54
- Titles now generates in user language #40
- Host name and proxy address configuration support added #26
- Added Bing styling (citations with links that open in new tabs) #23
- Spinner indicating messages are loading
- Unsupported code language errors & misc. reported errors fixed
v0.0.2
Changes
- Adds Sydney (jailbroken Bing AI) to the model menu, thanks to DavesDevFails for bringing it up: #13.
- No limits on Bing AI through Sydney
- Keep token usage small (words/messages/convo) to avoid any Jailbreak errors
- Bing/Sydney now correctly cite links, with sources appended to message. More styling to come.
- Minor fixes/improvements
- Ensures new conversations are created on all model state changes
- Model menu won't close on model deletion
- Re-enabled ChatGPT free since it might be working for most people (not for me)
- First outside contributions! Thank you HyunggyuJang for #11 & #12