* Use AR model for embeddings features
* endpoints
* Embeddings CRUD UI
* Add presets. Hide a couple more settings
* system specs
* Seed embedding definition from old settings
* Generate search bit index on the fly. cleanup orphaned data
* support for seeded models
* Fix run test for new embedding
* fix selected model not set correctly
This adds registration and last known IP information and email to scanning context.
This provides another hint for spam scanner about possible malicious users.
For example registered in India, replying from Australia or
email is clearly a throwaway email address.
To quickly select backfill candidates without comparing SHAs, we compare the last summarized post to the topic's highest_post_number. However, hiding or deleting a post and adding a small action will update this column, causing the job to stall and re-generate the same summary repeatedly until someone posts a regular reply. On top of this, this is not always true for topics with `best_replies`, as this last reply isn't necessarily included.
Since this is not evident at first glance and each summarization strategy picks its targets differently, I'm opting to simplify the backfill logic and how we track potential candidates.
The first step is dropping `content_range`, which serves no purpose and it's there because summary caching was supposed to work differently at the beginning. So instead, I'm replacing it with a column called `highest_target_number`, which tracks `highest_post_number` for topics and could track other things like channel's `message_count` in the future.
Now that we have this column when selecting every potential backfill candidate, we'll check if the summary is truly outdated by comparing the SHAs, and if it's not, we just update the column and move on
When enabling spam scanner it there may be old unscanned posts
this can create a risky situation where spam scanner operates
on legit posts during false positives
To keep this a lot safer we no longer try to hide old stuff by
the spammers.
Adds a comprehensive quota management system for LLM models that allows:
- Setting per-group (applied per user in the group) token and usage limits with configurable durations
- Tracking and enforcing token/usage limits across user groups
- Quota reset periods (hourly, daily, weekly, or custom)
- Admin UI for managing quotas with real-time updates
This system provides granular control over LLM API usage by allowing admins
to define limits on both total tokens and number of requests per group.
Supports multiple concurrent quotas per model and automatically handles
quota resets.
Co-authored-by: Keegan George <kgeorge13@gmail.com>
Disabling streaming is required for models such o1 that do not have streaming
enabled yet
It is good to carry this feature around in case various apis decide not to support streaming endpoints and Discourse AI can continue to work just as it did before.
Also: fixes issue where sharing artifacts would miss viewport leading to tiny artifacts on mobile
This update adds some structure for handling errors in the spam config while also handling a specific error related to the spam scanning user not being an admin account.
* FEATURE: smart date support for AI helper
This feature allows conversion of human typed in dates and times
to smart "Discourse" timezone friendly dates.
* fix specs and lint
* lint
* address feedback
* add specs
This PR fixes an issue where LLM enumerator would error out when `SiteSetting.ai_spam_detection = true` but there was no `AiModerationSetting.spam` present.
Typically, we add an `LlmDependencyValidator` for the setting itself, however, since Spam is unique in that it has it's model set in `AiModerationSetting` instead of a `SiteSetting`, we'll add a simple check here to prevent erroring out.
Previously we had the behaviour for model settings so that when you try and set a model, it runs a test and returns an error if it can't run the test successfully. The error then prevents you from setting the site setting.
This results in some issues when we try and automate things. This PR updates that so that the test runs and discreetly logs the changes, but doesn't prevent the setting from being set. Instead we rely on "run test" in the LLM config along with ProblemChecks to catch issues.
Previous version was prone to the bug:
https://github.com/ruby-concurrency/concurrent-ruby/issues/1075
This is particularly bad cause we could have a DB connection
attached to the thread and we never clear it up, so after N hours
this could start exhibiting weird connection issues.
- Add spam_score_type to AiSpamSerializer for better integration with reviewables.
- Introduce a custom filter for detecting AI spam false negatives in moderation workflows.
- Refactor spam report generation to improve identification of false negatives.
- Add tests to verify the custom filter and its behavior.
- Introduce links for all spam counts in report
This commit adds an "unavailable" state for the AI semantic search toggle. Currently the AI toggle disappears when the sort by is anything but Relevance which makes the UI confusing for users looking for AI results. This should help!
In a previous refactor, we moved the responsibility of querying and storing embeddings into the `Schema` class. Now, it's time for embedding generation.
