Previously chat stream iterator wasn't closed when response streaming
for offline chat model threw an exception.
This would require restarting the application. Now application doesn't
hang even if current response generation fails with exception
GPT-4o-mini is cheaper, smarter and can hold more context than
GPT-3.5-turbo. In production, we also default to gpt-4o-mini, so makes
sense to upgrade defaults and tests to work with it
- Background
Llama.cpp allows enforcing response as json object similar to OpenAI
API. Pass expected response format to offline chat models as well.
- Overview
Enforce json output to improve intermediate step performance by
offline chat models. This is especially helpful when working with
smaller models like Phi-3.5-mini and Gemma-2 2B, that do not
consistently respond with structured output, even when requested
- Details
Enforce json response by extract questions, infer output offline
chat actors
- Convert prompts to output json objects when offline chat models
extract document search questions or infer output mode
- Make llama.cpp enforce response as json object
- Result
- Improve all intermediate steps by offline chat actors via json
response enforcement
- Avoid the manual, ad-hoc and flaky output schema enforcement and
simplify the code
This is a more robust way to extract json output requested from
gemma-2 (2B, 9B) models which tend to return json in md codeblocks.
Other models should remain unaffected by this change.
Also removed request to not wrap json in codeblocks from prompts. As
code is doing the unwrapping automatically now, when present
- Allow free tier users to have unlimited chats with default chat model. It'll only be rate-limited and at the same rate as subscribed users
- In the server chat settings, replace the concept of default/summarizer models with default/advanced chat models. Use the advanced models as a default for subscribed users.
- For each `ChatModelOption' configuration, allow the admin to specify a separate value of `max_tokens' for subscribed users. This allows server admins to configure different max token limits for unsubscribed and subscribed users
- Show error message in web app when hit rate limit or other server errors
Currently, the search model config display for admins only shows the id of the search model config, which is not very informative.
The changes enhances the admin console by displaying the name of the search model config (name), as well as the bi-encoder model (bi_encoder) and cross-encoder model (cross_encoder) along the id.
Previously `force' was passed as a query param to the single indexing API. After the recent API updates, it is meant to select the API method to use (PATCH vs PATCH). Converting `force' argument to a bool fixes implementing this new behavior
- Major
- Improve doc search actor performance on vague, random or meta questions
- Pass user's name to document and online search actors prompts
- Minor
- Fix and improve openai chat actor tests
- Remove unused max tokns arg to extract qs func of doc search actor
- Issue
Previously the doc search actor wouldn't extract good search queries
to run on user's documents for broad, vague questions.
- Fix
The updated extract questions prompt shows and tells the doc search
actor on how to deal with such questions
The doc search actor's temperature was also increased to support more
creative/random questions. The previous temp of 0 was meant to
encourage structured json output. But now with json mode, a low temp is
not necessary to get json output
- Use temperature of 0 by default for extract questions offline chat
actor
- Use temperature of 0.2 for send_message_to_model_offline (this is
the default temperature set by llama.cpp)
* Add ability to cycle through the chat history in the chat input on Obsidian (similar to terminal history navigation)
* Add mod key shortcut to cycle through chat history in chat input
* Add shortcut help text in chat input placeholder
---------
Co-authored-by: Debanjum Singh Solanky <debanjum@gmail.com>
### Overview
Support exclude file filter in user search queries
### Details
- All of the exclude file filter terms need to be satisfied
- Any one of the include file filter terms should be satisfied
### Example
- **Search Query**: *what happened yesterday? -file:"tasks.org" -file:"work.md" file:"diary.org" file:"journal.org*
- **Behavior**: Query will try find relevant notes in any of `journal.org` or `diary.org` and not in `tasks.org` and not in `work.md`
### Details
* Add support for exclusion file filters
* Translate file filter to valid Django DB entry filter regex
* Exclude all files when multiple exclude file filter in query
Previously we were applying an "Or" filter, which would exclude any
file mentioned in a query with multiple exclude file filter.
This is not what we naturally mean when we ask excluding a file in a query
* Rename, rearrange, deduplicate and add file filter tests
Closes#728
---------
Co-authored-by: Debanjum Singh Solanky <debanjum@gmail.com>
Previously required the automation page to be refreshed to see updates
to the automation in the edit automation card. This would be seen when
user tries to edit an automation multiple times (without a page refresh)
Previously, the code incorrectly treated all non-nil values as true, leading to
the index being re-indexed with the force flag whenever the user selected to
update the index.
- Pass the new conversation id as kwarg for the scheduled_chat function
- For edit automations, re-use the original conversation id
- Parse images correctly for image automations
- Use color to provide visual feedback when hover, click on feedback
buttons
- Use color to provide visual feedback when hover on speech, copy
buttons click
- Add cooldown period before being able to send feedback on that message again.
Avoids inadvertent multiple consecutive clicks on feedback buttons
- Since the .gitignore will ignore any of the assets in the src/ folder when building the package wheel, we need to output the static assets to another folder just for the python pypi package. Use /compiled for this.
