| 1 | // cloud_budget.go — phase 3 of docs/ai-provisioning.md: the allowance and |
| 2 | // pool math that turns the usage ledger (#857) into a budget answer. Money |
| 3 | // only ever appears here as CONFIG (rates, allowance, pool — env, set by the |
| 4 | // operator) and as arithmetic over the ledger's quantities; no payment |
| 5 | // machinery lives in this file. The forwarding of over-allowance usage to |
| 6 | // Stripe Billing Meters is deliberately NOT built yet — it lands with #859's |
| 7 | // AI add-on product, after the rates and allowance below are reviewed and |
| 8 | // published in docs/tier-economics.md. |
| 9 | // |
| 10 | // The budget model, per the locked decisions (Paul 2026-07-14): |
| 11 | // - Every space gets a small included monthly allowance (in EUR of model |
| 12 | // cost, computed from per-model rates over the four token counters). |
| 13 | // - A fleet-wide monthly pool bounds the aggregate subsidy: however many |
| 14 | // free spaces opt in, the total included spend has a ceiling — the |
| 15 | // creates_log circuit-breaker pattern, applied to tokens. |
| 16 | // - Exhausted (space over allowance, or pool drained) means the sidecar |
| 17 | // drops to the deterministic command floor with one honest notice |
| 18 | // (internal/bot/meter.go); nothing breaks, nothing is deleted. |
| 19 | // - Unconfigured = measure-only: every report answers "ok", exactly the |
| 20 | // phase-1 behavior. BYO-key and self-hosted sidecars never report at all. |
| 21 | package main |
| 22 | |
| 23 | import ( |
| 24 | "encoding/json" |
| 25 | "fmt" |
| 26 | "log" |
| 27 | "time" |
| 28 | ) |
| 29 | |
| 30 | // aiRates prices one model's tokens, EUR per million, one rate per counter — |
| 31 | // cache reads and writes price differently from fresh input, which is why the |
| 32 | // ledger keeps all four dimensions distinct. |
| 33 | type aiRates struct { |
| 34 | In float64 `json:"in"` |
| 35 | Out float64 `json:"out"` |
| 36 | CacheWrite float64 `json:"cw"` |
| 37 | CacheRead float64 `json:"cr"` |
| 38 | } |
| 39 | |
| 40 | // parseAIRates reads LCHAT_CLOUD_AI_RATES — a JSON object keyed by model id, |
| 41 | // e.g. {"claude-sonnet-5":{"in":3,"out":15,"cw":3.75,"cr":0.3}}. Rates are |
| 42 | // EUR per million tokens. Empty = no pricing = the budget never bites. |
| 43 | func parseAIRates(v string) (map[string]aiRates, error) { |
| 44 | if v == "" { |
| 45 | return nil, nil |
| 46 | } |
| 47 | rates := map[string]aiRates{} |
| 48 | if err := json.Unmarshal([]byte(v), &rates); err != nil { |
| 49 | return nil, fmt.Errorf("bad LCHAT_CLOUD_AI_RATES: %v", err) |
| 50 | } |
| 51 | for model, r := range rates { |
| 52 | if r.In < 0 || r.Out < 0 || r.CacheWrite < 0 || r.CacheRead < 0 { |
| 53 | return nil, fmt.Errorf("negative rate for %q", model) |
| 54 | } |
| 55 | } |
| 56 | return rates, nil |
| 57 | } |
| 58 | |
| 59 | // budgetMonth is the allowance window — a UTC calendar month, matching the |
| 60 | // ledger's day grain. |
| 61 | func budgetMonth() string { return time.Now().UTC().Format("2006-01") } |
| 62 | |
| 63 | // usageCostEUR prices ledger rows against the rate book. Unknown models and |
| 64 | // non-token dimensions cost zero — activity counters (charter_runs) are |
| 65 | // measurement, not money, and a model missing from the rate book is an |
| 66 | // operator gap worth logging, not a reason to guess a price. |
| 67 | func usageCostEUR(rows []botUsageRow, rates map[string]aiRates) float64 { |
| 68 | var eur float64 |
| 69 | for _, row := range rows { |
| 70 | r, ok := rates[row.Model] |
| 71 | if !ok { |
| 72 | if row.Model != "" { |
| 73 | log.Printf("cloud: budget: no rate for model %q — its tokens price as 0 (add it to LCHAT_CLOUD_AI_RATES)", row.Model) |
| 74 | } |
| 75 | continue |
| 76 | } |
| 77 | perM := map[string]float64{ |
| 78 | "llm_input_tokens": r.In, |
| 79 | "llm_output_tokens": r.Out, |
| 80 | "llm_cache_write_tokens": r.CacheWrite, |
| 81 | "llm_cache_read_tokens": r.CacheRead, |
| 82 | }[row.Dimension] |
| 83 | eur += float64(row.Quantity) / 1e6 * perM |
| 84 | } |
| 85 | return eur |
| 86 | } |
| 87 | |
| 88 | // budgetState answers one space's state for a flush response: "ok" or |
| 89 | // "exhausted". Ordered cheap-to-dear: no config short-circuits before any |
| 90 | // ledger read, the per-space allowance before the fleet-wide sum. |
| 91 | func (s *cloudServer) budgetState(spaceID int64) string { |
| 92 | if len(s.aiRates) == 0 || s.aiAllowanceEUR <= 0 { |
| 93 | return "ok" // measure-only until the operator prices the budget |
| 94 | } |
| 95 | month := budgetMonth() |
| 96 | rows, err := s.db.usageForMonth(spaceID, month) |
| 97 | if err != nil { |
| 98 | log.Printf("cloud: budget: space %d month read: %v", spaceID, err) |
| 99 | return "ok" // a ledger blip must never silence a space's bot |
| 100 | } |
| 101 | if usageCostEUR(rows, s.aiRates) >= s.aiAllowanceEUR { |
| 102 | return "exhausted" |
| 103 | } |
| 104 | if s.aiPoolEUR > 0 { |
| 105 | fleet, err := s.db.fleetUsageForMonth(month) |
| 106 | if err != nil { |
| 107 | log.Printf("cloud: budget: fleet month read: %v", err) |
| 108 | return "ok" |
| 109 | } |
| 110 | if usageCostEUR(fleet, s.aiRates) >= s.aiPoolEUR { |
| 111 | log.Printf("cloud: budget: the fleet AI pool for %s is spent — subsidized AI paused fleet-wide", month) |
| 112 | return "exhausted" |
| 113 | } |
| 114 | } |
| 115 | return "ok" |
| 116 | } |
| 117 | |
amadan