The motivation behind these changes is to isolate vector characteristics in simple objects to later replace them with a DB-backed version, similar to what we did with LLM configs.
* REFACTOR: A Simpler way of interacting with embeddings' tables.
This change adds a new abstraction called `Schema`, which acts as a repository that supports the same DB features `VectorRepresentation::Base` has, with the exception that removes the need to have duplicated methods per embeddings table.
It is also a bit more flexible when performing a similarity search because you can pass it a block that gives you access to the builder, allowing you to add multiple joins/where conditions.
This introduces a comprehensive spam detection system that uses LLM models
to automatically identify and flag potential spam posts. The system is
designed to be both powerful and configurable while preventing false positives.
Key Features:
* Automatically scans first 3 posts from new users (TL0/TL1)
* Creates dedicated AI flagging user to distinguish from system flags
* Tracks false positives/negatives for quality monitoring
* Supports custom instructions to fine-tune detection
* Includes test interface for trying detection on any post
Technical Implementation:
* New database tables:
- ai_spam_logs: Stores scan history and results
- ai_moderation_settings: Stores LLM config and custom instructions
* Rate limiting and safeguards:
- Minimum 10-minute delay between rescans
- Only scans significant edits (>10 char difference)
- Maximum 3 scans per post
- 24-hour maximum age for scannable posts
* Admin UI features:
- Real-time testing capabilities
- 7-day statistics dashboard
- Configurable LLM model selection
- Custom instruction support
Security and Performance:
* Respects trust levels - only scans TL0/TL1 users
* Skips private messages entirely
* Stops scanning users after 3 successful public posts
* Includes comprehensive test coverage
* Maintains audit log of all scan attempts
---------
Co-authored-by: Keegan George <kgeorge13@gmail.com>
Co-authored-by: Martin Brennan <martin@discourse.org>
* UX: Improve rough edges of AI usage page
* Ensure all text uses I18n
* Change from <button> usage to <DButton>
* Use <AdminConfigAreaCard> in place of custom card styles
* Format numbers nicely using our number format helper,
show full values on hover using title attr
* Ensure 0 is always shown for counters, instead of being blank
* FEATURE: Load usage data after page load
Use ConditionalLoadingSpinner to hide load of usage
data, this prevents us hanging on page load with a white
screen.
* UX: Split users table, and add empty placeholders and page subheader
* DEV: Test fix
Instead of a stacked chart showing a separate series for positive and negative, this PR introduces a simplification to the overall sentiment dashboard. It comprises the sentiment into a single series of the difference between `positive - negative` instead. This should allow for the data to be more easy to scan and look for trends
Instead of a stacked chart showing a separate series for positive and negative, this PR introduces a simplification to the overall sentiment dashboard. It comprises the sentiment into a single series of the difference between `positive - negative` instead. This should allow for the data to be more easy to scan and look for trends.
* FEATURE: first class support for OpenRouter
This new implementation supports picking quantization and provider pref
Also:
- Improve logging for summary generation
- Improve error message when contacting LLMs fails
* Better support for full screen artifacts on iPad
Support back button to close full screen
Refactor dialect selection and add Nova API support
Change dialect selection to use llm_model object instead of just provider name
Add support for Amazon Bedrock's Nova API with native tools
Implement Nova-specific message processing and formatting
Update specs for Nova and AWS Bedrock endpoints
Enhance AWS Bedrock support to handle Nova models
Fix Gemini beta API detection logic
* FEATURE: Backfill posts sentiment.
It adds a scheduled job to backfill posts' sentiment, similar to our existing rake task, but with two settings to control the batch size and posts' max-age.
* Make sure model_name order is consistent.
This PR fixes an issue where the tag suggester for edit title topic area was suggesting tags that are already assigned on a post. It also updates the amount of suggested tags to 7 so that there is still a decent amount of tags suggested when tags are already assigned.