- Auto focus on email input on login screen for smoother login experience
- Use file icon associated with search page results. Improve search bar
- Show logged in user's email in nav menu for context
- Use previous icons with eyes for search, agents and automations items in nav menu
Hierarchical documents like org-mode, markdown have their ancestry
shown in first line. Remove it to show cleaner, deduplicated reference
text from org-mode, markdown files
Utilize chat footer space more efficiently. This is especially useful
on small screens
- Send button is anyway only enabled when there is text in chat input
- Otherwise voice message button is better to show by default
- Remove invalid call to styles.main
- Remove unnecessary top padding above side pane to keep side pane at
consistent position across web app
- Use same pageLayout styles and styling structure on agent like
automation
- Vertically center automation section and page title on it's row
- Fix applying flex vs grid with tailwind
- Remove x axis footer padding on small screens to preserve space,
keep equal spacing between footer items
- Add 1rem margin to buttons to not have overlap in boundary
- Add 1rem y-axis padding to chat footer to not have focus boundary
leave the footer boundary on smaller screens
Installing Khoj as PWA was supported in previous web UX as well. This
just adds link to the existing webmanifest to continue support for
installing Khoj as PWA with new web UX
Previously the rename wasn't updating the chat session title. We'd
have to refresh the page or side pane to get latest chat session names
after rename action.
Previously the footer's right border wasn't visible on small screens
due to usage of w-full
Use mr-1 on send button instead of px-1 on chat input parent to
eualize chat footer buttons spacing
- Show informative toast messages on copy, delete of API keys
- Onle show API keys card in non anonymous mode. API keys aren't
required (and is disabled on server side) in anon mode. Not showing
card at all in anon mode reduces chance of unnecessary confusion
Style profile pircture button on nav menu
- Use primary colored ring around subscribed user profile on nav menu
- Use gray colored ring around non-subscribed user profile on nav menu
- Use upper case initial as profile pic for user with no profile pic
- Click anywhere on nav menu item to trigger action
Previously the actual clickable area was smaller than the width of
the nav menu item
- Move the nav menu into the chat history side panel component, so that they both show up on one line
- Update all pages to use it with the new formatting
- in mobile, present the sidebar button, home button, and profile button evenly centered in the middle
- Pass userConfig from Home as prop to chatBodyData component with
loading state
- Pass loading state of userConfig to allow components to handle
rendering dependent elements once it is loaded
Use updated format for HTTP streamed responses from the Khoj server in the new chat UX
Remove references to the websocket connected field, as websocket use has been deprecated
Otherwise the Khoj's chat response is filling up in between the
streamed message and already rendered references section at the bottom
of the message
Define OnlineContext type to simplify typing online context param
across other interfaces and functions
- There were some state mismatches in configuring a whatsapp number. This commit fixes those issues and uses an external library for phone number validation
- Note that the SSR for next doesn't support rendering on the client-side, so it'll only update it one big chunk
- Fix unique key error in the chatmessage history for incoming messages
- Remove websocket value usage in the chat history side panel
- Remove other websocket code from the chat page
- Why
Profile section and billing section looked too empty (1 card each).
Combining them makes the setting page look more complete. Shows
subscription options early on
- Details
- Made Futurist text orange
- Made Unsubscribe a down button instead of cloud slash
- Updated toast title to subscription
- Improve Futurist trial title and description
- Remove now unnecessary button to Save in Card with dropdown
- Use toast to show success, failure (not working)
- Rename language to search, Move it to features section. Add icon to
the card
- Update references in new and old web client settings
- Arrange new client settings props and add header comments similar to
- config response for code readability
- Add a lot more suggestions cards, improve mobile rendering of suggestion cards, improve alignment of chat input, shift message when starts recording voice, remove dead code
* Converted navigation menu into a dropdown menu
* Moved collapsed side panel menu icons into top row
* Auto refresh when conversation is deleted to update side panel and route back to main page if deletion is on current conversation
* Highlight the current conversation in the side panel
* Dynamic homepage messages with current day and time of day.
* `colorutils` upgraded to have more expansive tailwind color options and dynamic class name generation.
* Converted create agent button alert into shadcn `ToolTip`
* Colored lines and icons for agents in chat window
* Cleaned up border styling in dark mode
* fixed three dot menu in side panel to be more easier to click
* Add the KhojLogo import in the nav menu and use a default user profile icon when not authenticated
* Get rid of custom --box-shadow CSS variable
* Pass the agent metadat through the chat body data in order to style the send button
* Add login to the unauthenticated login view, redirecto to home if conversation history not loaded
* Set a max height for the input text area
* Simplify tailwind class names
---------
Co-authored-by: sabaimran <narmiabas@gmail.com>
## Major: Breaking Changes
- Move API endpoints under /configure/<type>/model to /api/model/<type>
- Move API endpoints under /api/configure/content/ to /api/content/
- Accept file deletion requests by clients during sync
- Split /api/v1/index/update into /api/content PUT, PATCH API endpoints
## Minor: Create New API Endpoint
- Create API endpoints to get user content configurations
Related: #852
Changes for new agents page
- Modernized agent cards
- Responsive design to support mobile users
- Button for users to create their own agents (coming soon)
- Optimized to use tailwind and icon utils
- Side panel added for quick access to conversations
Previous logic was more brittle to break with simple unbalanced
'{' or '}' string present in the event data. This method of trying to
identify valid json obj was fairly brittle. It only allowed json
objects or processed event as raw strings.
Now we buffer chunk until we see our unicode magic delimiter and only
then process it.
This is much less likely to break based on event data and the
delimiter is more tunable if we want to reduce rendering breakage
likelihood further
- Add support for text to speech, speech to text. Add loading and responsive indicators to reflect state.
