Skip to content

g1t/docs/BILLING_OPERATIONS.md

977 lines64,062 bytesCodeBlame

Pick any line to see why it is the way it is: the commit, the pull request and issue it came from, and what the agent was thinking.

Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily1# Billing operations: costs, margin and prices
2
3How g1t checks what it charges against what Cloudflare charges it, keeps
4prices at cost plus 20%, and tells staff when the margin slips. Internal.
5Code: `services/billing/src/costs.rs` (reading the bill), `margin.rs`
6(reconciliation, drift, alerts), `pricing.rs` (versions, proposals,
Spend caps: a monthly budget for comped workspaces and a daily breaker on what g1t pays7notice), `keeper.rs` (sandbox and Workers for Platforms measurements),
8`budget.rs` (what g1t pays for itself, and its caps; see
9[Spend caps](#spend-caps)).
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily10Page: sudo **Costs & margin** (`/costs`).
11
12Several Cloudflare products g1t runs on are new. Artifacts bills
13"operations" from 2026-10-14 without defining them (see
14[ARTIFACTS.md](ARTIFACTS.md), M1). So nothing here hard-codes a product
15list or a unit: every line Cloudflare bills is kept, and how a line maps
16to what g1t sells is data you change from sudo, without a deploy.
17
18## Data sources
19
20| Source | What | Where it lands |
21| --- | --- | --- |
22| Billable usage, `GET /accounts/{account}/billable-usage?from=&to=` | One row per service per day in FOCUS columns: `ServiceFamilyName`, `ServiceName`, `ChargePeriodStart`, `PricingQuantity`, `ContractedCost` / `BilledCost` / `ListCost`. Every product g1t uses appears once it is used: Workers, Workers for Platforms, D1, KV, R2, Queues, Containers, Durable Objects, Artifacts, Browser Rendering, Workers AI, Vectorize, Cloudflare for SaaS, Email. Inside an included amount the cost is 0. | `cost_lines`, source `billable_usage` |
Costs: Cloudflare's count for a pull request's working copy is shared out to its repository's workspace (repos pull_owners)23| GraphQL `artifactsEventsAdaptiveGroups` | Artifacts' own count by `date`, `eventType` and `repositoryName`. Operations are `create`, `fork`, `push`, `pull`, `delete`; errors (`rateLimited`, `serverError`, …) are kept but not counted. | `cost_lines`, source `artifacts_events`; per workspace (from the store key `<workspace>--<repo>`; a pull request's working copy, `pulls--<id>`, is its repository's workspace's, from repos' `pull_owners`) in `own_counts` as `cloudflare_git` |
Merge branch 'worktree-agent-a633ac0f7f66d419d'24| GraphQL `aiGatewayRequestsAdaptiveGroups`, filtered to `AI_GATEWAY_ID` | What AI Gateway priced g1t's own provider traffic at, by `date`, `provider`, `model` and `wholesale`: `count`, `sum.cost` (dollars), `sum.tokensIn`/`tokensOut`/`cacheReadTokens`/`cacheWriteTokens`. Field names checked against Cloudflare's schema (introspection of `AccountAiGatewayRequestsAdaptiveGroups{Sum,Dimensions,Filter_InputObject}`). An adaptive (sampled) dataset: an estimate, close at g1t's volumes. Only g1t's hosted models go through this gateway: a workspace's own provider is called at its own address, never here. | `cost_lines`, source `ai_gateway`, product `ai_gateway_requests`: per day and model a line `<provider>_<model>` (requests, at the gateway's cost), and at no cost `…__tokens`, `…__cache_read_tokens`, `…__cache_write_tokens`; Cloudflare-billed (unified billing) requests are prefixed `wholesale__`. Mapped to `models` (migration 0036). A re-read day replaces all its gateway lines. |
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily25| The ledger | Every charge: its cost at the price book's cost, what it was charged at price, what paid for it. | read, never written |
26| `pending_usage` | Month-end meters (git, storage, scans, embeddings, the cache) as they stand. | snapshotted daily into `pending_days` |
27| `plan_payments` | The plan's $20. | read |
One operation mapping, owned by repos; billing reads it instead of keeping its own28| repos `git_operations` | Operations customers are charged for, per workspace, counted by repos through its `operation_mapping`. | `own_counts` meter `git_operations` |
Costs: Cloudflare's subscriptions read from Cloudflare each day, the estimate only until then; sudo's costs split into Costs & margin and Bill & pricing29| Subscriptions, `GET /accounts/{account}/subscriptions` | What g1t pays each month whatever it uses (Workers Paid, add-ons): each subscription that is paid, trialing or awaiting payment, at its price over its frequency. Not on the billable-usage bill. Read in the daily run with the bill's token; a failure is logged and the last read stays. | `cf_subscriptions` (one row) |
One operation mapping, owned by repos; billing reads it instead of keeping its own30| repos `artifacts_usage` | Every raw meter of the git store (`git.fetch`, `git.receive_pack`, `binding.*`, …) per day and workspace, with repos' `operation_mapping`. | `own_counts` meters `artifacts_<raw meter>`, and `cost_operations` (raw counts × the mapping's `cost_operations`: what g1t expects Cloudflare to bill) |
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily31
32A meter's slug is Cloudflare's name lower-cased with words joined by `_`
33and the "(First … included)" note dropped: `Workers for Platforms CPU ms
34(First 60M ms are included)` under `Workers` is product `workers`, meter
35`workers_for_platforms_cpu_ms`. Several rows of the same day and meter
36(regions, tiers) are added together before they are stored.
37
38## Credentials
39
40| Secret on g1t-billing | Permissions | Used for |
41| --- | --- | --- |
Costs: Cloudflare's subscriptions read from Cloudflare each day, the estimate only until then; sudo's costs split into Costs & margin and Bill & pricing42| `CLOUDFLARE_BILLING_TOKEN` (optional) | Account: **Billing Read**, Account: **Account Analytics Read**, for the g1t account only | Reading the bill, the Artifacts events and the subscriptions |
Billing: AI Gateway's analytics are read with the token that can see them, and a gateway that priced nothing is said43| `CLOUDFLARE_USAGE_TOKEN` (exists) | Billing Read, Account Analytics Read, AI Gateway Read | The keeper (settling runs from the gateway's logs); also the bill when `CLOUDFLARE_BILLING_TOKEN` is not set. AI Gateway's analytics are read with this token first and, if that is refused, with the bill's: Cloudflare answers a token without AI Gateway Read with no rows rather than an error, so the bill's token would read as a gateway that priced nothing |
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily44
45With neither, the daily run reconciles only what g1t counted itself, and
46the page says the bill cannot be read. Nothing fails. To set the scoped one:
47
481. Cloudflare dashboard → My Profile → API Tokens → Create Token → Custom token.
492. Permissions: Account · Billing · Read; Account · Account Analytics · Read.
503. Account resources: Include · the g1t account. No zone permissions.
514. `cd services/billing && npx wrangler secret put CLOUDFLARE_BILLING_TOKEN`.
sudo: the costs run button is named for what it does, the whole nightly analysis525. In sudo, Costs & margin → **Run the analysis now**.
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily53
54Alerts are emailed through the `EMAIL` binding (Cloudflare Email Sending)
55to `COSTS_ALERT_EMAIL` (`hey@flagon.io`). An empty value sends none.
56
57## Schedule
58
59The daily cron (`17 4 * * *`, `keeper::DAILY`) runs, in order:
60
611. The keeper's measurements (sandbox seconds, app requests and CPU), each
62 a proposal now, not a direct change.
632. `costs_daily`:
Merge branch 'worktree-agent-a633ac0f7f66d419d'64 1. Read the bill, the Artifacts events and AI Gateway's analytics. The first run reads the last
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily65 31 days (GraphQL keeps 31); later runs the last 4, since Cloudflare
66 restates recent days, or back to the last day read after a gap.
67 Lines are upserted on `(day, source, product, meter)`, so a re-read
68 replaces, never adds.
69 2. g1t's own counts for the same days (replaced per day).
70 3. Snapshot `pending_usage` into `pending_days`.
sudo: the costs run button is named for what it does, the whole nightly analysis71 4. Reconcile the last 31 days (further back after a gap) into
72 `margin_days` and `workspace_costs`
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily73 (replaced per day).
74 5. Drift over the last 7 days into `cost_drift`.
75 6. Unit costs over the last 30 days, proposed to the price book.
76 7. Apply price versions whose date has come.
77 8. Open, update and close margin alerts; email new ones.
78 9. Email owners on the plan about rises to come.
79
Costs: Cloudflare's subscriptions read from Cloudflare each day, the estimate only until then; sudo's costs split into Costs & margin and Bill & pricing80**Run the analysis now** at the top of sudo's Costs & margin and Bill & pricing pages
sudo: the costs run button is named for what it does, the whole nightly analysis81(`admin_run_costs`) runs all of step 2 at once, alerts included, with no
82need to wait for 04:17 UTC. Running it twice is safe: every step replaces
83what it wrote.