Add support for versioned artifacts with improved diff handling
* Add versioned artifacts support allowing artifacts to be updated and tracked
- New `ai_artifact_versions` table to store version history
- Support for updating artifacts through a new `UpdateArtifact` tool
- Add version-aware artifact rendering in posts
- Include change descriptions for version tracking
* Enhance artifact rendering and security
- Add support for module-type scripts and external JS dependencies
- Expand CSP to allow trusted CDN sources (unpkg, cdnjs, jsdelivr, googleapis)
- Improve JavaScript handling in artifacts
* Implement robust diff handling system (this is dormant but ready to use once LLMs catch up)
- Add new DiffUtils module for applying changes to artifacts
- Support for unified diff format with multiple hunks
- Intelligent handling of whitespace and line endings
- Comprehensive error handling for diff operations
* Update routes and UI components
- Add versioned artifact routes
- Update markdown processing for versioned artifacts
Also
- Tweaks summary prompt
- Improves upload support in custom tool to also provide urls
- Added a new admin interface to track AI usage metrics, including tokens, features, and models.
- Introduced a new route `/admin/plugins/discourse-ai/ai-usage` and supporting API endpoint in `AiUsageController`.
- Implemented `AiUsageSerializer` for structuring AI usage data.
- Integrated CSS stylings for charts and tables under `stylesheets/modules/llms/common/usage.scss`.
- Enhanced backend with `AiApiAuditLog` model changes: added `cached_tokens` column (implemented with OpenAI for now) with relevant DB migration and indexing.
- Created `Report` module for efficient aggregation and filtering of AI usage metrics.
- Updated AI Bot title generation logic to log correctly to user vs bot
- Extended test coverage for the new tracking features, ensuring data consistency and access controls.
This change adds a simpler class for sentiment classification, replacing the soon-to-be removed `Classificator` hierarchy. Additionally, it adds a method for classifying concurrently, speeding up the backfill rake task.
This PR updates the logic for the location map so it permits only the desired prompts through to the composer/post menu. Anything else won't be shown by default.
This PR also adds relevant tests to prevent regression.
### 🔍 Overview
With the recent changes to allow DiscourseAi in the translator plugin, `detect_text_locale` was needed as a CompletionPrompt. However, it is leaking into composer/post helper menus. This PR ensures we don't not show it in those menus.
We are adding a new method for generating and storing embeddings in bulk, which relies on `Concurrent::Promises::Future`. Generating an embedding consists of three steps:
Prepare text
HTTP call to retrieve the vector
Save to DB.
Each one is independently executed on whatever thread the pool gives us.
We are bringing a custom thread pool instead of the global executor since we want control over how many threads we spawn to limit concurrency. We also avoid firing thousands of HTTP requests when working with large batches.
* FEATURE: allow mentioning an LLM mid conversation to switch
This is a edgecase feature that allow you to start a conversation
in a PM with LLM1 and then use LLM2 to evaluation or continue
the conversation
* FEATURE: allow auto silencing of spam accounts
New rule can also allow for silencing an account automatically
This can prevent spammers from creating additional posts.
Two changes worth mentioning:
`#instance` returns a fully configured embedding endpoint ready to use.
All endpoints respond to the same method and have the same signature - `perform!(text)`
This makes it easier to reuse them when generating embeddings in bulk.
* PERF: Preload only gists when including summaries in topic list
* Add unique index on summaries and dedup existing records
* Make hot topics batch size setting hidden
The `topic_query_create_list_topics` modifier we append was always meant to avoid an N+1 situation when serializing gists. However, I tried to be too smart and only preload these, which resulted in some topics with *only* regular summaries getting removed from the list. This issue became apparent now we are adding gists to other lists besides hot.
Let's simplify the preloading, which still solves the N+1 issue, and let the serializer get the needed summary.
* FEATURE: Make emotion /filter ordering match the dashboard table
This change makes the /filter endpoint use the same criteria we use
in the dashboard table for emotion, so it is not confusing for users.
It means that only posts made in the period with the emotion shall be
shown in the /filter, and the order is simply a count of posts that
match the emotion in the period.
It also uses a trick to extract the filter period, and apply it to
the CTE clause that calculates post emotion count on the period, making
it a bit more efficient. Downside is that /filter filters are evaluated
from left to right, so it will only get the speed-up if the emotion
order is last. As we do this on the dashboard table, it should cover
most uses of the ordering, kicking the need for materialized views
down the road.