- When streaming for speech to text, show incremental transcription in the message input field
- When streaming text to speech, and a pause button in the chat message to allow user to stop playback
* V1 of the new automations page
Implemented:
- Shareable
- Editable
- Suggested Cards
- Create new cards
- added side panel new conversation button
- Implement mobile-friendly view for homepage
- Fix issue of new conversations being created when selected agent is changed
- Improve center of the homepage experience
- Fix showing agent during first chat experience
- dark mode gradient updates
---------
Co-authored-by: sabaimran <narmiabas@gmail.com>
- Deduplicate code to collect chat telemetry by relying on
end_llm_response event
- Log time to first token and total chat response time for latency
analysis of Khoj as an agent. Not just the latency of the LLM
- Remove duplicate timer in the image generation path
Do not need response generator to stuff compiled references in chat
stream using "### compiled references:" separator.
References are now sent to clients as structured json while streaming
## Overview
- Gemma 2 is a new open model family by Google. They've released a 9B, 29B param model. A 2B model is also expected.
- It performs really well on the Chatbot arena and shows good performance when testing within Khoj as well.
- Llama.cpp support for Gemma 2 architecture seems to have stabilized
- If Gemma 2 performs well in further testing, it can be made the default offline chat model for Khoj
- Once the 2B param model is released, the model size to download can be automatically chosen based on (V)RAM available
## Major
- Support Gemma 2 for Offline Chat
- Improve and fix chat model prompts for better, consistent context
## Minor
- Fix and improve offline chat actor, director tests
- Improve offline chat truncation to consider chat message delimiter tokens
Previously loading animation would be at top of message. Moving it to
bottom is more intuitve and easier to track.
Remove white-space: pre from list elements. It was adding too much y
axis padding to chat messages (and train of thought)
- Details
Only return notes refs, online refs, inferred queries and generated
response in non-streaming mode. Do not return train of throught and
other status messages
Incorporate missing logic from old chat API router into new one.
- Motivation
So we can halve chat API code by getting rid of the duplicate logic
for the websocket router
The deduplicated code:
- Avoids inadvertant logic drift between the 2 routers
- Improves dev velocity
- Overview
Use simpler HTTP Streaming Response to send status messages, alongside
response and references from server to clients via API.
Update web client to use the streamed response to show train of thought,
stream response and render references.
- Motivation
This should allow other Khoj clients to pass auth headers and recieve
Khoj's train of thought messages from server over simple HTTP
streaming API.
It'll also eventually deduplicate chat logic across /websocket and
/chat API endpoints and help maintainability and dev velocity
- Details
- Pass references as a separate streaming message type for simpler
parsing. Remove passing "### compiled references" altogether once
the original /api/chat API is deprecated/merged with the new one
and clients have been updated to consume the references using this
new mechanism
- Save message to conversation even if client disconnects. This is
done by not breaking out of the async iterator that is sending the
llm response. As the save conversation is called at the end of the
iteration
- Handle parsing chunked json responses as a valid json on client.
This requires additional logic on client side but makes the client
more robust to server chunking json response such that each chunk
isn't itself necessarily a valid json.
- Convert functions in SSE API path into async generators using yields
- Validate image generation, online, notes lookup and general paths of
chat request are handled fine by the web client and server API
* Update the automations UI to be a more suitable color distribution based on new designs
* Use accented colors for the metadata, update dark mode colors
* Update form to use icons as well and render more pretty inline form labels
Pull out /api/configure/phone API endpoints into /api/phone for
more concise and sufficiently explanatory API path
Refactor Flow
1. Rename /api/configure/phone -> /api/phone
Now the API to configure all the AI models is under /api/models.
This provides better organization and API hierarchy. The /configure
url segment was redundant.
- Rename POST /api/phone to PATCH /api/phone
- Rename GET /api/configure to GET /api/settings
Refactor Flow
1. Move out POST /user/name to main api.py
2. Rename /api/configure/<type>/model -> /api/model/<type>
3. Rename @api_configure to @api_mode
4. Rename file api_config.py to api_model.py
Pull out /api/configure/content API endpoints into /api/content to
allow for more logical organization of API path hierarchy
This should make the url more succinct and API request intent more
understandable by using existing HTTP method semantics along with the
path.
The /configure URL path segment was either
- redundant (e.g POST /configure/notion) or
- incorrect (e.g GET /configure/files)
Some example of naming improvements:
- GET /configure/types -> GET /content/types
- GET /configure/files -> GET /content/files
- DELETE /configure/files -> DELETE /content/files
This should also align, merge better the the content indexing API
triggered via PUT, PATCH /content
Refactor Flow
1. Rename /api/configure/types -> /api/content/types
2. Rename /api/configure -> /api
3. Move /api/content to api_content from under api_config
- Remove unused full_corpus boolean. The full_corpus=False code path
wasn't being used (accept for in a test)
- The full_corpus=True code path used was ignoring file deletion
requests sent by clients during sync. Unclear why this was done
- Added unit test to prevent regression and show file deletion by
clients during sync not ignored now
- This utilizes PUT, PATCH HTTP method semantics to remove need for
the "regenerate" query param and "/update" url suffix
- This should make the url more succinct and API request intent more
understandable by using existing HTTP method semantics
- Add day of week to system prompt of openai, anthropic, offline chat models
- Pass more context to offline chat system prompt to
- ask follow-up questions
- know where to find information about khoj (itself)
- Fix output mode selection prompt. Log error if model does not select
valid option from list of valid output modes provided
- Use consistent names for question, answers passed to
extract_questions_offline prompt
- Log which model extracts question, what the offline chat model sees
as context. Similar to debug log shown for openai models
- Pass system message as the first user chat message as Gemma 2
doesn't support system messages
- Use gemma-2 chat format
- Pass chat model name to generic, extract questions chat actors
Used to figure out chat template to use for model
For generic chat actor argument was anyway available but not being
passed, which is confusing
- Update references to the settings page to use new url across docs
and code
- Rename desktop and web settings page to settigns.html instead of
config[ure].html
- Use name, id for every [search|chat|voice|pain]_model_option
- Rename current_model_state field to more intuitive enabled_content_source
- Update references to the update fields in config.html
- Deprecate khoj-assistant pypi package. Use more accurate and
succinct pypi project name, khoj
- Update references to sye khoj pypi package in docs and code instead
of the legacy khoj-assistant pypi package
- Update pypi workflow to publish to both khoj, khoj-assistant for now
- Update stale python 3.9 support mentioned in our pyproject. Can't
support python 3.9 as depend on latest django which support >=3.10
- Put logic to get config data, detailed or basic into router helpers module
- Use the get config func across the config pages on web clients
- Put configure content and get_notion_auth_url funcs in router helper
module to avoid circular import
Migrates the Automations page to React, mostly keeping the overall design consistent with organization. Use component library, with some changes in color. Add easier management with straightforward form and editing experience.
