Chainlit

Fixed

- Langchain Callback handler IndexError
- Attempt to fix websocket issues

Added

- The `User` class now has a `display_name` field. It will not be persisted by the data layer.
- The logout button will now reload the page (needed for custom auth providers)

Changed

- Directly log step input args by name instead of wrapping them in "args" for readability.

Fixed

- Langchain Callback handler ValueError('not enough values to unpack (expected 2, got 0)')

Changed

- hide_cot becomes cot and has three possible values: hidden, tool_call, full
- User feedback are now scoring an entire run instead of a specific message
- Slack/Teams/Discord DM threads are now split by day
- Slack DM now also use threads
- Avatars are always displayed at the root level of the conversation

Removed

- disable_feedback has been removed
- root_message has been removed

Added

- Messages are now editable. You can disable this feature with `config.features.edit_message = false`
- `cl.chat_context` to help keeping track of the messages of the current thread
- You can now enable debug_mode when mounting Chainlit as a sub app by setting the `CHAINLIT_DEBUG` to `true`.

Fixed

- Message are now collapsible if too long
- Only first level tool calls are displayed
- OAuth redirection when mounting Chainlit on a FastAPI app should now work
- The Langchain callback handler should better capture chain runs
- The Llama Index callback handler should now work with other decorators

Added

- Mistral AI instrumentation