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily84
85## Reconciliation math
86
87Every Cloudflare line goes to one of g1t's products ("buckets") by
88`cost_map`: the row for its product with the longest matching meter
89prefix, `*` last. A line no row claims goes to `unmapped`.
90
91| Bucket | Cloudflare | Paid for by (`revenue_map`) |
92| --- | --- | --- |
93| `sandboxes` | Containers, Durable Objects compute duration | `sandbox`, `self_hosted`, `builds` |
94| `deployments` | Workers for Platforms | `deployments` |
95| `git` | Artifacts operations (and its events, as counts) | `git` |
96| `repo_storage` | Artifacts storage | `storage` |
97| `actions_cache` | R2 | `cache` |
98| `embeddings` | Workers AI, Vectorize | `context` |
99| `security` | (Workers CPU, under `platform`) | `security` |
100| `domains` | Cloudflare for SaaS | `domains` |
Merge branch 'worktree-agent-a633ac0f7f66d419d'101| `models` | not Cloudflare: AI Gateway's settled cost on the ledger; AI Gateway's own daily total (`ai_gateway_requests`) beside it, to check it | every other task (agent runs) |
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily102| `platform` | Workers, D1, KV, Queues, Email, Browser Rendering, other Durable Objects | the plan's price |
103
104For each day and bucket:
105
Costs: a statement that keeps usage sold, running g1t, subscriptions and what was given away (comped, free use, trial, pool) apart, and says who was paid; free use carries its own cost; the run button says it is running106- **Cloudflare cost** = Σ the bucket's lines' cost, as billed: after the
107 included allowances, so a month inside them costs $0 here as on
108 Cloudflare's Billable usage page. `models` uses the ledger's cost of the
109 tokens instead; that is paid to the model providers and is not on
Merge branch 'worktree-agent-a633ac0f7f66d419d'110 Cloudflare's bill. Its "Cloudflare" column is what AI Gateway priced the
111 same traffic at, which drift compares with the ledger (below); it is
112 never added to the cost.
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily113- **Own cost** = Σ the ledger's `cost_micros` for the bucket's keys (the
114 price book's cost when charged), plus month-end deltas. A workspace's own
115 model provider is no cost to g1t.
116- **Value** = what customers were charged at price: `-amount_micros` plus
117 what the plan's included usage, a trial, the open-source pool or g1t paid.
118 g1t's own (comped) workspaces are valued at cost plus the margin.
119- **Cash** = what workspaces paid: `-amount_micros`, and the plan's price.
Merge Stripe Tax, the card fee on card payments, and one free workspace per person120 Never tax or card fees: a payment credits the balance, and
121 `plan_payments`, without them (see [Tax and the card fee](#tax-and-the-card-fee)).
Costs: margin is measured on what was sold; comped workspaces, free periods, the trial and the pools are given away, a budget shown beside it122- **Given away** = the part of the cost that went on usage g1t paid for
Costs: a statement that keeps usage sold, running g1t, subscriptions and what was given away (comped, free use, trial, pool) apart, and says who was paid; free use carries its own cost; the run button says it is running123 itself on purpose, by why:
124 - **comped**: all of a comped workspace's cost, every bucket;
125 - **free use**: a free period's usage, the overruns g1t covered
126 (`ledger.given_micros`), and all of a workspace's cost on a day it had
127 nothing priced (free allowances);
128 - **trial** and **open-source pool**: what `trial_micros` and
Merge branch 'worktree-agent-a633ac0f7f66d419d'129 `oss_micros` paid;
130 - **discount**: what a discount on an account's custom terms took below
131 cost plus the margin (`ledger.discount_micros`, see
132 [Margin floor](#margin-floor)). The usage is valued at its price, so a
Billing: credits with a kind and expiry, discounts instead of comped, and safer charging133 discounted sale never reads as margin lost;
134 - **promotional credit** and **goodwill credit**: what credit staff gave
135 paid for, when it is spent (`given_credit_promotional_micros`,
136 `given_credit_goodwill_micros`, migration 0038). That usage's charge is
137 taken out of cash, so it is never money in. A refund is not here: see
Merge branch 'main' into worktree-agent-a69aeabc4b0deeb97138 [Credits from g1t](#credits-from-g1t);
139 - **testing resets**: what a workspace's usage cost g1t before staff
140 reset its billing (`reset_costs`, `given_reset_micros`, migration
141 0046). The model calls and Cloudflare usage still happened, so the
142 reconciliation reads the kept rows back as that workspace's usage on
143 their days: valued as before, no cash, all of it given. See
144 [Resetting a test workspace](#resetting-a-test-workspace).
Costs: a statement that keeps usage sold, running g1t, subscriptions and what was given away (comped, free use, trial, pool) apart, and says who was paid; free use carries its own cost; the run button says it is running145
146 Otherwise a workspace's day is split by those shares of its value at
147 price, and the same shares of each of its buckets' cost are given, its
148 part of running g1t included. The Team plan's included usage is sold:
149 the plan's price paid for it. Stored on `margin_days` (`given_micros`
Merge branch 'worktree-agent-a633ac0f7f66d419d'150 and `given_<why>_micros`, `given_discount_micros` from migration 0036)
Merge branch 'main' into worktree-agent-a69aeabc4b0deeb97151 and `workspace_costs` (`given_micros`). Sudo's Costs & margin page lists
152 each why by name, testing resets included (`givenResetMicros`).
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily153- **Month-end meters**: a day's figure is that day's `pending_days`
154 snapshot less the day before's, within a month. Their month-end ledger
155 entries are left out, so nothing is counted twice.
Costs: a statement that keeps usage sold, running g1t, subscriptions and what was given away (comped, free use, trial, pool) apart, and says who was paid; free use carries its own cost; the run button says it is running156- **Product margin** = (value − cost) / value.
157- **Sudo's statement** keeps apart:
158 - **Usage sold**: cash for usage against the cost of the usage buckets
159 less what was given. Its margin is the headline; at cost plus 20% it
160 sits near 16.7%.
161 - **Running g1t**: the plan's price against `platform` less its given
162 share.
Costs: Cloudflare's subscriptions read from Cloudflare each day, the estimate only until then; sudo's costs split into Costs & margin and Bill & pricing163 - **Cloudflare subscriptions**: what Cloudflare lists, a month, over
164 the range (`cf_subscriptions`); until a read has worked,
165 `CLOUDFLARE_FIXED_MONTHLY_MICROS`, an estimate.
Costs: a statement that keeps usage sold, running g1t, subscriptions and what was given away (comped, free use, trial, pool) apart, and says who was paid; free use carries its own cost; the run button says it is running166 - **Not mapped**: billed, charged for by nothing.
167 - **Given away**: by why. A budget, watched under g1t's own spend, never
168 shown as a loss.
169 - **All in**: money in against all of it, with the figure without what
170 was given beside it. **Who g1t paid** splits the cost into Cloudflare
171 and the model providers.
172
173 The overall alert is (Σ cash − (Σ cost − Σ given)) / Σ cash.
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily174- **Quantities**: where a mapping names an `own_meter`, Cloudflare's
175 billed quantity of those lines (or, without one, Artifacts' operation
176 events) against g1t's own count.
177
178**Shared costs to workspaces.** A bucket's cost is shared in proportion
179to, first available: Cloudflare's own per-workspace count
180(`cloudflare_<bucket>`, today the Artifacts events by repository), g1t's
Costs: a statement that keeps usage sold, running g1t, subscriptions and what was given away (comped, free use, trial, pool) apart, and says who was paid; free use carries its own cost; the run button says it is running181own count, what its usage cost (so free use carries its own cost), what
182each was charged for it. `platform`
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily183and `unmapped` are shared by each workspace's share of all usage that
184day. Shares are whole micros that add up to the bill exactly (largest
185remainder).