Use system preference for determining dark mode if not explicitly set.
- Updated references panel
- Use subtle coloring for chat cards
- Chat streaming with train of thought
- Side panel with limited sessions, expandable
- Manage conversation file filters easily from the side panel
- Updated nav menu, easily go to agents/automations/profile
- Upload data from the chat UI (on click attachment icon)
- Slash command pop-up menu, scrollable and selectable
- Dark mode-enabled
- Mostly mobile friendly
- Pass Loading message, class name via props to both inline and normal
loading spinners
- Pass loading conversation message to loading spinner when chat
history is being fetched
- Create profile card componennt. Use it for agent profile card
- Pass agent persona from khoj server via API
- Put link to agent profile page in the hover card to make it 2 clicks
away. Othewise inadvertent clicks on agent in chat view lead away to
agent page
- Use tailwind line-clamp extension to clamp card to first two lines
- Reuse class name when get slash command icons
- Previous chat input styling didn't have the cursor centered in the
chat input text area. But it did allow seeing multi line chat inputs
for context
- Major
- Ask for prompt in prose
- Remove seed from SD3 image generation to improve diversity of output
for a given prompt
Otherwise for conversations with similar sounding
prompts, the images would be almost exactly the same. This maybe
another indicator of SD3's inability to capture detailed
instructions
- Consistently use "prompt" wording instead of "query" in improved
image generation prompts.
Previously a mix of those terms were being used, which could confuse
the chat model
- Minor
- Add day of week to prompt
- Remove 2-5 sentence limit on instructions to SD3. It seems to be
able to follow longer instructions just with less fidelity than
DALLE. And the 2-5 sentence instruction limit wasn't being adhered to
- Improve ability to edit, improve the image based on follow-up
instructions by the user
- Align prompts for DALLE and SD3. Only difference is to wrap text to
be rendered in quotes for SD3. This improves it's ability to render
requested text. DALLE cannot render text as well or consistently
- Because we're using a FastAPI api framework with a Django ORM, we're running into some interesting conditions around connection pooling and clean-up. We're ending up with a large pile-up of open, stale connections to the DB recurringly when the server has been running for a while. To mitigate this problem, given starlette and django run in different python threads, add a middleware that will go and call the connection clean up method in each of the threads.
- Transcribe on holding Ctrl+s keyboard shortcut
- Transcribe on holding the transcribe button pressed via mouse too
- Make the transcribe button robust to inadvertent touches by using timeout
- Do not transcribe, trigger auto-send on silences. Silence detection
is super rudimentary, just blocks standard emanations by whisper
when no speech
### Fix
- Fix degrade in speed when indexing large files
- Resolve org-mode indexing bug by splitting current section only once by heading
- Improve summarization by fixing formatting of text in indexed files
### Improve
- Improve scaling user, admin flows to delete all entries for a user
- Split once by heading (=first_non_empty) to extract current section body
Otherwise child headings with same prefix as current heading will
cause the section split to go into infinite loop
- Also add check to prevent getting into recursive loop while trying
to split entry into sub sections
Adding files to the DB for summarization was slow, buggy in two ways:
- We were updating same text of modified files in DB = no of chunks
per file times
- The `" ".join(file_content)' code was breaking each character in the
file content by a space. This formats the original file content
incorrectly before storing in the DB
Because this code ran in the main file indexing path, it was slowing down
file indexing. Knowledge bases with larger files were impacted more strongly
- Delete entries by batch to improve efficiency of query at scale
- Share code to delete all user entries between it's async, sync methods
- Add indicator to show when files being deleted on web config page
The Khoj CSP interferes with other Obsidian features and plugins as
CSP is applied page wide.
For now chat message sanitization via Dompurify should suffice.
Enable CSP when can scope it to only the Khoj Obsidian plugin.
Maybe better to fallback to non-summarize behavior if summarize intent
is just inferred but we can't actually summarize because the single
file added to conversation isn't satisfied
- Simplify quick jump between Khoj side pane and main editor view using keyboard shortcuts
- Enable voice chat in Obsidian to make interactions with Khoj more seamless
- Overview
Khoj wil be able to do online search out of the box, even for self-hosted users
- Default to Jina search, reader API when no Serper.dev, Olostep API keys
- Run online searches in parallel to process multiple queries faster
- Details
- Jina provides a [reader API](https://github.com/jina-ai/reader) for online search and web page reading
It requires no API key. This provides a good default to enable
online search for self-hosted readers requiring no additional setup.