* Remove zero score in filter
* add table tooltip
* lint
1. Keep source in a "details" block after rendered so it does
not overwhelm users
2. Ensure artifacts are never indexed by robots
3. Cache break our CSS that changed recently
* FIX: Misc fixes for sentiment in the admin dashboard
- Fixes missing filters for the main graph
- Fixes previous 30 days trend in emotion table
Also moves links to individual cells in emotion table, so admins can
drill down to the specific time period on their reports.
* lints
We use `includes` instead of `joins` because we want to eager-load summaries, avoiding an extra query when summarizing. However, Rails will complain unless you explicitly inform them you plan to use that inside a `WHERE` clause.
The topic query is used differently, and we can't assume the modifier will always receive an AR relation. Let's scope it to `Discourse#filters` instead of most lists.
This is a significant PR that introduces AI Artifacts functionality to the discourse-ai plugin along with several other improvements. Here are the key changes:
1. AI Artifacts System:
- Adds a new `AiArtifact` model and database migration
- Allows creation of web artifacts with HTML, CSS, and JavaScript content
- Introduces security settings (`strict`, `lax`, `disabled`) for controlling artifact execution
- Implements artifact rendering in iframes with sandbox protection
- New `CreateArtifact` tool for AI to generate interactive content
2. Tool System Improvements:
- Adds support for partial tool calls, allowing incremental updates during generation
- Better handling of tool call states and progress tracking
- Improved XML tool processing with CDATA support
- Fixes for tool parameter handling and duplicate invocations
3. LLM Provider Updates:
- Updates for Anthropic Claude models with correct token limits
- Adds support for native/XML tool modes in Gemini integration
- Adds new model configurations including Llama 3.1 models
- Improvements to streaming response handling
4. UI Enhancements:
- New artifact viewer component with expand/collapse functionality
- Security controls for artifact execution (click-to-run in strict mode)
- Improved dialog and response handling
- Better error management for tool execution
5. Security Improvements:
- Sandbox controls for artifact execution
- Public/private artifact sharing controls
- Security settings to control artifact behavior
- CSP and frame-options handling for artifacts
6. Technical Improvements:
- Better post streaming implementation
- Improved error handling in completions
- Better memory management for partial tool calls
- Enhanced testing coverage
7. Configuration:
- New site settings for artifact security
- Extended LLM model configurations
- Additional tool configuration options
This PR significantly enhances the plugin's capabilities for generating and displaying interactive content while maintaining security and providing flexible configuration options for administrators.
Implement streaming tool call implementation for Anthropic and Open AI.
When calling:
llm.generate(..., partial_tool_calls: true) do ...
Partials may contain ToolCall instances with partial: true, These tool calls are partially populated with json partially parsed.
So for example when performing a search you may get:
ToolCall(..., {search: "hello" })
ToolCall(..., {search: "hello world" })
The library used to parse json is:
https://github.com/dgraham/json-stream
We use a fork cause we need access to the internal buffer.
This prepares internals to perform partial tool calls, but does not implement it yet.
This re-implements tool support in DiscourseAi::Completions::Llm #generate
Previously tool support was always returned via XML and it would be the responsibility of the caller to parse XML
New implementation has the endpoints return ToolCall objects.
Additionally this simplifies the Llm endpoint interface and gives it more clarity. Llms must implement
decode, decode_chunk (for streaming)
It is the implementers responsibility to figure out how to decode chunks, base no longer implements. To make this easy we ship a flexible json decoder which is easy to wire up.
Also (new)
Better debugging for PMs, we now have a next / previous button to see all the Llm messages associated with a PM
Token accounting is fixed for vllm (we were not correctly counting tokens)
This PR fixes an issue where clicking to regenerate a summary was still showing the cached summary. To resolve this we call resetSummary() to reset all the summarization related properties before creating a new request.
This change introduces a job to summarize topics and cache the results automatically. We provide a setting to control how many topics we'll backfill per hour and what the topic's minimum word count is to qualify.
We'll prioritize topics without summary over outdated ones.
Fixes encoding of params on LLM function calls.
Previously we would improperly return results if a function parameter returned an HTML tag.
Additionally adds some missing HTTP verbs to tool calls.
The custom field "discourse_ai_bypass_ai_reply" was added so
we can signal the post created hook to bypass replying even
if it thinks it should.
Otherwise there are cases where we double answer user questions
leading to much confusion.