186
187## Drift (last 7 days)
188
189| Kind | When | What to do |
190| --- | --- | --- |
One operation mapping, owned by repos; billing reads it instead of keeping its own191| Count | g1t's count and Cloudflare's differ by more than the mapping's `drift_percent` (10%) | Find out what Cloudflare counts: compare its events with `own_counts` `artifacts_*` and `cost_operations`. If it counts more (binding reads, `ls-refs`), either change repos' `operation_mapping` so customers are charged for what Cloudflare counts, or leave it and let the per-unit cost rise (below). |
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily192| Cost | Cloudflare charged more than `drift_percent` away from the price book's cost of the same usage, with at least `min_daily_cost` | A price is stale: check the proposals. |
Merge branch 'main' into worktree-agent-a69aeabc4b0deeb97193| Cost, on `models` | What AI Gateway priced g1t's own provider traffic at over the 7 days, against the ledger's model cost for the same days (billed to g1t: comped, free and trial use included, a workspace's own provider not) plus the model cost testing resets kept for those days (`reset_costs`), more than the `ai_gateway_requests` mapping's `drift_percent` (10%) apart, with at least `min_daily_cost`. A ledger with none of the gateway's cost is drift too, and so is a gateway that priced nothing against a ledger with at least `min_daily_cost` of model cost (no percentage): that is not agreement, it is a token that cannot see AI Gateway, or calls that went around it | The gateway higher: model calls g1t paid for and charged no one: runs not settled yet (they catch up within the hour), runs with no session, a run started without a billing ticket, or something else on g1t's gateway. The ledger higher: runs that reached a provider without the gateway. The detail adds why the gateway's own figure may be off: prompt-cache read and write tokens (the gateway prices them at its rates for cache tokens, which can lag the provider's; check against the provider's invoice), requests Cloudflare billed itself (unified billing: on Cloudflare's bill, not a provider's), and models with no price. A testing reset in the window is named in the detail: one that kept its cost says how much of the ledger's side it is; one from before resets kept their cost (the audit log has it, `reset_costs` does not) says the gateway's figure includes usage the ledger no longer has, so that part is not a leak, and the day it leaves the 7 days; while such a reset is in the window the `models` leak is not raised. Days are UTC by when a request ran (gateway) and when a charge was entered (ledger), so a run across midnight shifts a little between days; the 7-day sum absorbs it. |
Merge branch 'worktree-agent-a633ac0f7f66d419d'194| Unpriced | Over the 7 days, a model in AI Gateway's analytics with tokens and $0 cost, or runs settled with `runs.gateway_note` (the gateway could not price all of a run) | The gateway has no price for a model g1t runs: add it in the gateway (custom cost) or route away from it. Until then those runs are charged no less than the sandbox reported (Claude Code's own price table), never $0 silently. |
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily195| Leak | Cost of at least `min_daily_cost` and nothing charged for it (never for `platform`), or a meter in `unmapped` | Map the meter (below), or decide it is overhead (`platform`). |
196
197## Prices: versions, proposals, notice
198
199- `price_versions` holds every price ever, never edited. `prices` is the
200 version in force. The daily run applies a version once its
201 `effective_at` has come, and adds the public `price_changes` record.
202 Ledger entries made from the price book carry `price_version` (the
203 version ids, comma-separated), so a past statement is always explained
204 by the prices of its day.
205- Proposals come from the keeper (sandbox seconds, app requests and CPU)
206 and the reconciler (mappings with `scale_to_own`: today git operations).
207 For git operations: Cloudflare's rate per its own operation (the median
208 over charged days of cost ÷ quantity) × (Cloudflare's operations ÷ g1t's)
209 × 1,000. If Cloudflare counts three for each one g1t counts, the per-1,000
210 price triples. At least 1,000 of g1t's operations are needed.
211- Decision (`pricing::decide`): under 2% is noise; more than 4× either way
212 is suspect and waits for staff; within `auto_apply_percent` (25%) it is
213 applied on its own when `auto_apply` is on; anything else waits.
214- Notice: a fall applies at once. A rise applies `notice_days` (14) after
215 the decision, and for a monthly meter (git, storage, cache, domains,
216 embeddings, scans) at the start of the month after that, so no month
217 is charged at two prices. Owners of workspaces on the plan are emailed
218 once per rise (`price_notices`), and the pricing page lists it with
219 "takes effect". Rises are never retroactive; margin protection is for
220 new usage once notice has run.
221- Staff approve or reject in sudo. A rejection needs a note.
222
223## Changing a mapping
224
Costs: Cloudflare's subscriptions read from Cloudflare each day, the estimate only until then; sudo's costs split into Costs & margin and Bill & pricing225In sudo, Costs & margin → Bill & pricing → **Mappings**: Cloudflare's product and meter
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily226prefix (as **Cloudflare's lines** lists them; `*` for the rest of the
227product), g1t's product, and optionally:
228
229- **Price meter**: the price book meter the line measures.
230- **Own meter**: g1t's count of the same units (`own_counts.meter`).
231- **Scale to g1t's count**: price one of g1t's units at as many of
232 Cloudflare's as it took (proposals as above).
233- **Drift threshold**.
234
sudo: the costs run button is named for what it does, the whole nightly analysis235It applies from the next run; **Run the analysis now** applies it at once.
One operation mapping, owned by repos; billing reads it instead of keeping its own236Every change is in the audit log (`cost_mapping`).
237
238## Which raw meters are operations
239
240There is one mapping, and the repos service owns it: `operation_mapping`
241in g1t-repos' database, one row per raw meter with `cost_operations` (how
242many operations Cloudflare bills for it) and `billable_operations` (how
243many the customer is charged for). Change it with repos'
244`set_operation_mapping` RPC (services only), or
245`npx wrangler d1 execute g1t-repos --remote` until sudo has a form. A
246change applies to counts from then on, never to what was counted.
247Billing keeps no mapping of its own: it reads repos' `git_operations`
248(already mapped) for what customers are charged, and `artifacts_usage`
249(raw counts with the mapping) for `cost_operations`. Migration 0023 drops
250the `billable_units` table 0022 made for this, which was never written.
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily251
Models' margin read -14%: usage nothing paid for is valued at price, not $0252**Charged at price** (a product's value) is what each day's usage was paid:
253charged to a card or credit, or drawn from the plan's included usage, a
254trial, a pool or a gift. Usage nothing paid for, as in a free period, is
255valued at price (cost plus the margin), since it was given away at its price
256rather than sold for nothing; so are g1t's own workspaces. Runs on a
257workspace's own model provider have no cost to g1t. Every daily run
258reconciles the whole 31-day window from what is already kept, so a change in
259how a day is valued reaches every day sudo shows.
260
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily261## Alerts runbook
262
263| Alert | Raised when | First steps |
264| --- | --- | --- |
265| Margin under the floor | A product's value against cost under `margin_floor_percent` (10%) for `alert_days` (3) days running, each with at least `min_daily_cost` | Open the product on Costs & margin. Cost up? Check proposals (approve a rise; it waits out the notice). Value down? A mapping or `revenue_map` may have moved. |
Costs: margin is measured on what was sold; comped workspaces, free periods, the trial and the pools are given away, a budget shown beside it266| All of g1t under the floor | The same for money in against the cost of what was sold: every cost less what was given away (comped workspaces, free periods, the trial, the pools), which is a budget watched in budget.rs. While less than $1 a day comes in, it says the dollars, not a percentage | Look at which products moved; check `platform` (it has no revenue of its own and grows with traffic). Before launch, with little paid usage, expect it. |
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily267| Leak | Drift of kind leak | Map the meter, or decide it is overhead. |
268| Drift | Count drift | See Drift above. Cloudflare's definitions change in beta: ask them in writing ([ARTIFACTS.md](ARTIFACTS.md), §7). |
Margin alerts measure what is sold, and say dollars when a percentage would mislead269| Costs more than it pays | A workspace's shared cost over 30 days above what its usage was priced at (`value_micros`, whoever paid: card, trial, gift or included usage) × `anomaly_factor`, at least `anomaly_floor`; not comped workspaces | Shown on Reach out as "Costs more than it pays": its usage is priced below what it costs. Abuse (Abuse & fraud page) or a gap in pricing. Not emailed. A trial or gift paying for usage does not raise it. |
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily270
271Alerts close on their own when the condition clears. Open ones are
272emailed again weekly. The red bar on every sudo page shows margin,
273overall and leak alerts.