- Jina search API also returns webpage contents with the results, so
just use those directly when Jina Search API used instead of
trying to read webpages separately. The extract relevant content from
webpage step using a chat model is still used from the
`read_webpage_and_extract_content' func in this case.
- Parse search results from Jina search API into same format as
Serper.dev for accurate rendering of online references by clients
- Run online searches in parallel with AsyncIO to process multiple queries faster
- Support Stable Diffusion 3 via API
Server Admin needs to setup model similar to DALLE-3 via Django Admin Panel
- Use shorter prompt generator to prompt SD3 to create better images
- Allow users to set paint model to use from web client config page
Jina AI provides a search and webpage reader API that doesn't require
an API key. This provides a good default to enable online search for
self-hosted readers requiring no additional setup.
Jina search API also returns webpage contents with the results, so
just use those directly when Jina Search API used instead of
trying to read webpages separately. The extract relvant content from
webpage step using a chat model is still used from the
`read_webpage_and_extract_content' func in this case.
Parse search results from Jina search API into same format as
Serper.dev for accurate rendering of online references by clients
- Added support for uploading .jpeg, .jpg, and .png files to Khoj from Web, Desktop app
- Updating indexer to generate raw text and entries using RapidOCR
- Details
* added support for indexing images via ocr
* fixed pyproject.toml
* Update src/khoj/processor/content/images/image_to_entries.py
Co-authored-by: Debanjum <debanjum@gmail.com>
* Update src/khoj/processor/content/images/image_to_entries.py
Co-authored-by: Debanjum <debanjum@gmail.com>
* removed redudant try except blocks
* updated desktop js file to support image formats
* added tests for jpg and png
* Fix processing for image to entries files
* Update unit tests with working image indexer
* Change png test from version verificaition to open-cv verification
---------
Co-authored-by: Debanjum <debanjum@gmail.com>
Co-authored-by: sabaimran <narmiabas@gmail.com>
This should improve fluidity of keyboard interactions with Khoj on
Obsidian.
Open Khoj chat view via keybinding or command pallete and ask
question using only the keyboard, with no mouse clicks required
- Automatically carry out voice chats with Khoj from within Obsidian
When send voice message, Khoj will auto respond with voice as well
- Listen to past Khoj messages as speech
- Add circular loading spinner to use while message is being converted
to speech
* Add a leader election mechanism to circumvent runtime issues for multiple schedulers
- Reduce the load on the DB and risk of issues on the service side by limiting the execution environment to one elected leader at a given time. This one is responsible for managing all of the execution of the jobs, though all workers are capable of adding and removing jobs
* Set a max duration for the schedule leader position (12 hrs), add some error if automation not added successfully
* rough sketch of desktop shortcuts. many bugs to fix still
* working MVP of desktop shortcut khoj
* UI fixes
* UI improvements for editable shortcut message
* major rendering fix to prevent clipboard text from getting lost
* UI improvements and bug fixes
* UI upgrades: custom top bar, edit sent message and color matching
* removed debug javascript file
* font reverted to Noto Sans
* cleaning up the code and removing diffs
* UX fixes
* cleaning up unused methods from html
* front end for button to send user back to main window to continue conversation
* UX fix for window and continue conversation support added
* migrated common js functions into chatutils.js
* Fix window closing issue in macos by
1. Use a helper function to determine if the window is open by seeing if there's a browser window with shortcut.html loaded
2. Use the event listener on the window to handle teardown
* removed extra comment and renamed continue convo button
---------
Co-authored-by: sabaimran <narmiabas@gmail.com>
- Add an experimental feature used for fact-checking falsifiable statements with customizable models. See attached screenshot for example. Once you input a statement that needs to be fact-checked, Khoj goes on a research spree to verify or refute it.
- Integrate frontend libraries for [Tailwind](https://tailwindcss.com/) and [ShadCN](https://ui.shadcn.com/) for easier UI development. Update corresponding styling for some existing UI components.
- Add component for model selection
- Add backend support for sharing arbitrary packets of data that will be consumed by specific front-end views in shareable scenarios
Initialize our migration to use Next.js for front-end views via Agents. This includes setup for getting authenticated users, reading in available agents, setting up a pop-up modal when you're clicking on an agent, and allowing users to start new conversations with agents.
Best attempt at an in-place migration, though there are some noticeable differences.
Also adds view for chat that are not being used, but in experimental phase.
Previously the text_to_image helper would only trigger the image
generation flow if OpenAI client was setup. This is not required
anymore as offline chat model + sd3 API works. So remove that check
Khoj will find and display notes similar to the current entry in the side pane when
1. find similar is open in side pane and
2. cursor has moved to a new entry
### Major
- Find similar notes to current note at cursor automatically in background
- Only show headings of search result and increase default results count
### Minor
- Pass absolute path of file to index from khoj.el emacs client
- Update help message to only show the smaller set of new keybindings
- Fix edge cases in loading some chat sessions
To improve the developer experience for front-end development, we're migrating to Next.js. In order to do this migration page-by-page, we're using static site generation via Next.js. This also helps us avoid making cross site requests from front-end to back-end for the time being, while giving a ramp to separating out server and client if needed for scale down the road.
Dev instructions for using the next.js setup are in the added README.