This also slightly refactors code making the controller smaller
The new `/admin/plugins/discourse-ai/ai-personas/stream-reply.json` was added.
This endpoint streams data direct from a persona and can be used
to access a persona from remote systems leaving a paper trail in
PMs about the conversation that happened
This endpoint is only accessible to admins.
---------
Co-authored-by: Gabriel Grubba <70247653+Grubba27@users.noreply.github.com>
Co-authored-by: Keegan George <kgeorge13@gmail.com>
* FEATURE: Fast-track gist regeneration when a hot topic gets a new post
* DEV: Introduce an upsert-like summarize
* FIX: Only enqueue fast-track gist for hot hot hot topics
---------
Co-authored-by: Rafael Silva <xfalcox@gmail.com>
* FIX/REFACTOR: FoldContent revamp
We hit a snag with our hot topic gist strategy: the regex we used to split the content didn't work, so we cannot send the original post separately. This was important for letting the model focus on what's new in the topic.
The algorithm doesn’t give us full control over how prompts are written, and figuring out how to format the content isn't straightforward. This means we're having to use more complicated workarounds, like regex.
To tackle this, I'm suggesting we simplify the approach a bit. Let's focus on summarizing as much as we can upfront, then gradually add new content until there's nothing left to summarize.
Also, the "extend" part is mostly for models with small context windows, which shouldn't pose a problem 99% of the time with the content volume we're dealing with.
* Fix fold docs
* Use #shift instead of #pop to get the first elem, not the last
This changeset contains 4 fixes:
1. We were allowing running tests on unsaved tools,
this is problematic cause uploads are not yet associated or indexed
leading to confusing results. We now only show the test button when
tool is saved.
2. We were not properly scoping rag document fragements, this
meant that personas and ai tools could get results from other
unrelated tools, just to be filtered out later
3. index.search showed options as "optional" but implementation
required the second option
4. When testing tools searching through document fragments was
not working at all cause we did not properly load the tool
* FIX: Llm selector / forced tools / search tool
This fixes a few issues:
1. When search was not finding any semantic results we would break the tool
2. Gemin / Anthropic models did not implement forced tools previously despite it being an API option
3. Mechanics around displaying llm selector were not right. If you disabled LLM selector server side persona PM did not work correctly.
4. Disabling native tools for anthropic model moved out of a site setting. This deliberately does not migrate cause this feature is really rare to need now, people who had it set probably did not need it.
5. Updates anthropic model names to latest release
* linting
* fix a couple of tests I missed
* clean up conditional
A new feature_context json column was added to ai_api_audit_logs
This allows us to store rich json like context on any LLM request
made.
This new field now stores automation id and name.
Additionally allows llm_triage to specify maximum number of tokens
This means that you can limit the cost of llm triage by scanning only
first N tokens of a post.
This changeset:
1. Corrects some issues with "force_default_llm" not applying
2. Expands the LLM list page to show LLM usage
3. Clarifies better what "enabling a bot" on an llm means (you get it in the selector)
* Display gists in the hot topics list
* Adjust hot topics gist strategy and add a job to generate gists
* Replace setting with a configurable batch size
* Avoid loading summaries for other topic lists
* Tweak gist prompt to focus on latest posts in the context of the OP
* Remove serializer hack and rely on core change from discourse/discourse#29291
* Update lib/summarization/strategies/hot_topic_gists.rb
Co-authored-by: Rafael dos Santos Silva <xfalcox@gmail.com>
---------
Co-authored-by: Rafael dos Santos Silva <xfalcox@gmail.com>
Splits persona permissions so you can allow a persona on:
- chat dms
- personal messages
- topic mentions
- chat channels
(any combination is allowed)
Previously we did not have this flexibility.
Additionally, adds the ability to "tether" a language model to a persona so it will always be used by the persona. This allows people to use a cheaper language model for one group of people and more expensive one for other people
On very large sites, the rare cache misses for Related Topics can take around 200ms, which affects our p99 metric on the topic page. In order to mitigate this impact, we now have several tools at our disposal.
First, one is to migrate the index embedding type from halfvec to bit and change the related topic query to leverage the new bit index by changing the search algorithm from inner product to Hamming distance. This will reduce our index sizes by 90%, severely reducing the impact of embeddings on our storage. By making the related query a bit smarter, we can have zero impact on recall by using the index to over-capture N*2 results, then re-ordering those N*2 using the full halfvec vectors and taking the top N. The expected impact is to go from 200ms to <20ms for cache misses and from a 2.5GB index to a 250MB index on a large site.
