feat: move agentic loop and session management to backend#46
Merged
alexinthesky merged 1 commit intomainfrom Feb 14, 2026
Merged
feat: move agentic loop and session management to backend#46alexinthesky merged 1 commit intomainfrom
alexinthesky merged 1 commit intomainfrom
Conversation
alexinthesky
changed the title
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
alexinthesky
added a commit
that referenced
this pull request
Feb 13, 2026
…ition **Issue 1: Playwright using 2 workers despite config setting workers: 1** Root cause: Top-level `fullyParallel: true` in playwright.config.ts was overriding project-level `workers: 1` settings for chromium-llm-tests. Fix: Remove top-level `fullyParallel: true` to respect per-project configuration. Projects can now properly control their own parallelism: - chromium-session-tests: workers: 1, fullyParallel: false - chromium-llm-tests: workers: 1, fullyParallel: false - chromium: workers: 6, fullyParallel: true (default) **Issue 2: Agent runs being canceled mid-stream (500 error after ~500ms)** Root cause: When a new session is created during `sendMessage()`, the session ID changes from null to the new ID. This triggers the reconnect useEffect cleanup, which aborts the AbortController *while the SSE stream is still active*, causing: ``` rpc error: code = Canceled desc = context canceled ``` Flow that caused the bug: 1. User sends message (sessionId is null) 2. Backend creates session, returns sessionId 3. Frontend calls `setCurrentSessionIdDirect(result.sessionId)` 4. sessionManager.currentSessionId changes (null → new ID) 5. Reconnect useEffect dependency triggers cleanup 6. Cleanup aborts AbortController 7. Active SSE stream in `reconnectToAgentRun` gets canceled 8. Send button never re-enables, E2E tests timeout Fix: In reconnect effect cleanup, only abort if `!isGenerating`. This prevents aborting the controller while a message is actively being streamed. The abort will still happen when truly needed (user navigates away, explicitly stops generation, or session changes while idle). **Impact:** - E2E tests will now properly run LLM tests with single worker - Agent runs will complete successfully without mid-stream cancellation - Send button will re-enable after responses complete **Note on LLM_API_KEY:** Secret is correctly configured and LLM calls are succeeding. The test failures were caused by the SSE abort race condition, not missing API key. Fixes #46 Co-Authored-By: Claude Sonnet 4.5 <[email protected]>
This was referenced Feb 13, 2026
alexinthesky
added a commit
that referenced
this pull request
Feb 13, 2026
- Fix double close on channel in RunBroadcaster unsubscribe - Fix broadcaster race condition in AppendEvent - Replace hardcoded rgba() colors with semantic Tailwind classes - Remove dead code for cookie-based LLM auth (Grafana 12 strips cookies) - Add unit tests for ReasoningIndicator component All tests passing: - Frontend unit tests: 552 passed - Backend Go tests: all passed - TypeScript type checking: passed - ESLint: passed (warnings in unmodified code only) Co-Authored-By: Claude Sonnet 4.5 <[email protected]>
alexinthesky
added a commit
that referenced
this pull request
Feb 13, 2026
Fixes two medium-severity issues identified by cursor bot: 1. Cancelled runs with no events marked as failed - Check if run is still cancellable when no events produced - Set status to "cancelled" if run was cancelled early - Prevents status mismatch between endpoint response and stored status 2. Redis SubscribeAndSnapshot duplicate-event race condition - Add sequence numbers to SSEEvent for ordering - Implement atomic sequence tracking in RunStore (in-memory counter) - Implement atomic sequence tracking in RedisRunStore (Redis INCR) - Add client-side deduplication in agentClient (skip sequence <= lastSeen) - Prevents duplicate events when reconnecting to in-progress runs Low-severity issues deferred to follow-up issue #48: - Empty assistant message on cancelled runs - GetBroadcaster dead code removal All tests passing: - Frontend unit tests: 552 passed - Backend Go tests: all passed - TypeScript type checking: passed - ESLint: passed (warnings in unmodified code only) Co-Authored-By: Claude Sonnet 4.5 <[email protected]>
… test consolidation Move the agentic loop from the frontend to a Go backend implementation with SSE streaming, detached execution (survives browser tab close), and automatic reconnection. Includes multi-org support, RBAC enforcement at tool execution, token-aware context window management, and session persistence. Key changes: - Backend agentic loop (pkg/agent/) with LLM client, tool execution, context window - Detached execution with atomic subscribe+snapshot for reconnection - Session management moved from frontend to backend (in-memory + Redis stores) - Frontend simplified to thin HTTP client (backendSessionClient.ts) - Reasoning/thinking content dropped when tool calls are present - E2E tests consolidated from 17 files to 8 with all 36 tests passing (0 skipped) - All LLM-dependent tests use consistent wait patterns and fast prompts Co-Authored-By: Claude Opus 4.6 <[email protected]>
alexinthesky
force-pushed
the
vk/4e22-improve-code-str
branch
from
26bbfd5 to
9d5540c
Compare
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
gespi1
approved these changes
Feb 13, 2026
alexinthesky
changed the title
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
2 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.
Note
High Risk
Introduces new backend chat execution, persistence, and SSE streaming paths and changes RBAC enforcement semantics, which can impact authorization and run/session correctness across orgs and deployments.
Overview
Moves the AI chat agentic loop into the Go backend via a new
/api/agent/*API, including SSE event streaming, run tracking (in-memory or Redis), cancellation, and token-aware context trimming. The backend now callsgrafana-llm-app’s OpenAI-compatible endpoint using a Grafana service-account token and executes MCP tool calls with execution-time RBAC enforcement.Reworks RBAC from hardcoded tool-name lists to MCP
ToolAnnotations(readOnlyHint) propagated from MCP servers, and adds proxy support for tool lookup plus per-server header injection + dial timeouts. Adds backend session CRUD/current-session tracking and persists assistant/tool-call results from streamed events.Updates local/dev/test plumbing: adds
server:fullmulti-org compose setup, optional Prometheus datasource provisioning env vars, switches E2E secrets fromOPENAI_API_KEYtoLLM_API_KEY, and serializes Playwright workers while adding a stop/queue UX inChatInputfor overlapping sends.Written by Cursor Bugbot for commit 9d5540c. This will update automatically on new commits. Configure here.