274
Merge branch 'worktree-agent-a633ac0f7f66d419d'275## Model costs
276
277Every model call g1t pays for is an agent run's (the `claude` CLI in the
Merge the AI Gateway: Anthropic's Messages API on a workspace's tokens278sandbox, `crates/runner`) or a customer's AI Gateway request (a workspace
AI Gateway: OpenAI's format, open models, and your own providers279token with `models:write` at `models.g1t.sh/anthropic` or
280`models.g1t.sh/openai/v1`, `gateway.rs`, to Claude on Anthropic or to open
281models on Workers AI through the same AI Gateway); the only other model is
282Workers AI's embeddings for g1t's own search, which are on Cloudflare's bill
Merge the AI Gateway: Anthropic's Messages API on a workspace's tokens283(`embeddings`). How each reaches the ledger:
Merge branch 'worktree-agent-a633ac0f7f66d419d'284
285| Call | Who pays | Run and session | Ledger cost | Settled to the gateway |
286| --- | --- | --- | --- | --- |
287| Agent run through the model proxy (`services/models`) on g1t's hosted models | g1t | `runs` row; session `ms_…` in `cf-aig-metadata` | On finish, the sandbox's figure (Claude Code's `total_cost_usd`, at its own price table, cache tokens included) | Yes, every 15 minutes |
288| Agent run straight to the gateway (no `MODELS_URL`) | g1t | `runs` row; session `rs_…` in `cf-aig-metadata` (`services/runner` `gatewaySession`) | As above | Yes |
289| Agent run with no gateway (`AI_GATEWAY_ID` empty, self-hosting) | g1t's key | `runs` row, no session | The sandbox's figure | No: nothing to settle against |
Merge branch 'model-routing'290| Agent run on a workspace's own provider | The workspace | `runs` row, `billed_to = 'workspace'`, session `ms_…` (the proxy counts its tokens by it) | No model cost (none to g1t); the agent rate on `agent_tokens_own`, line `<run>/agent-own` | No; never on g1t's gateway. Closed by the cron's `settle_own_runs`, which charges tokens counted late |
Merge branch 'main' into worktree-agent-a69aeabc4b0deeb97291| AI Gateway request on g1t's key | The workspace, from included usage and AI credit (`gateway_admit` refuses at $0, over the limit, or with no plan) | No run; `gateway_requests` row; session `gw_<token>_<YYYYMMDDHH>` in `cf-aig-metadata` (`task: gateway`) | Per request: its tokens at `gateway_models` (the table, migrations 0045 and 0047: list price per million by kind, five-minute and one-hour cache writes apart, and for a model priced by prompt length such as Claude Haiku 5.5 a `threshold` above which the whole request is at the `over_` prices) plus the `gateway_models` meter's markup (0 in beta); a `usage` line with task `gateway`, reference the request id, quantity 1 | No, not yet: the session tag is there for it. **Cost** drift on `models` compares the gateway's total with these lines too. Requests billed other than by tokens (fast mode, `inference_geo`, fallbacks, server tools, containers) are refused by the proxy |
AI Gateway: OpenAI's format, open models, and your own providers292| AI Gateway request on a workspace's own provider (any model connection whose `gateway_models` takes the model) | The workspace's provider | `gateway_requests` row, `own_key = 1`, `provider` and `connection` | None | No; never on g1t's gateway |
Merge branch 'worktree-agent-a633ac0f7f66d419d'293| A sandbox that died before reporting | g1t | as its route | Charged from the gateway when settled | Yes |
294| Embeddings (indexing) | g1t | none (Workers AI) | Month-end `context` meter | No: Cloudflare's bill, `embeddings` bucket |
295| Embeddings (queries, search and agent context) | g1t | none | None: not charged, by design | No: in Cloudflare's `embeddings` line, shared out |
296
297**Settling.** A run's charge is corrected to what AI Gateway priced its
298session's requests at (`settled_cost` in `keeper.rs`). The gateway's
299figure is trusted in full: it is not held to the $100 cap on a sandbox's
300own report. It is never taken below what the sandbox reported when it
301cannot be the whole cost: a request with tokens and no cost (a model the
302gateway has no price for) or more logs than are read (2,000). Such a run
303keeps `runs.gateway_note`, its correction says why, and it raises the
Merge the AI Gateway: Anthropic's Messages API on a workspace's tokens304**Unpriced** drift. For agent runs g1t keeps no token rates of its own:
305the first figure is Claude Code's, the final one the gateway's. AI Gateway
306requests are the exception: they are charged from `gateway_models` (one
307row per model offered), which has to follow the provider's price list by
AI Gateway: OpenAI's format, open models, and your own providers308hand until they are settled like runs. It is not part of the price book's
309`price_versions`: a price change is a migration that updates the row and
Merge branch 'main' into worktree-agent-a69aeabc4b0deeb97310its `updated_at` (as 0047 did for Sonnet 5.5's cache reads), and the
AI Gateway: OpenAI's format, open models, and your own providers311`gateway_models` meter's markup is the only price-book number on it. Open
312models need `WORKERS_AI_TOKEN` (a Cloudflare API token with Workers AI on
313g1t's account) on the model proxy; without it, and without
314`AI_GATEWAY_TOKEN` holding that permission, they are refused with `503`.
Merge branch 'worktree-agent-a633ac0f7f66d419d'315
316**The daily total.** AI Gateway's analytics for the day (above) against
317the ledger's model cost is the check that nothing slips past: a model call
318with no run, or a run never settled, shows as **Cost** drift on `models`.
319The gateway's per-request `cost` is its estimate from its own price list:
320it can be off for prompt-cache tokens, for requests Cloudflare bills
321itself, and for models it has no price for. The drift's detail says when
322any of those were in the window; the provider's invoice is the last word.
323
324### Margin floor
325
326A sold charge is cost × (1 + `MARGIN_PERCENT`), rounded up (`margin_on`;
327`charge_micros` for a sandbox's own report). Terms change it only as
328follows (`Terms::discounted`, `Billing::charged`):
329
330- **Standard**: charged in full.
Billing: credits with a kind and expiry, discounts instead of comped, and safer charging331- **A 100% discount** (what was "comped"; see [Discounts](#discounts)),
332 `FREE_WHILE_BUILDING`, the plan's included usage, the trial,
Merge branch 'worktree-agent-a633ac0f7f66d419d'333 the open-source pool, and overruns g1t covers: given, and counted by why
334 (above).
335- **Custom, with a discount**: the discount comes off, and what it took
336 below cost plus the margin is written on the entry as
337 `ledger.discount_micros` and counted as given (**discount**), so the sale
338 is valued at its price and charged plus given is never under cost plus the
339 margin. On a settlement correction it moves with the charge (less than
340 nothing when the charge comes down).
341- **Goodwill credits** (overages) are separate, given by staff on purpose:
342 their margin part first, the cost only up to the cap, each audited.
343
344Every usage path goes through this: `finish_run`, settling, sandbox time,
345features and builds (`charge_feature`), and the month-end meters.
346
Billing: credits with a kind and expiry, discounts instead of comped, and safer charging347## Discounts
348
349An account's terms are standard, or custom: a **discount** from 1 to 100%,
350a limit of its own, or both, with a reason (the terms' note) and an
351optional end date. What used to be "comped" is a **100% discount**
352(`Terms::full_discount`; migration `0039_discounts_not_comped.sql` moved
353every `comped` row to `custom` at 100%, and code reads a leftover `comped`
354row as 100%). In SQL, `sales::FULL_DISCOUNT_SQL`.
355
356- **Charging.** Every charge records what the discount took off it
357 (`ledger.discount_micros`), 100% included: the entry is charged nothing
358 and the discount is its whole price. Migration 0039 backfilled the
359 discount on a 100%-discounted workspace's earlier entries that were
360 charged nothing and paid by nothing, at cost plus 20%.
361- **What a 100% discount still does as "comped" did.** The plan is on
362 without its price (`PlanKind::Internal`), trust is `internal` (no limit
363 on unpaid usage), nothing is invoiced or closed, and g1t's own spend on
364 it is held to the monthly budget (the terms' limit, at cost; see
365 [Spend caps](#spend-caps)).
366- **The statement and Usage.** The customer sees every usage line at its
367 price (`StatementLine.price_micros`: charged, plus what paid for it, plus
368 the discount), the discount per day or project and in the totals
369 (`StatementTotals.price_micros`, `discount_micros`, `discount_percent`),
370 and the CSV has price and discount columns. The Usage page measures at
371 price for a discounted account (`Usage.discount_micros`,
372 `discount_percent`).
373- **Margin.** A 100% discount's usage is given away as before, in the
374 bucket still named `comped` (`given_comped_micros`); sudo calls it
375 **100% discounts**. A partial discount's part below cost plus the margin
376 is `given_discount_micros` (**partial discounts**). Both are kept apart
377 from margin on what was sold.
378- **sudo.** The workspace's **Terms** form takes a discount (None, 25%,
379 50%, 100%, or Custom, a whole percent; the custom field shows by CSS
380 alone), a limit, an end date and the reason. Badges and filters say
381 *100% discount* or *N% off*. Each change is audited (`terms`), such as
382 `standard → 100% discount, monthly budget $150.00: g1t's own`.
383
384## Credits from g1t
385
386Staff give a workspace credit from sudo; the code is
387`services/billing/src/grants.rs`, the tables `credit_grants` and
388`ledger.credit_kind` (migration `0038_staff_credits.sql`).
389
390### Giving credit
391
392sudo → the workspace (or an enterprise, choosing one of its workspaces) →
393**Give credit**:
394
3951. **Amount**: $10, $20, $25, $50, $100, or **Custom** (up to $10,000).
396 Up to $100 it is one step; over $100, type the workspace's slug as well.
3972. **Kind**: promotional (a welcome, a referral, an event), goodwill (an
398 apology), or refund (money back for something that went wrong: say what
399 it refunds and, optionally, the day).
4003. **Expires**: never, 30, 90 or 365 days, or the end of a chosen day
401 (UTC). A refund never expires.
4024. **Note**: required. It is on the statement and in the owners' email.
403
404The form needs no JavaScript: the fields for one choice (the custom amount,
405a refund's details, the expiry date) show by CSS alone, and all show where
406`:has()` is not supported. `admin_credit` checks everything again.
407
408A grant is a `crd_…` ledger line (kind `top_up`, so never a payment) with
409`credit_kind`, and a `credit_grants` row. The balance rises at once. The
410owners are emailed through identity's `notify_owners` (the same path as
411limit notices). It is audited as `credit`. The Overages queue's one-click
412goodwill credit is a grant too, of kind goodwill.
413
414The inbox is not told: its items are threads on a repository, built from
415events, and a credit is a workspace's. That needs a workspace-level inbox
416thread first.