Another tool is migrating our index type from IVFFLAT to HNSW, which can increase the cache misses performance even further, eventually putting us in the under 5ms territory.
Co-authored-by: Roman Rizzi <roman@discourse.org>
This introduces another configuration that allows operators to
limit the amount of interactions with forced tool usage.
Forced tools are very handy in initial llm interactions, but as
conversation progresses they can hinder by slowing down stuff
and adding confusion.
This adds chain halting (ability to terminate llm chain in a tool)
and the ability to create uploads in a tool
Together this lets us integrate custom image generators into a
custom tool.
* FEATURE: allows forced LLM tool use
Sometimes we need to force LLMs to use tools, for example in RAG
like use cases we may want to force an unconditional search.
The new framework allows you backend to force tool usage.
Front end commit to follow
* UI for forcing tools now works, but it does not react right
* fix bugs
* fix tests, this is now ready for review
Previous to this change we could flag, but there was no way
to hide content and treat the flag as spam.
We had the option to hide topics, but this is not desirable for
a spam reply.
New option allows triage to hide a post if it is a reply, if the
post happens to be the first post on the topic, the topic will
be hidden.
This allows our users to add the Ollama provider and use it to serve our AI bot (completion/dialect).
In this PR, we introduce:
DiscourseAi::Completions::Dialects::Ollama which would help us translate by utilizing Completions::Endpoint::Ollama
Correct extract_completion_from and partials_from in Endpoints::Ollama
Also
Add tests for Endpoints::Ollama
Introduce ollama_model fabricator
This allows custom tools access to uploads and sophisticated searches using embedding.
It introduces:
- A shared front end for listing and uploading files (shared with personas)
- Backend implementation of index.search function within a custom tool.
Custom tools now may search through uploaded files
function invoke(params) {
return index.search(params.query)
}
This means that RAG implementers now may preload tools with knowledge and have high fidelity over
the search.
The search function support
specifying max results
specifying a subset of files to search (from uploads)
Also
- Improved documentation for tools (when creating a tool a preamble explains all the functionality)
- uploads were a bit finicky, fixed an edge case where the UI would not show them as updated
Caveats
- No streaming, by design
- No tool support (including no XML tools)
- No vision
Open AI will revamt the model and more of these features may
become available.
This solution is a bit hacky for now
Polymorphic RAG means that we will be able to access RAG fragments both from AiPersona and AiCustomTool
In turn this gives us support for richer RAG implementations.
Previously we waited 1 minute before automatically titling PMs
The new change introduces adding a title immediately after the the
llm replies
Prompt was also modified to include the LLM reply in title suggestion.
This helps situation like:
user: tell me a joke
llm: a very funy joke about horses
Then the title would be "A Funny Horse Joke"
Specs already covered some auto title logic, amended to also
catch the new message bus message we have been sending.
* FIX: we were never reindexing old content
Embedding backfill contains logic for searching for old content
change and then backfilling.
Unfortunately it was excluding all topics that had embedding
unconditionally, leading to no backfill ever happening.
This change adds a test and ensures we backfill.
* over select results, this ensures we will be more likely to find
ai results when filtered
This improves the site setting search so it performs a somewhat
fuzzy match.
Previously it did not handle seperators such as "space" and a
term such as "min_post_length" would not find "min_first_post_length"
A more liberal search algorithm makes it easier to the AI to
navigate settings.
* Minor fix, {{and parameter.enum parameter.enum.length}} is non
obviously broken.
If parameter.enum is a tracked array it will return the object
cause embers and helper implementation.
This corrects an issue where enum keeps on selecting itself by
mistake.
This allows callers of embedding based search to bypass hyde.
Hyde will expand the search term using an LLM, but if an LLM is
performing the search we can skip this expansion.
It also introduced some tests for the controller which we did not have
* FEATURE: LLM Triage support for systemless models.
This change adds support for OSS models without support for system messages. LlmTriage's system message field is no longer mandatory. We now send the post contents in a separate user message.
