Utf-16 annotation indices#555
Closed
crspeller wants to merge 8 commits into
Closed
Conversation
* Integrate Bifrost LLM gateway library for unified provider access Replace individual LLM provider implementations (OpenAI, Anthropic, Azure, Bedrock, Cohere, Mistral) with the Bifrost gateway library for chat completions. Key changes: - Add new bifrost/ package implementing llm.LanguageModel interface - BifrostLLM adapter handles message conversion, tool calling, and streaming - Maps ServiceConfig to Bifrost provider configuration - Update bots/bots.go to use Bifrost for supported providers - Keep asage provider as fallback for unsupported types - Retain openai package for transcription and embeddings (not covered by Bifrost) - Retain anthropic/bedrock packages for FetchModels and evals This provides a unified API for multiple LLM providers while maintaining backwards compatibility with existing configurations. * Add Bifrost support for embeddings and transcription Extend Bifrost integration to cover additional LLM capabilities: Embeddings: - Add BifrostEmbeddingProvider implementing embeddings.EmbeddingProvider interface - Support CreateEmbedding and BatchCreateEmbeddings via Bifrost gateway - Add "bifrost" provider type to embeddings configuration - Support OpenAI, Azure, Cohere, and Bedrock providers for embeddings Transcription: - Add BifrostTranscriber for audio-to-text conversion via Bifrost - Support VTT format output compatible with existing subtitles parsing - Configurable model and provider selection This provides unified access to embedding and transcription APIs through the Bifrost gateway, allowing users to choose their preferred provider while maintaining a consistent interface. * Remove legacy provider packages in favor of Bifrost This commit completes the migration to using Bifrost as the unified LLM gateway by: - Deleting openai/, anthropic/, bedrock/, and asage/ packages - Removing ASage service type support from configuration - Updating bots.go to use Bifrost exclusively for LLM and transcription - Updating search/embeddings.go to use Bifrost for embedding providers - Fixing type assignment issues in getLLM with interface types - Using service DefaultModel for transcription (fallback to whisper-1) - Cleaning up config.go by removing OpenAI-specific helper functions All LLM providers are now accessed through the Bifrost gateway library, providing a consistent interface across OpenAI, Anthropic, Azure, OpenAI-compatible, and Bedrock providers. * Improve Bifrost configuration to use built-in features - Use NetworkConfig.BaseURL for OpenAI Compatible and other providers that support custom API endpoints (Azure uses AzureKeyConfig instead) - Pass OpenAI OrgID via NetworkConfig.ExtraHeaders - Configure built-in retry logic (2 retries with 1s-10s backoff) - Remove unused httpClient from providerAccount struct (Bifrost manages its own HTTP client internally) * Update Bifrost integration for newer library API * Responses API * Fix CI * Mod tidy * Add reasoning and annotation streaming to Bifrost responses * Restore missing functionality in Bifrost migration - Add auto-run tool resolution wrapper with recursive loop - Honor ReasoningDisabled per-request config - Route native web search through Responses API for all providers - Stream reasoning/thinking events in chat API path - Restore Anthropic thinking budget auto-calculation - Preserve thinking blocks in conversation history - Merge consecutive same-role messages for Anthropic - Fix transcription model to always use whisper-1 - Pass embedding dimensions in requests - Remove unused httpClient parameter from Bifrost * Fix review issues in Bifrost migration * /v1 issue * Fix remaining e2e failures: reasoning summaries, mock model, and embedding URL - Set reasoning.Summary = "auto" for OpenAI Responses API so reasoning summary events are emitted in the stream (matches old SDK behavior) - Add defaultModel to mock service configs (Bifrost requires it) - Pass APIURL for OpenAI embedding provider for consistency - Normalize embedding provider URLs to strip /v1 suffix * Fix some time issues * Fix e2e flake mode * Lint * Annotations fixed * Fix for Anthropic annotations --------- Co-authored-by: Claude <noreply@anthropic.com>
Bring agents-v2 up to date with master, incorporating 8 commits: token accounting, bot config hot-reload, license gating, e2e observability, model constant updates, MCP OAuth fix, docs, and RHS prompt hint removal. Resolved conflicts in bots/bots.go (imports), evals/evals.go (model constants), and accepted deletion of openai/ files superseded by the bifrost migration. Made-with: Cursor
* Add live system-console e2e flow for real provider Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Exercise model dropdown in live e2e flow Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Fix Responses API payload unsupported summary field Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Run live system-console test in real API CI job Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Stabilize live system-console test form selectors Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Make live system-console CI test retry-safe Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Relax model persistence assertion for dropdown labels Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Force clean session before regular-user login Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Support Client4 getPosts in live flow polling Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Avoid strict text collision on mention