417
418### How it is spent
419
420Credit is spent before anything prepaid, the soonest-expiring grant first
421(never-expiring last, then the oldest). Given while the workspace owes, it
422pays what is owed first, the most recent usage first. What each grant paid
423for is never stored: `grants::replay` works it out from the ledger in order,
424so the charge paths do not know about credit and the answer is always what
425the ledger says. A charge that comes down (a settled run) gives back to
426the grant that paid last, while it can still be spent.
427
428### Expiry and revoking
429
430- **Expiry.** The daily run (`expire_credits`, before the reconciliation)
431 closes grants past `expires_at` and enters what was left as a negative
432 `crd…_expired` line; audited as `credit_expired`. A grant past its expiry
433 pays for nothing even before the run.
434- **Revoke.** sudo → the workspace's **Credits** (or **Credits & refunds**)
435 → **Revoke unused**, with why (`admin_revoke_credit`): what is left, as a
436 `crd…_revoked` line, audited as `credit_revoked`. What was spent stays
437 spent.
438
439Neither takes the balance below zero: at most the balance, if a refunded
440payment left less there than the credit.
441
442### How margin treats them
443
444| Kind | When spent | On the day it was given |
445| --- | --- | --- |
446| Promotional | The usage is valued at its price, its charge comes out of cash and is given (`given_credit_promotional_micros`) | Nothing |
447| Goodwill | The same, as `given_credit_goodwill_micros` | Nothing |
448| Refund | Paid for: cash, as any usage | Its amount (less what was revoked) comes off cash on the day it refunds, shared over that day's paid usage |
449
450A refund gives back money already collected, so counting it as given would
451make it look like a budget g1t chose to spend. Taking it off cash for the
452day it refunds says that day's sale was worth less, and counting what it
453later pays for as cash keeps money in equal to what was collected. The
454refund's day is clamped to the last 30 days, the days the reconciliation
455recomputes; an older one lands on the oldest. Refunds never expire, so
456cash taken back is never stranded.
457
458What credit paid of a month-end meter (storage, git, scans) is its own
459row on the day it was charged, since those meters are reconciled from
460snapshots. Sudo's Costs & margin lists promotional and goodwill credit
461under **Given away**, and below the statement the range's credits given,
462spent, and refunded. **Credits & refunds** (`/credits`, `admin_credits`)
463lists every grant (by kind, month, staff and workspace) and the last 12
464months by kind: given, spent, expired, revoked.
465
Usage, Billing settings and prepaid AI credit; fixes from the UX audit466### Purchased and scoped credit (prepaid AI)
Billing: credits with a kind and expiry, discounts instead of comped, and safer charging467
Usage, Billing settings and prepaid AI credit; fixes from the UX audit468`credit_grants` also has `scope` (`all`, or `models`: model usage only,
469`grants::is_model_usage`, which includes the agent rate) and `source`
470(`staff`, `purchase`, `promo_code`, `upgrade`), and `CreditKind::Purchased`.
471Spending takes credit scoped to models first, then the soonest-expiring. A
472grant scoped to models given while the workspace owes pays only what models
473owed, never other usage (`replay`). Purchased credit is money paid in: its
474ledger line is a payment (Stripe's id, never `crd…`, `credit_kind`
475`purchased`, statement kind *AI credit*), and the usage it pays for stays
476money in, never given. Staff cannot give it (`admin_credit` refuses the
477kind). Code: `services/billing/src/ai.rs`; migration `0040_ai_credit.sql`.
Billing: credits with a kind and expiry, discounts instead of comped, and safer charging478
Usage, Billing settings and prepaid AI credit; fixes from the UX audit479- **Buying.** `buy_ai_credit` opens Stripe Checkout (payment mode, $10 to
480 $1,000, a second line *Card processing fee* when the `card_fee` cost
481 setting is on, `setup_future_usage=off_session`), recorded in `checkouts`
482 with `feature = 'ai_credit'`, `amount_cents` the credit and `fee_cents`
483 the fee. The credit is entered by whichever comes first, the person coming
484 back (`confirm_ai_credit`, `?ai_credit=cs_…`) or
485 `checkout.session.completed`: both claim the row `open → paid`, the
486 grant's id is the session's id (`INSERT OR IGNORE`) and the ledger's
487 reference is unique, so a payment is credited exactly once. Expires 365
488 days after purchase (the daily `expire_credits`).
489- **Owed.** AI credit props up the balance but is money only for models, so
490 what is owed is `max(0, AI credit left − balance)` (`owed_with`; at a
491 month's close, `models_left_before` the month's start).
492- **Auto-reload.** `ai_reload` (settings; off by default) and `ai_reloads`
493 (one row per attempt). Each cron run (and a run that would be refused)
494 calls `reload_now`: below the threshold, it charges the customer's default
495 payment method off-session for the target less the balance (whole
496 dollars, at least $10, within the month's maximum), with the idempotency
497 key `reload/<workspace>/<YYYY-MM>/<n>` (a retry after a crash is the same
498 PaymentIntent), and grants purchased credit with the PaymentIntent's id. A
499 decline or a payment needing the person turns auto-reload off
500 (`failed_at`, `error`), emails the owners and audits `ai_reload_failed`.
501- **At $0.** `start_run` on g1t's models refuses with `payment_required`
502 when the workspace is on the paid plan (not a 100% discount, not an
503 enterprise), its included usage is used, and AI credit is $0 or less.
504- **Upgrade credit.** The first time a plan subscription is recorded active
505 (`features::record`), $5 of promotional credit scoped to models, id
506 `crd_upgrade_<workspace>`, expiring in a year: given, never revenue. Never
507 for a workspace with a 100% discount.
508
509### The agent rate and models' markup
510
511Price-book meters (migration 0040, each with versions and a public change):
512`agent_models` (per provider dollar; markup 20% until 2026-10-08, then 0,
513a fall applied at once), `agent_tokens` ($0 until 2026-10-22, then $0.25 a
514million tokens: a rise, after the 14 days' notice, emailed to owners on the
515plan by `tell_owners_of_rises`), `gateway_models` (markup 0 during beta),
516`card_fee_percent` (29,000 micros per dollar) and `card_fee_fixed`
517(300,000). Changing any is a price-book change, never a deploy. `finish_run`
518and `settle` charge models at `agent_models`' markup; `charge_agent_rate`
Merge branch 'model-routing'519charges the weighted tokens of the run's session since it was last charged
520(`runs.agent_tokens`, the weighted tokens charged so far, claimed with a
521compare-and-set), on a line `<run>/agent` (later `<run>/agent/<tokens>`),
522with `quantity` the weighted tokens. What it counts is the more of what
523`token_usage` holds for the session and what the sandbox reported with its
524cost (`finish_run`'s `tokens`, from Claude Code's closing `usage`), each
525weighted by kind. **Card fee switch:** sudo → Costs → Guardrails → *Card fee
Merge Stripe Tax, the card fee on card payments, and one free workspace per person526on card payments* (`cost_settings.card_fee`, `on`/`off`, on by default). It
527covers every card payment now, not only AI credit: see
528[Tax and the card fee](#tax-and-the-card-fee).
Usage, Billing settings and prepaid AI credit; fixes from the UX audit529
Merge branch 'model-routing'530**On a workspace's own model key** (migration `0041_agent_rate_own_key.sql`):
531the run keeps its model session (`runs.session_id`, `ms_…`) so the proxy's
532counts reach it, and the agent rate is charged at `agent_tokens_own` ($0 until
5332026-10-22, then $0.25 a million, a rise from nothing with its notice), on
534`<run>/agent-own` (later `<run>/agent-own/<tokens>`), `billed_to = 'g1t'`
535(g1t's own charge: it counts toward limits and spend), named *Agent rate,
536your own model key* on Usage and the statement. The model is never charged.
537`settle_runs` skips these runs (nothing on g1t's gateway); the cron's
538`settle_own_runs` closes them 5 minutes after they finish (3 hours after
539they start, for a sandbox that never reported) and charges tokens counted
540late. Runs from before have no session and are never charged the rate.
541
542### The agent rate's token weights
543
544How much each kind of token counts toward the agent rate, on g1t's models
545and own keys alike, is four price-book meters (migration
546`0042_agent_rate_weights.sql`): `agent_token_weight_input`, `_output`,
547`_cache_read` and `_cache_write`, each a weight in millionths in
Billing: cache reads count a tenth toward the agent rate548`cost_micros` (1,000,000 counts a token once). They started at 1, which is
549what the rate always counted; since 2026-10-08 cache reads count a tenth
550(100,000; migration `0044_cache_reads_count_a_tenth.sql`), as model providers
551price them. Input, output and cache writes count once. A cached agent run reads most of its context from
Merge branch 'model-routing'552cache (about 90% of its tokens on a typical Sonnet implement run), so the
553cache-read weight is the lever: at 1 the rate adds about 44% to such a run's
554model cost; at 0.1, far less.