This adds scaffolding for including the built files in the python package as well as the docker images. Docker setup has been tested locally. In order to verify the build is working as expected, we can navigate to the {khoj_host}:42110/experimental and verify that the experiment page comes up.
This setup works with serving static files included in the src/interface/web folder from the Django app. The key bit for understanding the setup is in the yarn export command in package.json.
* Enable speech to text responses in khoj chat
- Current issue: reads out all the markdown formatting, plus waits for the whole result to be streamed before playing it
* Extract content from markdown-formatted text
* Add a loader for while you're waiting for Khoj's response
* Add user configuration option for chat model options, allow server side configuration for option list
* Join up APIs, views, admin pages to allow configuring custom voice models
When create new conversation session, automatically request query. As
that is expected next action after creating new session
Pass session-id to khoj-chat to allow reuse from
create-new-conversation func
When delete conversation session, do not call load chat session.
Unnecessary action.
Use thread-last to improve code flow in new, delete conversation funcs
* Add magic link email sign-in option
* Adding backend routes and model changes to keep state of email verification code and status
* Test and fix end to end email verification flow
* Add documentation for how to use the magic link sign-in when self-hosting Khoj
* Add magic link sign in to public conversation page
Previously the cursor would move to the Khoj side pane on opening it.
This would break user's flow, especially when find similar triggers
automatically
New behavior maintains smoother update of auto find similar without
disrupting user browsing
Previously it would show complete result body this would make the
result width variable and hard to track all the returned results
Showing just heading makes it easier to track
- Call find similar on current element if point has moved to new
element
- Delete the first result from find-similar search results as that'll
be the current note (which is trivially most similar to itself)
- Determine find-similar based text formating at the rendering layer
rather than at the top level find-similar func
* Uses entire file text and summarizer model to generate document summary.
* Uses the contents of the user's query to create a tailored summary.
* Integrates with File Filters #788 for a better UX.
To add multiple allowed Khoj domains pass them as a comma separated
list of domains via the KHOJ_DOMAIN environment variable
Resolve comment in issue #662
Given img src enforcement via CSP required loosening. Soft enforce it
via a regex replace of img HTML elements if the src isn't from the
whitelisted set of source prefixes.
Currently allowed source prefixes are
- app: for local images
- data: for inline generated images
- https://generated.khoj.dev: for cloud generated images
- Create and use a function to convert markdown to sanitized html
- Remove unused Latex delimiter handling as Katex isn't used in
Khoj chat on Obsidian
- Open Khoj in Emacs Side pane
Open Khoj chat, search in right pane to allow for ambient engagement
- Improve Khoj Chat
- Show online references used for chat
- Make chat API call async to not block user interactions
- Fix loading chat history, references in khoj.el chat buffer
- Improve Khoj Search, Find Similar functions
- Make calls to Khoj search API async to not block user interactions
- Support Conversation Sessions
- Create transient menu to open, create, delete conversation sessions from the Khoj Emacs client
- C-x o to switch to search org content conflicts with switch buffer shortkey
This is more apparent in the async search scenario as it prevents
perform other actions while async search is in progress
- Also switching content type wouldn't scale to all the content types
Khoj will support without causing more conflicting keybinding
Khoj side pane occupies a vertically split bottom right side pane.
If the bottom right window is not a vertical split, create a new
vertical split pane for khoj, otherwise reuse the existing window
- Fix getting file filters for not found conversations
- Allow iamge rendering in automation emails
- Fix nearest 15th minute calculation in automations creation
* added support for uploading multiple files at a time.
* optimized multiple file upload to use a batch upload
* allowing files to upload even if there is one unsupported file
See the currently active window in context while doing chat, search
or find similar operations in a side pane.
This is similar to how we've moved Khoj on Obsidian into the side pane
as well
# Major
- Disambiguate Text output mode to disambiguate from Default data source lookup
- Fix showing headings in intermediate step in generating chat response
- Remove "Path" prefix from org ancestor heading in compiled entry
# Minor
- Fix OpenAI chat actor, director unit tests
* Add language-specific syntax highlighting via highlight.js
- Add highlight.js to our assets CDN for fast load and compliance with the CSP
- See other stylesheets options here: https://cdnjs.com/libraries/highlight.js
* Bonus: set min-height to prevent increasing length of the sessions pane
* Fix references rendering and add highlight.js in public conversation
* Fix multilingual font rendering; fallback to an Arabic language font which contains more Asian characters. Close#756
* Tune font-sizes and styling to accomodate new fonts with old sizing
- Move connection-status styling out from inline html into css block
- Remove start typing chat-input height jitter
- align new-conversation button, text
- use relative font sizes instead of absolute font sizes in most places
---------
Co-authored-by: Debanjum Singh Solanky <debanjum@gmail.com>
* UI update for file filtered conversations
* Interactive file menu #UI to add/remove files on each conversation as references.
* Backend changes implemented to load selected file filters from a conversation into the querying process.
---------
Co-authored-by: sabaimran <narmiabas@gmail.com>
Previously if default output was selected by Khoj, we'd end up doing
an documents search as well, even when Khoj selected internet or
general data source to lookup.
This update disambiguates the default information mode from the text
output mode. To avoid doing documents search when not deemed necessary
by Khoj
Prevent XSS attacks by enforcing Content-Security-Policy (CSP) in apps.
Do not allow loading images, other assets from untrusted domains.
