Scaffold Go plugin skeleton with Ollama client, conversation store, rate limiter

server/rate.go

@@ -0,0 +1,64 @@
+package main
+
+import (
+    "sync"
+    "time"
+)
+
+// RateLimiter implements a per-user token bucket rate limiter.
+type RateLimiter struct {
+    mu    sync.Mutex
+    users map[string]*bucket
+    rate  int // requests per minute
+}
+
+type bucket struct {
+    tokens     float64
+    lastRefill time.Time
+}
+
+// NewRateLimiter creates a rate limiter with the given rate (requests/minute).
+func NewRateLimiter(ratePerMinute int) *RateLimiter {
+    if ratePerMinute <= 0 {
+        ratePerMinute = 10
+    }
+    return &RateLimiter{
+        users: make(map[string]*bucket),
+        rate:  ratePerMinute,
+    }
+}
+
+// Allow checks if a request from the given user ID should be allowed.
+func (rl *RateLimiter) Allow(userID string) bool {
+    rl.mu.Lock()
+    defer rl.mu.Unlock()
+
+    b, exists := rl.users[userID]
+    now := time.Now()
+
+    if !exists {
+        b = &bucket{tokens: 1, lastRefill: now}
+        rl.users[userID] = b
+    }
+
+    elapsed := now.Sub(b.lastRefill).Seconds()
+    refill := elapsed * float64(rl.rate) / 60.0
+    b.tokens += refill
+    if b.tokens > 1 {
+        b.tokens = 1
+    }
+    b.lastRefill = now
+
+    if b.tokens >= 1 {
+        b.tokens--
+        return true
+    }
+    return false
+}
+
+// Reset clears all rate limiter state.
+func (rl *RateLimiter) Reset() {
+    rl.mu.Lock()
+    defer rl.mu.Unlock()
+    rl.users = make(map[string]*bucket)
+}