555
Billing: cache reads count a tenth toward the agent rate556To change a weight (cache reads went to a tenth this way):
Merge branch 'model-routing'557
5581. Add the version, effective at once (a lower weight is a fall):
559
560 ```sql
561 INSERT INTO price_versions (id, meter, version, cost_micros, markup_percent, effective_at, reason, created_by, created_at)
562 VALUES ('pv_agent_token_weight_cache_read_2', 'agent_token_weight_cache_read', 2, 100000, 0,
563 '2026-10-08T00:00:00Z', 'Cache reads count a tenth toward the agent rate', 'staff', '2026-10-08T00:00:00Z');
564 ```
565
566 in a migration, or with `npx wrangler d1 execute g1t-billing --remote`
567 until sudo has a form.
5682. The daily run applies it once `effective_at` has come and writes the
569 public `price_changes` record (**Run the analysis now** applies it at
570 once). Raising a weight later is a rise: give it an `effective_at` 14
571 days out, and owners on the plan are emailed.
5723. Check `/pricing`: the agent rate's row lists the weights, and Usage's
573 agent-rate lines name them.
574
575`weighted` (`ai.rs`) rounds down to a whole token; a weight is never below 0.
576
Usage, Billing settings and prepaid AI credit; fixes from the UX audit577### Budgets
578
579The owners' spend limit is the budget. `limits.alert_levels` (comma
580separated, default every level), `limits.pause_at_limit` (default 1; off,
581100% is a warning, never a stop; the trust ceiling still stops work) and
582`limits.budget_webhook` (an https address, not g1t's; posted once per alert
583level a month by `warn_limits`). `set_budget` sets them, with `keepLimit`
584to leave the limit itself alone. A free workspace (trust `new`) has no
585spend limit to set: `set_spend_limit` and `set_budget` say so.
586
Billing: credits with a kind and expiry, discounts instead of comped, and safer charging587### Earlier credits
588
589Migration 0038 makes every earlier `Credit from g1t:` line a goodwill
590grant with no expiry: the old form asked for "a refund or goodwill" with
591no way to tell them apart, and goodwill never reads as money in.
592
Mission control shows model usage, yours and the workspace's: tokens, cost, active days, cache share, each day, and the mix593## Token usage
594
595The model proxy (`services/models`) reads Anthropic's `usage` from every
596`/v1/messages` answer, streamed or whole, on g1t's models and on a
597workspace's own provider alike (OpenAI-shaped providers are translated
598first). Count-tokens requests are not answers and are skipped. After the
599answer, it calls `record_tokens`, which adds input, output, cache reads and
600cache writes to one row per day, workspace, person, session and model in
601`token_usage` (migration `0030_token_usage.sql`). The person is who the run
602was for, from the model session's `requested_by`; never g1t's agent. A
603report that fails is dropped and never affects the answer.
604
605`token_usage` reads a window (42 days by default, 366 at most) for the
606workspace or one person: totals, every day's tokens and the active days,
607with `costMicros` the window's run charges from the ledger, measured as
Merge branch 'model-routing'608`usage` measures them. Usage's report also lists tokens by model
609(`UsageReport.models`). The agent rate is charged on these counts (above);
610a run's model is still priced from AI Gateway's logs, never from them.
Mission control shows model usage, yours and the workspace's: tokens, cost, active days, cache share, each day, and the mix611
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily612## Tables (migration `0022_costs_and_margin.sql`)
613
One operation mapping, owned by repos; billing reads it instead of keeping its own614`cost_lines`, `cost_map`, `revenue_map`, `own_counts`,
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily615`pending_days`, `margin_days`, `workspace_costs`, `cost_drift`,
616`margin_alerts`, `price_versions` (seeded with every current price as
617version 1), `price_proposals`, `price_notices`, `cost_settings` (the
618guardrails, seeded), the `actions_cache` price, and `ledger.price_version`.
619Every create is `IF NOT EXISTS` and every seed `INSERT OR IGNORE`; the one
One operation mapping, owned by repos; billing reads it instead of keeping its own620`ALTER` is applied once by D1's migration tracking. Migration
621`0023_one_operation_mapping.sql` drops `billable_units` (see above).
Merge branch 'worktree-agent-a633ac0f7f66d419d'622Migration `0036_model_costs_in_full.sql` adds `ledger.discount_micros`,
623`margin_days.given_discount_micros`, `runs.gateway_note` and the
Billing: credits with a kind and expiry, discounts instead of comped, and safer charging624`ai_gateway_requests` → `models` mapping. Migration
625`0038_staff_credits.sql` adds `credit_grants`, `ledger.credit_kind` and
626`margin_days.given_credit_{promotional,goodwill}_micros`, and backfills
Merge branch 'main' into worktree-agent-a69aeabc4b0deeb97627earlier credits (see [Credits from g1t](#credits-from-g1t)). Migration
628`0046_reset_costs.sql` adds `reset_costs` and `margin_days.given_reset_micros`
629(see [Resetting a test workspace](#resetting-a-test-workspace)).
Spend caps: a monthly budget for comped workspaces and a daily breaker on what g1t pays630
631## Spend caps
632
633Two caps keep what g1t pays for itself bounded while billing takes no
634real money. Both are measured at **cost** (what Cloudflare and the model
635providers charge g1t), never at price. Code: `services/billing/src/budget.rs`.
636Page: sudo **Costs & margin** → **g1t's own spend** (`/costs#spend`).
637
638### What counts as g1t's own spend
639
640Every charge that settles (an agent run's model cost from `finish_run` or
641AI Gateway's settlement, sandbox time from `record_sandbox`, a build from
642`charge_feature`) is split by what paid for it and g1t's part is added to
643`g1t_spend` (day, bucket, billing account):
644
645| Bucket | What |
646| --- | --- |
647| `comped` | All of a comped account's work (flagon-io) |
648| `trial` | The trial credit's share |
649| `oss` | The open-source pool's share |
650| `given` | A free workspace's overrun past its last bit of trial |
651| `unpaid` | Charged, but with no real money behind it: Stripe's test key, or `FREE_WHILE_BUILDING` |
652
653The plan's included usage and on-demand charges count as revenue only
654with live payments; in test mode they are `unpaid`. A workspace's own
655model provider costs g1t nothing and is not counted. Month-end meters
656(git, storage, scans, embeddings, the cache) are not counted here; the
657daily reconciliation covers them. Migration `0024_spend_caps.sql`
658backfills the current month from the ledger.
659
660### Caps
661
662| Cap | Variable (g1t-billing) | Default | At the cap |
663| --- | --- | --- | --- |
664| A comped account's monthly budget | `COMPED_MONTHLY_CEILING_MICROS`, or the account's own **Limit** in its terms | $150 a month | New work on the account (agents, checks, workflows, builds) is refused with "<name>'s monthly budget for g1t's own agents is used up … Staff can raise it in sudo". Runs already going finish; the per-run cap still applies to them. It lifts when staff raise the budget or the month turns (UTC). |
665| The daily breaker | `PLATFORM_DAILY_SPEND_CAP_MICROS` | $75 a day (UTC) | New agent runs on g1t's hosted models that g1t would pay for are refused until 00:00 UTC. Not paused: agents on the workspace's own model provider, checks and builds, and workspaces paying with live payments on the plan (not given by staff) or an enterprise contract. In test mode that exemption covers no one. |
666
Costs: Cloudflare's subscriptions read from Cloudflare each day, the estimate only until then; sudo's costs split into Costs & margin and Bill & pricing667`0` turns either off. Cloudflare's subscriptions are read from Cloudflare
668each day (`cf_subscriptions`) and shown on the page only;
669`CLOUDFLARE_FIXED_MONTHLY_MICROS` ($30) stands in until a read works.
Spend caps: a monthly budget for comped workspaces and a daily breaker on what g1t pays670
671The checks are cheap: `reserve` reads today's total (one indexed sum) and,
672for a comped account, its month's comped rows. Refusals come back as
673`paused`, which the compute gate honours for every plan, internal and
674enterprise included (`packages/contracts/src/compute.ts`). The runner tells
675billing whether an agent run is on hosted models (`hostedModel` on
676`reserve`); a caller that does not say is treated as hosted.
677
678### Alerts
679
680All to `COSTS_ALERT_EMAIL` (`hey@flagon.io`), through the `EMAIL` binding:
681
682- **Comped budget**: at 50, 75, 90 and 100%, once each per account and month
683 (`budget_alerts`), checked every 15 minutes. A jump past several levels
684 sends only the highest.
685- **Breaker**: at once, from the charge that trips it; if that email fails,
686 the 15-minute cron sends it (`spend_breaker.told_at`).
687
688While the breaker is open or a comped budget is used up, every sudo page
689shows a red **Spend cap** bar.
690
691### Raising and lifting
692
693- **Raise a comped budget**: sudo → the workspace → Billing → **Terms**, set
694 **Limit $** to the new monthly budget (blank goes back to the default),
695 with a note. It applies to the next start; nothing to deploy. The change
696 is in the account's audit log.
697- **Lift the breaker for today**: sudo → Costs & margin → **g1t's own
698 spend** → **Lift for today**, with why (`admin_lift_breaker`; audit action
699 `breaker_lifted`). It resets by itself at 00:00 UTC.