- Only allow loading assets from trusted domains
like 'self', khoj.dev, ipapi for geolocation, google (fonts, img)
- images from khoj domain, google (for profile pic)
- assets from khoj domain
- Do not allow iframe src
- Allow unsafe-inline script and styles for now as markdown-it escapes html
in user, khoj chat
- Add hostURL to CSP of the Desktop, Obsidian apps
Given web client is served by khoj server, it doesn't need to
explicitly allow for khoj.dev domain. So if user self-hosting, it'll
automatically allow the domain in the CSP (via 'self')
Whereas the Obsidian, Desktop clients allow configure the server URL.
Note *switching server URL breaks CSP until app is reloaded*
* The command menu (triggered by "/") now has a clickable list of possible commands, that automatically fill into the chat when pressed.
* The `/help` command now searches `khoj.dev` pages to provide useful assistance to the user.
---------
Co-authored-by: raghavt3 <raghavt3@illinois.edu>
Co-authored-by: sabaimran <65192171+sabaimran@users.noreply.github.com>
### Details
- **Chat with Khoj from right pane on Obsidian**
- Modal was too ephemeral, couldn't have it open for reference, quick jump to Khoj chat
- **Stream intermediate steps taken by Khoj** for generating response to the chat pane
Gives more transparency into Khoj 'thinking' process, e.g internet, notes searches performed, documents read etc.
The feedback allows us to tune our messages to elicit better responses by Khoj
- Add ability to **copy message to clipboard, paste chat messages directly into current file**
- Jump to **Search**, **Find Similar** functions from navigation bar on the Khoj Obsidian side pane
- Improve spacing, use consistent colors in chat message references and buttons
Resolves#789, #754
* Updating the API / UI to support sharing of automations
* Allow people to see the automations even when not logged in, and add an overlay effect
* Handle unauthenticated users taking actions
* Support showing pre-filled automation details on the config automations page
* Redirect user to login if they try to add an automation while unauthenticated
- Dedupe the code to add action buttons to chat messages
- Update the renderIncrementalMessage function to also add the action
buttons to newly generated chat messages by Khoj
* Disable automation recurrence at minute level frequency
* Set a max lifetime for django's connections to the db
* Disable any automation that has a non-numeric first digit (i.e., recuring on the minute level)
* Re-enable automations
---------
Co-authored-by: sabaimran <narmiabas@gmail.com>
Previously clicking inline links would open the URL directly in the
Desktop app. This was strange and it didn't provide any way to go back
to Khoj desktop app UI from the opened link
- Don't propagate max_tokens to the openai chat completion method. the max for the newer models is fixed at 4096 max output. The token limit is just used for input
* Add support for chatting with Anthropic's suite of models
- Had to use a custom class because there was enough nuance with how the anthropic SDK works that it would be better to simply separate out the logic. The extract questions flow needed modification of the system prompt in order to work as intended with the haiku model
- Pass file path of reference along with the compiled reference in
list of references returned by chat API converts
- Update the structure of references from list of strings to list of
dictionary (containing 'compiled' and 'file' keys)
- Pull out the compiled reference from the new references data struct
wherever it was is being used
Simplify, reuse, standardize code to render messages with references
in the obsidian, web and desktop clients. Specifically:
- Reuse function to create reference section, dedupe code
- Create reusable function to generate image markdown
- Simplify logic to render message with references
- Setup websocket using Khoj web app as reference.
- Moved the geolocating code to chat view out from the general pane
view
- Use loading spinner from web instead of the thinking emoji
It'll replace any highlighted text with the chat message or if not
text is highlighted, it'll insert the chat message at the last cursor
position in the active file
- Jump to chat, show similar actions from nav menu of Khoj side pane
- Add chat, search icons from web, desktop app
- Use lucide icon for find similar (for now)
- Match proportions of find similar icon to khoj other icons via css, js
- Use KhojPaneView abstract class to allow reuse of common functionality like
- Creating the nav bar header in side pane views
- Loading geo-location data for chat context
This should make creating new views easier
* Update suggested automations
* add a schedule picker when creating an automation
* Create a new conversation in flow of the automation scheduling in order to send a preview and deliver more consistent results
* Start adding in scaffolding to manually trigger a test job for an automation
* Add support for manually triggering automations for testing
* Schedule automation asynchronously
* Update styling of the preview button
* Improve admin lookup experience and prevent jobs from being scheduled to run every minute of everyday
* Ignore mypy issues on job info short description
### Description and Rationale for Changes
This feature includes thumbs up and thumbs down buttons on Khoj's chat responses that provide automated feedback. When a thumbs up/down button is clicked, the code sends an email to team@khoj.dev with the following:
* user query
* khoj's response
* whether the sentiment of the user was good or bad.
This is critical in improving Khoj's nondeterministic LLM model for a better user experience.
### List of Changes
* new endpoint in `api_chat.py` (/feedback) that can be used to trigger mail sending).
* thumbs up and thumbs down buttons implemented in `chat.html`
* new function in `routers/email.py` to handle feedback email sending via resend
* `feedback.html` template for a formatted email with the feedback.