* Models using Ollama can also disable system prompts
New `ai_pm_summarization_allowed_groups` can be used to allow
visibility of the summarization feature on PMs.
This can be useful on forums where a lot of communication happens
inside PMs.
Creating a new model, either manually or from presets, doesn't initialize the `provider_params` object, meaning their custom params won't persist.
Additionally, this change adds some validations for Bedrock params, which are mandatory, and a clear message when a completion fails because we cannot build the URL.
* FIX: Add tool support to open ai compatible dialect and vllm
Automatic tools are in progress in vllm see: https://github.com/vllm-project/vllm/pull/5649
Even when they are supported, initial support will be uneven, only some models have native tool support
notably mistral which has some special tokens for tool support.
After the above PR lands in vllm we will still need to swap to
XML based tools on models without native tool support.
* fix specs
* DEV: Remove old code now that features rely on LlmModels.
* Hide old settings and migrate persona llm overrides
* Remove shadowing special URL + seeding code. Use srv:// prefix instead.
Using RAG fragments can lead to considerably big system messages, which becomes problematic when models have a smaller context window.
Before this change, we only look at the rest of the conversation to make sure we don't surpass the limit, which could lead to two unwanted scenarios when having large system messages:
All other messages are excluded due to size.
The system message already exceeds the limit.
As a result, I'm putting a hard-limit of 60% of available tokens. We don't want to aggresively truncate because if rag fragments are included, the system message contains a lot of context to improve the model response, but we also want to make room for the recent messages in the conversation.
Using assistant role for system produces an error because
they expect alternating roles like user/assistant/user and so on.
Prompts cannot start with the assistant role.
This allows summary to use the new LLM models and migrates of API key based model selection
Claude 3.5 etc... all work now.
---------
Co-authored-by: Roman Rizzi <rizziromanalejandro@gmail.com>
* FIX: Use base64 encoded images in AI Image Caption via LLaVa
This fixed a regression introduced in #646 where we started sending
schemaless URLs for our LLaVa service, which doesn't handle it well.
Moving to base64 encoded images solves:
- The service needing to download images
Now the service running LLaVa doesn't need internet access
- Secure uploads compat
Every image is treated the same, less branching for secure uploads
- Image Size problems
Discourse is now responsible for ensure a max size for images
- Troublesome dev env
Previously to this commit you would need a dev env that was internet
acessible to use llava image captions
Introduces custom AI tools functionality.
1. Why it was added:
The PR adds the ability to create, manage, and use custom AI tools within the Discourse AI system. This feature allows for more flexibility and extensibility in the AI capabilities of the platform.
2. What it does:
- Introduces a new `AiTool` model for storing custom AI tools
- Adds CRUD (Create, Read, Update, Delete) operations for AI tools
- Implements a tool runner system for executing custom tool scripts
- Integrates custom tools with existing AI personas
- Provides a user interface for managing custom tools in the admin panel
3. Possible use cases:
- Creating custom tools for specific tasks or integrations (stock quotes, currency conversion etc...)
- Allowing administrators to add new functionalities to AI assistants without modifying core code
- Implementing domain-specific tools for particular communities or industries
4. Code structure:
The PR introduces several new files and modifies existing ones:
a. Models:
- `app/models/ai_tool.rb`: Defines the AiTool model
- `app/serializers/ai_custom_tool_serializer.rb`: Serializer for AI tools
b. Controllers:
- `app/controllers/discourse_ai/admin/ai_tools_controller.rb`: Handles CRUD operations for AI tools
c. Views and Components:
- New Ember.js components for tool management in the admin interface
- Updates to existing AI persona management components to support custom tools
d. Core functionality:
- `lib/ai_bot/tool_runner.rb`: Implements the custom tool execution system
- `lib/ai_bot/tools/custom.rb`: Defines the custom tool class
e. Routes and configurations:
- Updates to route configurations to include new AI tool management pages
f. Migrations:
- `db/migrate/20240618080148_create_ai_tools.rb`: Creates the ai_tools table
g. Tests:
- New test files for AI tool functionality and integration
The PR integrates the custom tools system with the existing AI persona framework, allowing personas to use both built-in and custom tools. It also includes safety measures such as timeouts and HTTP request limits to prevent misuse of custom tools.