700- **Change a default**: edit the variable in `services/billing/wrangler.jsonc`
701 and deploy g1t-billing.
Billing keeps Stripe's view itself: the saved card on the account, missed events replayed every 15 minutes, and the endpoint kept702
sudo: reset a test workspace's billing so it starts again as a new customer; refused on a live Stripe key, for comped workspaces and for an enterprise's703## Resetting a test workspace
704
705sudo → the workspace → **Reset billing (testing)** (`admin_reset_billing`)
706returns a workspace used for testing to how a new customer starts. It
707deletes the workspace's rows from every billing table: ledger and balance,
708plan and plan payments, limits and limit requests, trial grant, invoices,
709holds, card checks, alerts sent, price notices, month-end snapshots and
sudo: the billing reset no longer names sandbox_months (dropped in 0015), checked against the migrations by a test; a failed reset says why instead of an error page710closes, storage meters, token usage, spikes, sales records and
sudo: reset a test workspace's billing so it starts again as a new customer; refused on a live Stripe key, for comped workspaces and for an enterprise's711notes, `workspace_costs`, its workspace margin alert and its own billing
712account. It keeps `own_counts` (what Cloudflare's bill is compared with)
Merge branch 'main' into worktree-agent-a69aeabc4b0deeb97713and the audit log, which records the reset with the note, the number of
714rows and what g1t had paid for. The workspace, its members and its
715repositories are identity's and repos' and stay.
716
717**What g1t paid for is kept.** The wiped usage still happened: AI Gateway
718still prices its model calls and Cloudflare still bills its sandboxes. So,
719in the same batch as the deletes, the reset writes `reset_costs`: a row per
720day and bucket the workspace had cost on (the ledger's cost, month-end
721meters' cost, and the value the reconciliation gave it), plus one row for
722the reset itself (bucket `''`, nothing in it) so every reset is on record.
723`reset_at` is the same instant as the reset's `admin_actions` entry. The
724costs run reads the rows back as the workspace's usage on their days, all
725of it given away as **testing resets**: `models` drift compares AI
726Gateway with the ledger's model cost plus what resets kept, the statement
727lists it under **Given away**, and the workspace stays in **Who g1t paid**
728with its cost given. The rows are never wiped by a later reset, and a
729rename moves them. Re-running the analysis reads the same rows and gives
730the same answer.
731
732Resets before migration 0046 kept nothing; their wiped rows are gone and
733nothing is made up for them. The costs run finds them in the audit log
734(`admin_actions`, action `reset`) with no `reset_costs` at the same
735instant, and the `models` drift detail says AI Gateway's figure includes
736usage wiped by a testing reset of that workspace on that day, rather than
737calling it a leak. syntaqx was reset on 2026-10-07 after about $8.60 of
738model usage from 2026-10-02 to 2026-10-07; that usage is in the 7-day
739window until the run of 2026-10-13 and leaves it on 2026-10-14.
sudo: reset a test workspace's billing so it starts again as a new customer; refused on a live Stripe key, for comped workspaces and for an enterprise's740
741Billing refuses it while `STRIPE_SECRET_KEY` is a live key, for comped
sudo: a billing reset runs the costs analysis again so every figure is fresh; every submit button shows it is working (CSS only); no margin percentage on less than a cent sold742workspaces, and for a workspace an enterprise pays for. It then runs the
743costs analysis again (as **Run the analysis now** does), so the margin
744figures drop the workspace's past usage at once; if that run does not
745finish, the page says so and the button does it.
sudo: reset a test workspace's billing so it starts again as a new customer; refused on a live Stripe key, for comped workspaces and for an enterprise's746
Billing keeps Stripe's view itself: the saved card on the account, missed events replayed every 15 minutes, and the endpoint kept747## Stripe
748
749Billing keeps what it needs from Stripe so reads never wait on it, and
750hears of changes three ways (`webhooks.rs`, `stripe_sync.rs`).
751
Usage, Billing settings and prepaid AI credit; fixes from the UX audit752**API version.** Every request sends `Stripe-Version: 2025-02-24.acacia`
753(`stripe::STRIPE_VERSION`), the version billing's field reads are written
754for; without it Stripe answers at the account's default. Webhook events
755come at the destination's own version: billing reads an invoice's
756subscription from `subscription` or `parent.subscription_details.subscription`.
757Raising the version is a code change: read Stripe's upgrade notes for every
758field billing reads.
759
760**Failures.** No Stripe failure reaches a page as a 500: each payment page
761(`page_opened`), the portal, confirmations and plan changes turn it into
762`stripe::friendly` (Stripe's own message, never the request or a key), and
763log the full error with the workspace. Every payment page is recorded
764through one insert (`CHECKOUT_INSERT`), checked against the migrations by
765`every_checkout_insert_fills_the_table`, and has an idempotency key
766(`page/<purpose>/<workspace>/…/<10-minute bucket>`).
767
Stripe's webhook secret is a Worker secret, STRIPE_WEBHOOK_SECRET, from a destination made in Stripe's dashboard768**The webhook.** A destination made in Stripe's dashboard (Developers →
769Webhooks → Add destination) with the endpoint URL
770`https://api.g1t.sh/stripe/webhook`, in the mode of billing's key (at
771launch, make one in live mode and put its secret). Its signing secret
772(`whsec_…`: the destination, Signing secret, Reveal) is the billing
773Worker's secret:
774
775```sh
776cd services/billing && npx wrangler secret put STRIPE_WEBHOOK_SECRET
777```
778
779Without it every event is refused with 400. After rolling the secret in
780Stripe, put the new one; during the roll Stripe signs with both, so there
781is no gap. The event list need not be exact: billing adds any event it
782handles that the destination does not send (daily, or **Fix destination**
783in sudo → Stripe), and enables it again if Stripe disabled it. It never
784changes the secret. sudo → Stripe shows whether the secret is set, the
785destination and its status, missing events, and the latest events.
786Events are claimed once each in `stripe_events`; a handler that fails
787forgets its claim, and Stripe retries.
Billing keeps Stripe's view itself: the saved card on the account, missed events replayed every 15 minutes, and the endpoint kept788
789**What is kept, and how it stays current**
790
791| Kept | Where | Refreshed by |
792| --- | --- | --- |
793| Saved card (brand, last 4, expiry) | `accounts.card_*`, `card_synced_at` | `customer.updated`, `payment_method.*`, `setup_intent.succeeded`; forgotten when g1t sets a default card, so the next read asks once; the daily pass for cards older than 7 days |
794| Plans | `subscriptions` | `customer.subscription.*`, `invoice.*`; a read asks Stripe at most hourly once a period is over; the daily pass for rows older than a day |
795| Payments, refunds, disputes | ledger, `checkouts`, invoices | their events |
796
797**Every cron run (every 15 minutes)** replays missed events: Stripe's event
798list from an hour before `stripe_sync.through`, oldest first, through the
799same once-only claim. `through` moves to 5 minutes before now when all were
800handled, back to the first failure otherwise, and stays when more than
8011,000 events were listed. Claims stuck at `handling` for 10 minutes are
802dropped so the replay retries them. The first run reads 3 days back.
803
Stripe's webhook secret is a Worker secret, STRIPE_WEBHOOK_SECRET, from a destination made in Stripe's dashboard804**Daily** (`keeper::DAILY`): the destination at billing's address is
805enabled again if Stripe disabled it and given any missing event, audited as
806`stripe`/`webhook`; then up to 25 stale cards and 25 stale plans are read
807again.
Billing keeps Stripe's view itself: the saved card on the account, missed events replayed every 15 minutes, and the endpoint kept808
809**What still calls Stripe on a request**: starting a payment page, a plan
810or a card check; opening the billing portal; settling a page the person
811came back from; renaming a workspace (the customer's name). Nothing a page
812view reads.
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar813
Merge Stripe Tax, the card fee on card payments, and one free workspace per person814## Tax and the card fee
815
816Owner decision 2026-10-08. Code: `services/billing/src/tax.rs` (what is
817kept, the address hold), `stripe.rs` (every request's tax fields),
818`invoices.rs`, `webhooks.rs`, `ai.rs`; migration
819`0043_tax_and_card_fees.sql`.
820
821### What Stripe is asked
822
823Every price excludes tax (`tax_behavior=exclusive`) and carries tax code
824`txcd_10103001` (software as a service, business use; the card fee too,
825since a fee for paying for a sale follows the sale). Fields are valid for
826`2025-02-24.acacia`.
