Documentation
¶
Overview ¶
Package aiprofile implements AI_PROFILE enforcement, budget caps, feature flags, per-tier timeouts, and rate-limit coordination. Sprint P88-S05: T-05-06 through T-05-10.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func IsTierAllowed ¶
IsTierAllowed checks whether a tier is permitted by the current profile.
Types ¶
type BudgetConfig ¶
type BudgetConfig struct {
DailyUSD float64 // AI_DAILY_BUDGET_USD (0 = disabled / free tier only)
MonthlyUSD float64 // AI_MONTHLY_BUDGET_USD (0 = disabled / free tier only)
}
BudgetConfig holds the daily and monthly budget caps from the AI config block (CLI-R18: folded into .env.secrets; previously a dedicated .env.ai).
func LoadBudget ¶
func LoadBudget() BudgetConfig
LoadBudget reads budget env vars. Zero means disabled (free tier only, paid tiers blocked). Positive value is a hard cap in USD.
func (BudgetConfig) CheckBudget ¶
func (b BudgetConfig) CheckBudget(spentDaily, spentMonthly float64) error
CheckBudget evaluates whether a request should be allowed given current spend. Returns nil if allowed, or an error with a user-facing message if blocked. spentDaily and spentMonthly are in USD (converted from cost_micros / 1e6).
func (BudgetConfig) IsPaidTierBlocked ¶
func (b BudgetConfig) IsPaidTierBlocked() bool
IsPaidTierBlocked returns true when budget is zero (free tier only mode).
func (BudgetConfig) ShouldAlert ¶
func (b BudgetConfig) ShouldAlert(spentMonthly float64) bool
ShouldAlert returns true when monthly spend has reached the budget threshold.
type FeatureFlags ¶
type FeatureFlags struct {
AutoProvision bool // AI_FEATURE_AUTO_PROVISION (A2)
LocalAutoInstall bool // AI_FEATURE_LOCAL_AUTO_INSTALL (A1)
RoutingUI bool // AI_FEATURE_ROUTING_UI (A3 UI)
HealthDaemon bool // AI_FEATURE_HEALTH_DAEMON (A5)
}
FeatureFlags holds the AI_FEATURE_* master switches (T-05-08). All default to true (on). Flippable at runtime via admin settings.
func LoadFeatureFlags ¶
func LoadFeatureFlags() FeatureFlags
LoadFeatureFlags reads AI_FEATURE_* env vars. Missing or empty = true (on).
type Profile ¶
type Profile string
Profile represents the AI routing profile.
func ParseProfile ¶
ParseProfile parses an AI_PROFILE string into a typed Profile.
type Tier ¶
type Tier string
Tier represents a routing tier.
func AllowedTiers ¶
AllowedTiers returns the tiers permitted by the given profile.
func FilterTierChain ¶
FilterTierChain filters a tier chain based on the profile. Returns only allowed tiers in the original order. Used by the router at load time (T-05-06).
type TierTimeouts ¶
type TierTimeouts struct {
Local time.Duration // AI_TIMEOUT_LOCAL_MS
OAuth time.Duration // AI_TIMEOUT_OAUTH_MS
Pool time.Duration // AI_TIMEOUT_POOL_MS
Paid time.Duration // AI_TIMEOUT_PAID_MS
}
TierTimeouts holds per-tier timeout overrides from env vars (T-05-09). Zero means "use routing defaults from task config".
func LoadTierTimeouts ¶
func LoadTierTimeouts() TierTimeouts
LoadTierTimeouts reads AI_TIMEOUT_*_MS env vars.
func (TierTimeouts) TimeoutForTier ¶
func (t TierTimeouts) TimeoutForTier(tier Tier) time.Duration
TimeoutForTier returns the timeout override for the given tier, or zero if no override is set (caller should use its own default).
type TokenBucket ¶
type TokenBucket struct {
// contains filtered or unexported fields
}
TokenBucket implements a per-key rate limiter capping at 15 req/min (T-05-10). Thread-safe. Each key gets its own bucket.
func NewTokenBucket ¶
func NewTokenBucket() *TokenBucket
NewTokenBucket creates a rate limiter with 15 RPM per key.
func (*TokenBucket) Allow ¶
func (tb *TokenBucket) Allow(keyID int) bool
Allow checks whether a request for the given key ID is allowed. Returns true if allowed, false if rate-limited.
func (*TokenBucket) IsAvailable ¶
func (tb *TokenBucket) IsAvailable(keyID int) bool
IsAvailable returns true if the key is not currently in cooldown.
func (*TokenBucket) MarkRateLimited ¶
func (tb *TokenBucket) MarkRateLimited(keyID int, retryAfter time.Duration)
MarkRateLimited marks a key as rate-limited. If retryAfter > 0, uses that exact duration. Otherwise applies exponential backoff: 60s, 120s, 240s, up to 1h.
Source Files