Skip to main content

​
Overview

The VibeLearn worker is an Express.js HTTP server managed by Bun, running on port 37778. It handles:
  • Session lifecycle (init, observations)
  • The 5-step analysis pipeline
  • Upstream sync with retry queue
  • System health and admin

​
Startup

The worker is started by the SessionStart hook if not already running: 
# Manual start
bun plugin/scripts/worker-service.cjs start

# Status
curl http://127.0.0.1:37778/api/health | jq .

# Readiness (used by hooks)
curl http://127.0.0.1:37778/api/readiness
The worker is kept alive by plugin/scripts/worker-wrapper.cjs (Bun-managed). Environment variable VIBELEARN_MANAGED=true is set in the worker process.

​
API Endpoints

​
Session Management

MethodPathDescription
POST/api/sessions/initInitialize a new session
POST/api/sessions/observationsStore a tool observation
GET/api/sessions/:id/summaryGet session summary
POST /api/sessions/init 
{
  "contentSessionId": "session-abc",
  "project": "my-app",
  "prompt": "Add authentication"
}
POST /api/sessions/observations 
{
  "contentSessionId": "session-abc",
  "tool_type": "file_edit",
  "file_path": "src/auth.ts",
  "content": "export function...",
  "cwd": "/Users/me/my-app"
}

​
VibeLearn Analysis Pipeline

MethodPathDescription
POST/api/vibelearn/analyze/stackDetect tech stack
POST/api/vibelearn/analyze/staticAST pattern analysis
POST/api/vibelearn/analyze/conceptsLLM concept extraction
POST/api/vibelearn/analyze/quizLLM quiz generation
POST/api/vibelearn/syncSync to vibelearn.dev
All analysis endpoints accept { contentSessionId }. The concepts endpoint also accepts { last_assistant_message }. Concept extraction response: 
{
  "ok": true,
  "concepts": ["React Server Actions", "HMAC signing", "SQLite WAL mode"],
  "summary": "Built auth middleware with HMAC-signed tokens..."
}

​
VibeLearn Data

MethodPathDescription
GET/api/vibelearn/profileDeveloper mastery profile
GET/api/vibelearn/questions/pendingUnanswered quiz questions

​
System

MethodPathDescription
GET/api/healthFull health check
GET/api/readinessSimple ready check (used by hooks)
POST/api/admin/restartRestart worker (requires VIBELEARN_MANAGED=true)
POST/api/admin/shutdownShutdown worker

​
Analysis Pipeline Implementation

Each analysis endpoint is implemented in src/services/worker/http/routes/VibeLearnRoutes.ts: 
POST /analyze/stack
  └─ StackDetector.detect(filePaths)
       └─ reads package.json, pyproject.toml, go.mod etc.
       └─ writes to vl_stack_profiles

POST /analyze/static
  └─ StaticAnalyzer.analyze(sessionId)
       └─ reads modified files from disk (aggregated from observations)
       └─ tree-sitter AST parsing

POST /analyze/concepts
  └─ createAgentRunner() — tries Gemini → OpenRouter → Anthropic
  └─ ConceptExtractor.extract(prompt)
       └─ single LLM call
       └─ writes to vl_concepts + vibelearn_session_summaries

POST /analyze/quiz
  └─ QuizGenerator.generate(concepts)
       └─ single LLM call, skips concepts with mastery_score > 0.85
       └─ writes to vl_questions

POST /sync
  └─ UpstreamSync.sync(sessionId)
       └─ HMAC signs payload with user's API key
       └─ POST to api.vibelearn.dev/v1/sync
       └─ on failure: OfflineQueue.enqueue() → retried next sync

​
LLM Provider Selection

createAgentRunner() in VibeLearnRoutes creates a (prompt: string) => Promise<string> function that tries providers in order:
  1. Gemini — if VIBELEARN_GEMINI_API_KEY is set in settings
  2. OpenRouter — if VIBELEARN_OPENROUTER_API_KEY is set
  3. Anthropic — direct HTTP to api.anthropic.com using ANTHROPIC_API_KEY

​
Logs

# View today's logs
tail -f ~/.vibelearn/logs/vibelearn-$(date +%Y-%m-%d).log
Log level controlled by VIBELEARN_LOG_LEVEL in settings (debug, info, warn, error).

​
Route Registration

Routes are registered in WorkerService.registerRoutes(): 
this.server.registerRoutes(new SessionRoutes(this.dbManager));
this.server.registerRoutes(new VibeLearnRoutes(this.dbManager));
// ...other route handlers