827
828| Request | Tax | Card fee |
829| --- | --- | --- |
830| Checkout, plan or Security (`subscription_fields`) | `automatic_tax[enabled]`, `billing_address_collection=required`, `tax_id_collection[enabled]`, with a customer `customer_update[address]=auto` and `[name]=auto`; the subscription keeps automatic tax for every renewal | A second recurring line, *Card processing fee* |
831| Subscription on a saved card (`saved_subscription_fields`) | `automatic_tax[enabled]`; a customer Stripe Tax cannot place fails, and the person is sent to Checkout, which asks for the address | `items[1]`, the `card_fee` product |
832| Checkout, prepay (`prepay_fields`) | As above | By card only; none by bank transfer |
833| Checkout, AI credit (`credit_fields`) | As above | Its own line |
834| Checkout, card check (`card_check_fields`) | Setup mode: nothing charged, nothing taxed. `billing_address_collection=required`, and the card's address is copied onto a customer with none (`fill_address`) | — |
835| Auto-reload (`charge_saved`) | `POST /tax/calculations` first (credit and fee as lines), the PaymentIntent for the total, then `POST /tax/transactions/create_from_calculation` with the PaymentIntent as reference; a refund reverses its share (`create_reversal`, `mode=partial`) | In the calculation and the amount |
836| Workspace invoice, month close and threshold (`invoice_workspace`) | `automatic_tax[enabled]` on the draft; each item `tax_behavior`, `tax_code` | An item *Card processing fee*, when the default payment method is a card |
837| Enterprise invoice (`invoice_enterprise`) | The same | Never |
838| Products (`Stripe::product`) | Made with `tax_code`; one found without it is given it | The `card_fee` product, `metadata[g1t]=card_fee` |
839
840### What is kept
841
842A payment credits the balance with what it paid for, never its tax or fee:
843prepay credits the page's `amount_subtotal` less the fee line
844(`credit_prepayment`), a workspace invoice `amount_paid − tax − fee`
845(`credit_invoice`), AI credit its credit amount, and `plan_payments` the
846plan's invoice less its tax and *Card processing fee* lines
847(`stripe::invoice_split`). Each payment's tax and fee are rows in
848`tax_and_fees` (`<reference>/tax`, `<reference>/card_fee`; the enterprise's
849account id in `workspace` for its invoices), with the PaymentIntent, so a
850refund (`charge.refunded`) gives back the balance, tax and fee in
851proportion (`tax::refund_split`, negative rows under `refund/<charge>/…`).
852`workspace_invoices.fee_micros` and `tax_micros` sit beside each usage
853invoice.
854
855- **Statement.** *Tax* and *Card processing fees* are their own lines per
856 day (`StatementLine.passed_micros`), never in `charged_micros`; totals
857 `tax_micros`, `card_fee_micros`. The CSV has a row a day for each.
858- **Margin.** Cash never holds them, so margin is untouched. sudo → Costs
859 shows **Tax collected** and **Card fees passed on** for the range
860 (`OverallMargin.tax_collected_micros`, `card_fees_micros`). Tax is owed to
861 the authorities: file it from Stripe Tax's reports, never from g1t's.
862
863### No address
864
865Stripe Tax needs a country (in the US a ZIP code, in Canada a postal code
866or province: `stripe::address_places_customer`). Before a workspace
867invoice is drafted, g1t checks the customer; without an address, or when
868Stripe leaves the draft at `requires_location_inputs` or refuses with
869`customer_tax_location_invalid`, nothing is charged:
870`accounts.tax_address_needed_at` is set, the owners are emailed once
871(`notify_owners`), and Billing shows **Add a billing address**. Saving
872Invoice details with an address Stripe Tax can use clears it, as does a
873charge that goes through. Work is not stopped for it; the limits still
874apply. Auto-reload without an address fails like a declined card (turned
875off, owners told). An enterprise's invoice is not sent without an address:
876sudo → the enterprise → Invoices → **Billing address** (`admin_enterprise_address`,
877with its tax ID; audited `billing_address`).
878
879### The card fee
880
881`card_fee_cents` grosses Stripe's fee up so the amount paid for is left
882after it: `(amount + 30¢) / (1 − 2.9%)`, rounded up; $0.91 on $20, $1.06
883on $25. It is worked out on the amount before tax, so Stripe's fee on the
884tax itself (a few cents) is g1t's. It is shown before paying: the plan card
885and pricing page (`Plan.card_fee_cents`), AI credit (*Card processing fee
886$1.06, plus tax where it applies*), Prepay. Never on a bank transfer or an
887enterprise's (`send_invoice`) invoice. Meters stay at cost + 20% and models
888at the provider's price plus the agent rate (price versions in migration
8890040); the fee is passed through, not margin.
890
891### Tax-exempt customers and tax IDs
892
893g1t never sets `tax_exempt`. For a customer who sends an exemption
894certificate, set it in Stripe's dashboard (Customers → the customer → Tax
895status: Exempt, or Reverse charge); Billing then says so. Tax IDs come
896from Checkout (`tax_id_collection`) or Invoice details (`set_billing_details`,
897validated against Stripe's types in `details::TAX_ID_TYPES`); Stripe checks
898EU VAT numbers and Stripe Tax applies a reverse charge where it should.
899Billing shows Stripe's `verification.status`.
900
901### In Stripe's dashboard (not done by g1t)
902
9031. **Settings → Tax → Get started**: turn on Stripe Tax in live and test
904 mode.
9052. **Origin address**: Flagon, Inc.'s head office address.
9063. **Default tax code**: Software as a service, business use
907 (`txcd_10103001`); **default tax behavior**: exclusive.
9084. **Registrations**: add each jurisdiction where Flagon is registered to
909 collect (its home state at least; then states as thresholds are
910 crossed, which Stripe Tax's monitoring flags; the EU's OSS and the UK
911 if selling there). Stripe collects only where a registration exists.
9125. **Customer portal**: tax ID and address updates are already allowed
913 (`portal_configuration`).
9146. Refund an invoice through a **credit note**, so its tax is reversed in
915 Stripe Tax.
916
917## Free workspaces
918
919Owner decision 2026-10-08: one free workspace per person, and a free
920workspace adds no one. Billing answers one question, `free_workspaces`
921(`credits.rs`): which of the given workspaces are on no paid plan
922(`plan_kind_for` is `Free`). The plan, an enterprise's terms and a 100%
923discount (flagon-io) count as paid; with payments off nothing is free.
924Identity asks it (`services/identity/src/paid.rs`) and refuses with
925`payment_required`:
926
927| Where | Refused when |
928| --- | --- |
929| `create_workspace` | The person owns any free workspace. Several from before are kept (grandfathered); none can be added until each is paid for or deleted. |
930| `add_member`, `invite_member` | The workspace is free. |
931| `add_collaborator` | Someone outside a free workspace (a username who is not a member, or an address). Members' roles are fine. |
932| `accept_invite`, `respond_repo_invitation` | The invite's workspace (or repository's) is free now: it waits. A sign-up with such an invite makes the account without joining. |
933
934`@g1t` is never counted as someone added. If billing cannot be asked, the
935change is refused for now ("try again"), never let through. The site says
936so first (New workspace, People, a repository's Access, from the same RPC);
937the API and MCP pass identity's refusal on as `402`.
938
Teams and CODEOWNERS, labels and milestones, dependency updates, the security suite, and a clearer top bar939## The Security and quality activation
940
941A second monthly subscription a workspace can hold beside the plan
942(`Feature::Security`, `feature = 'security'` in `subscriptions`). It turns
943on the security suite's paid features for the workspace's private
944repositories: custom secret patterns, validity checks, delegated bypass,
945code scanning, dependency review and the security overview. Public
946repositories have them free; secret scanning, push protection,
947vulnerability alerts and security updates are free everywhere.
948
949- **Price.** The price book's `security_activation` meter (unit
950 `workspace-month`, `source` `list`, markup 0): 10,000,000 micros, $10,
951 from migration `0037_security_activation.sql` with its first
952 `price_versions` row and a public `price_changes` record. `plan()` and the
953 `prices` RPC read it (`features::security_plan_at`); if the price book
954 cannot be read, $10. Nothing in the web app hard-codes it. A change is a
955 new price version, like any other: noticed on the pricing page and
956 applied from its `effective_at` to new subscriptions. Subscriptions
957 already running keep the amount Stripe has until they are changed in
958 Stripe.
959- **Stripe.** Its own subscription and its own product, tagged
960 `metadata[g1t]=security` (the plan's is `plan`). Started from the Billing
961 page with `subscribe` (`feature: security`) on the checked card, or
962 through Checkout; ended with `cancel_subscription` (`feature: security`)
963 at the period's end. Its invoices count in `plan_payments` like the
964 plan's, as paid revenue.
965- **Who has it.** `has_feature(workspace, security)`: on with an active
966 subscription, with comped terms or as an enterprise's workspace, or when
967 Stripe is not configured. The plan's allowance (`allowances.plan`) does
968 not include it. Refusals are `PaymentRequired` with the price from the
969 price book and the Billing page's address.
970- **Where it is checked.** The security service, on each paid call for a
971 private repository (`suite::entitled`) and before using custom patterns
972 in a push (`patterns_for`). When billing cannot be reached it is taken
973 as off: a paid feature waits rather than running unpaid.
974- **Fixes.** "Fix with g1t" runs g1t's agent, charged as agent usage, never
975 to the activation.
976- **Sales figures.** MRR in sudo counts `feature = 'plan'` only; the
977 activation's subscriptions are not in it yet.

This file's history is long; its oldest lines are credited to the oldest commit read.