Overall, this PR significantly enhances the flexibility and extensibility of the Discourse AI system by allowing administrators to create and manage custom AI tools tailored to their specific needs.
Co-authored-by: Martin Brennan <martin@discourse.org>
Previously, we stored request parameters like the OpenAI organization and Bedrock's access key and region as site settings. This change stores them in the `llm_models` table instead, letting us drop more settings while also becoming more flexible.
* FEATURE: LLM presets for model creation
Previous to this users needed to look up complicated settings
when setting up models.
This introduces and extensible preset system with Google/OpenAI/Anthropic
presets.
This will cover all the most common LLMs, we can always add more as
we go.
Additionally:
- Proper support for Anthropic Claude Sonnet 3.5
- Stop blurring api keys when navigating away - this made it very complex to reuse keys
We no longer support the "provider:model" format in the "ai_helper_model" and
"ai_embeddings_semantic_search_hyde_model" settings. We'll migrate existing
values and work with our new data-driven LLM configs from now on.
Previously read tool only had access to public topics, this allows
access to all topics user has access to, if admin opts for the option
Also
- Fixes VLLM migration
- Display which llms have bot enabled
* DRAFT: Create AI Bot users dynamically and support custom LlmModels
* Get user associated to llm_model
* Track enabled bots with attribute
* Don't store bot username. Minor touches to migrate default values in settings
* Handle scenario where vLLM uses a SRV record
* Made 3.5-turbo-16k the default version so we can remove hack
- Display filtered search correctly, so it is not confusing
- When XML stripping, if a chunk was `<` it would crash
- SQL Helper improved to be better aware of Data Explorer
This is a rather huge refactor with 1 new feature (tool details can
be suppressed)
Previously we use the name "Command" to describe "Tools", this unifies
all the internal language and simplifies the code.
We also amended the persona UI to use less DToggles which aligns
with our design guidelines.
Co-authored-by: Martin Brennan <martin@discourse.org>
Native tools do not work well on Opus.
Chain of Thought prompting means it consumes enormous amounts of
tokens and has poor latency.
This commit introduce and XML stripper to remove various chain of
thought XML islands from anthropic prompts when tools are involved.
This mean Opus native tools is now functions (albeit slowly)
From local testing XML just works better now.
Also fixes enum support in Anthropic native tools
Add native Cohere tool support
- Introduce CohereTools class for tool translation and result processing
- Update Command dialect to integrate with CohereTools
- Modify Cohere endpoint to support passing tools and processing tool calls
- Add spec for testing tool triggering with Cohere endpoint
1. New tool to easily find files (and default branch) in a Github repo
2. Improved read tool with clearer params and larger context
* limit can totally mess up the richness semantic search adds, so include the results unconditionally.
Previoulsy on GPT-4-vision was supported, change introduces support
for Google/Anthropic and new OpenAI models
Additionally this makes vision work properly in dev environments
cause we sent the encoded payload via prompt vs sending urls
This change allows us to delete custom models. It checks if there is no module using them.
It also fixes a bug where the after-create transition wasn't working. While this prevents a model from being saved multiple times, endpoint validations are still needed (will be added in a separate PR).:
This is similar to code interpreter by ChatGPT, except that it uses
JavaScript as the execution engine.
Safeguards were added to ensure memory is constrained and evaluation
times out.
* FEATURE: Set endpoint credentials directly from LlmModel.
Drop Llama2Tokenizer since we no longer use it.
* Allow http for custom LLMs
---------
Co-authored-by: Rafael Silva <xfalcox@gmail.com>
When lazy load categories is enabled, the list of categories does not
have to fetched from the "site.json" endpoint because it is already
returned by "search.json".
This commit reverts commits 5056502 and 3e54697 because iterating over
all pages of categories is not really necessary.
- Introduce new support for GPT4o (automation / bot / summary / helper)
- Properly account for token counts on OpenAI models
- Track feature that was used when generating AI completions
- Remove custom llm support for summarization as we need better interfaces to control registration and de-registration
Roman Rizzi
Sam
Rafael dos Santos Silva
Natalie Tay
Keegan George
Mark VanLandingham
Martin Brennan
David Taylor
Joffrey JAFFEUX
Kris
Hoa Nguyen
PangBo
Loïc Guitaut
Bianca Nenciu