---------
Co-authored-by: mythicalcow <mythicalcow@linux.myguest.virtualbox.org>
Co-authored-by: sabaimran <narmiabas@gmail.com>
* Improve the automations UX
- Add suggested jobs to elimiinate some of the cold start problem
- Make each of the tasks cards that are clickable/editable
* Hide suggested automations that have already been added
* Add a footer and reapply styling when a save action is taken on a card
- Allows having it open on the side as you traverse your Obsidian notes
- Allow faster time to response, having responses visible for context
- Enables ambient interactions
* Make conversations optionally shareable
- Shared conversations are viewable by anyone, without a login wall
- Can share a conversation from the three dot menu
- Add a new model for Public Conversation
- The rationale for a separate model is that public and private conversations have different assumptions. Separating them reduces some of the code specificity on our server-side code and allows us for easier interpretation and stricter security. Separating the data model makes it harder to accidentally view something that was meant to be private
- Add a new, read-only view for public conversations
- Add parsing logic for LaTeX-format math equations in the web chat
- Add placeholder delimiters when converting the markdown to HTML in order to avoid removing the escaped characters
- Add the `<!DOCTYPE html>` specification to the page
- Pass user and calling_url to the scheduled chat too when modifying
params of automation
- Update to use user timezone even when update job via API
- Move timezone string to timezone object calculation into the
schedule automation method
- Previously it was a section in the settings page. Move it to
independent, top-level page to improve visibility of feature
- Calculate crontime from natural language on web client before
sending it to to server for saving new/updated schedule to disk.
- Avoids round-trip of call to chat model
- Convert POST /api/automation API endpoint into a direct request for
automation with query_to_run, subject and schedule provided via the
automation page. This allows more granular control to create automation
- Make the POST automations endpoint more robust; runs validation
checks, normalizes parameters
- Make the config page content use the same top level 3-column layout
as the khoj-header-wrapper
This ensures the content is aligned with heading pane width
- Let cards and other settings sections scale to the width of their
grid element. This utilizes more of the screen space and does it
consistently across the different settings pages
- Create new POST API endpoint to create automations
- Use it in the settings page on the web interface to create
new automations
This simplified managing automations from the setting page by allowing
both delete and create from the same page
- Render crontime string in natural language in message & settings UI
- Show more fields in tasks web config UI
- Add link to the tasks settings page in task scheduled chat response
- Improve task variables names
Rename executing_query to query_to_run. scheduling_query to
scheduling_request
- Make timezone aware scheduling programmatic, instead of asking the
chat model to do the conversion. This removes the need for
scratchpad and may let smaller models handle the task as well
- Make chat model infer subject for email. This should make the
notification email more readable
- Improve email by using subject in email subject, task heading. Move
query to email final paragraph, which is where task metadata should
go
- Using inferred_query directly was brittle (like previous job id)
- And process lock id had a limited size, so wouldn't work for larger
inferred query strings
- Pass timezone string from ipapi to khoj via clients
- Pass this data from web, desktop and obsidian clients to server
- Use user tz to render next run time of scheduled task in user tz
This takes the load of the task scheduling chat actor / prompt from
having to artifically differentiate query to create scheduled task
from a scheduled task run.
- The task scheduling actor was having trouble calculating the
timezone. Giving the actor a scratchpad to improve correctness by
thinking step by step
- Add more examples to reduce chances of the inferred query looping to
create another reminder instead of running the query and sharing
results with user
- Improve task scheduling chat actor test with more tests and
by ensuring unexpected words not present in response
There's a difference between running a scheduled task and notifying
the user about the results of running the scheduled task.
Decide to notify the user only when the results of running the
scheduled task satisfy the user's requirements.
Use sync version of send_message_to_model_wrapper for scheduled tasks
- Store scheduled job state in Postgres so job schedules persist
across app restarts
- Use Process Locks to only allow single worker to process a given job
type. This prevents duplicating job runs across all workers
- Detect when user intends to schedule a task, aka reminder
Add new output mode: reminder. Add example of selecting the reminder
output mode
- Extract schedule time (as cron timestring) and inferred query to run
from user message
- Use APScheduler to call chat with inferred query at scheduled time
- Handle reminder scheduling from both websocket and http chat requests
- Support constructing scheduled task using chat history as context
Pass chat history to scheduled query generator for improved context
for scheduled task generation
Previously the make delete API response failed, after deleting token.
Required a page refresh to see that the API token was actually gone.
This was happening because the response type of the delete token API
endpoint isn't a string, so it failed FastAPI response validation
checks.
- Allow self-hosted users to customize their open ai base url. This allows you to easily use a proxy service and extend support for other models.
- This also includes a migration that associates any existing openai chat model configuration with an openai processor configuration
- Make changing model a paid/subscriber feature
- Removes usage of langchain's OpenAI wrapper for better control over parsing input/output
- Allow passing completion args through completion_with_backoff
- Pass model_kwargs in a separate arg to simplify this
- Pass model in `model_name' kwarg from the send_message_to_model func
`model_name' kwarg is used by langchain, not `model' kwarg
- Make valid file extension checking case insensitive on Desktop app
- Skip indexing non-existent folders on Desktop app
- Pass auth headers to fix lazy load of chat messages on Desktop app
- Set chat-message height to height of content in web, desktop
Previous cross-encoder model was a few years old, newer models should
have improved in quality. Model size increases by 50% compared to
previous for better performance, at least on benchmarks
Most newer, better embeddings models add a query, docs prefix when
encoding. Previously Khoj admins couldn't configure these, so it
wasn't possible to use these newer models.
This change allows configuring the kwargs passed to the query, docs
encoders by updating the search config in the database.
Improve tool, online search, webpage links, docs search chat actor
prompts. Ensure works with hermes-2-pro and llama-3.
Be more specific about generating JSON and not saying anything else.
- Improve extract question prompts to explicitly request JSON list
- Use llama-3 chat format if HF repo_id mentions llama-3. The
llama-cpp-python logic for detecting when to use llama-3 chat format
isn't robust enough currently