assertion Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Assert live replies are bot-authored and non-error Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Prefer stable non-haiku model in live flow Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Prefer exact default-model match from dropdown Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Retry live replies when model returns transient error Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Polyfill crypto.randomUUID for insecure Docker test contexts The e2e tests run inside Docker where testcontainers resolves the Mattermost URL to the Docker bridge gateway IP (e.g. 172.17.0.1) instead of localhost. Chromium treats non-localhost HTTP origins as insecure contexts where crypto.randomUUID() is unavailable. The webapp's Add Service / Add Bot buttons call crypto.randomUUID() to generate IDs, which throws in insecure contexts. This polyfill uses crypto.getRandomValues() (available in all contexts) to provide a UUID v4 fallback. Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Use dated Anthropic model name compatible with Bifrost library The Bifrost gateway library (v1.3.11) does not recognize the newer claude-sonnet-4-6 model alias. Use the dated version claude-sonnet-4-20250514 which Bifrost can route properly. Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Strip Bifrost provider prefix from model IDs in config Bifrost's ListModels endpoint returns model IDs prefixed with the provider name (e.g. 'anthropic/claude-sonnet-4-20250514'). These prefixed IDs were being saved to the plugin config, but the chat API expects plain model names since the provider is specified separately. Strip the prefix in FetchModels (for new selections) and in NewFromServiceConfig (for existing saved configs). Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com> * Scope Bifrost provider prefix stripping to FetchModels only * Remove unused --------- Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com>
* Fix bot config changes not taking effect until plugin restart (#508) * Gate channel analysis by enterprise license (#511) Co-authored-by: Cursor Agent <cursoragent@cursor.com> * webapp: upgrade to React 18 dependency stack Co-authored-by: nick.misasi <nick.misasi@mattermost.com> * webapp: fix store typings for redux 5 Co-authored-by: nick.misasi <nick.misasi@mattermost.com> * fix bots compile by importing config package Add the missing config import so config.DeepCopyJSON resolves in EnsureBots, unblocking build, lint, and downstream test jobs in CI. Made-with: Cursor * Fix OpenAI Responses API rejecting reasoning summaries in input and test infra fixes - Remove reasoning summary echoing in convertToResponsesMessages; the Responses API rejects input entries that include a summary field - Add test asserting ResponsesReasoning is nil on input assistant messages - Make Anthropic/OpenAI model names configurable via ANTHROPIC_MODEL/OPENAI_MODEL env vars - Guard mmcontainer logStream writes to avoid 'write after end' race on teardown Made-with: Cursor * Fix bot config changes not taking effect until plugin restart (#508) * webapp: upgrade to React 18 dependency stack Co-authored-by: nick.misasi <nick.misasi@mattermost.com> * webapp: fix store typings for redux 5 Co-authored-by: nick.misasi <nick.misasi@mattermost.com> * Fix OpenAI Responses API rejecting reasoning summaries in input and test infra fixes - Remove reasoning summary echoing in convertToResponsesMessages; the Responses API rejects input entries that include a summary field - Add test asserting ResponsesReasoning is nil on input assistant messages - Make Anthropic/OpenAI model names configurable via ANTHROPIC_MODEL/OPENAI_MODEL env vars - Guard mmcontainer logStream writes to avoid 'write after end' race on teardown Made-with: Cursor * Fix lint: remove unused setEditorText and extra blank line Remove the unused setEditorText function and extra blank line in rhs_new_tab.tsx that were introduced during conflict resolution. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Revert non-React-18 changes to keep branch focused on webapp upgrade Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Fix e2e logStream write-after-end crash in CI Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Christopher Speller <crspeller@gmail.com> Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com>
Contributor
|
Cursor Agent can help with this pull request. Just |
🤖 LLM Evaluation ResultsOpenAI
❌ Failed EvaluationsShow 2 failuresOPENAI1. TestReactEval/[openai]_react_cat_message
2. TestThreadsOpenQuestionsFromExportedData/[openai]_open_questions_from_eval_timed_dnd.json
Anthropic
❌ Failed EvaluationsShow 1 failuresANTHROPIC1. TestReactEval/[anthropic]_react_cat_message
This comment was automatically generated by the eval CI pipeline. |
|
Creating a Plugin SpinWick test server |
|
Plugin Spinwick PR #555 🎉 Test server created! Access here: https://agents-pr-555-na8a3.test.mattermost.cloud
Installation ID: Credentials: Posted securely in this Mattermost channel - Look for PR #555 |
Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com>
|
Test server creation failed. Review the error details here. |
Co-authored-by: Christopher Speller <crspeller@users.noreply.github.com>
|
Test server creation failed. Review the error details here. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
masterbranch to resolve the merge conflict.master.Relevance
The underlying UTF-16 citation/indexing fix has already landed upstream, so this PR can be closed.
Ticket Link
N/A
Screenshots
N/A
Release Note