| 1 | --- |
| 2 | title: Spend |
| 3 | description: What the workspace spent and where it went, by agent, person, channel, model and product; suggestions to spend less that are checked against your own past work; the budgets from the workspace down to one task; a receipt for every task; and how it is priced. |
| 4 | --- |
| 5 | |
| 6 | **Spend** is the front of the workspace's money: |
| 7 | `g1t.sh/<workspace>/-/spend`, first under **Money** in the Workspace |
| 8 | sidebar. It says what was spent over a period, where it went, the budgets |
| 9 | that hold it, and what each task cost. Every figure on it comes from the |
| 10 | workspace's usage and agent records; nothing is estimated. |
| 11 | |
| 12 | **Usage** has every meter in detail, and **Billing and plans** has the |
| 13 | plan, the spend limit, payments and the statement. Spend links to both. |
| 14 | |
| 15 | ## Whose spend |
| 16 | |
| 17 | | View | Who sees it | What it counts | |
| 18 | | --- | --- | --- | |
| 19 | | **Workspace** | Owners and [billing managers](/guides/workspaces/#roles-that-add-to-a-member) | Everything the workspace used, every product, and every agent's work. | |
| 20 | | **You** | Everyone | What agents spent working for you: your chats with agents and the sessions you started. | |
| 21 | |
| 22 | Owners and billing managers switch between them with **Workspace** and |
| 23 | **You** at the top of the page. Everyone else sees **You**. |
| 24 | |
| 25 | ## The period |
| 26 | |
| 27 | Choose **This month**, **Last month**, **30 days** or **7 days**. Days are |
| 28 | UTC. Budgets are always this month's, whatever the period. |
| 29 | |
| 30 | ## The figures at the top |
| 31 | |
| 32 | | View | Figures | |
| 33 | | --- | --- | |
| 34 | | **Workspace** | **Spent**: usage at price over the period, every product, the same figure as [Usage](/guides/usage-and-billing/#the-usage-page). **Agents**: every agent's replies and sessions. **Charged**: what the workspace was charged this month, against its spend limit when it has one. | |
| 35 | | **You** | **Agents for you** over the period. **Your budget** this month, and what is spent against it. **Your chats with agents**: the replies agents wrote you. | |
| 36 | |
| 37 | **By day** shows the period one day at a time. Hover or focus a day for |
| 38 | its amount. |
| 39 | |
| 40 | ## Where it went |
| 41 | |
| 42 | Slice the period's spend with the tabs: |
| 43 | |
| 44 | | Slice | Shows | |
| 45 | | --- | --- | |
| 46 | | **Agents** | Each agent's share. A session's helpers and subagents count on the agent that started it. Open one for its own Spend tab. | |
| 47 | | **People** | Who asked. Routines and agent-to-agent work have no asker. Workspace view only. | |
| 48 | | **Channels** | Where it was asked: each channel, direct messages together, and channels you aren't in together, without their names. | |
| 49 | | **Models** | Each model's share. | |
| 50 | | **Kind of work** | Chat replies, sessions, routines, helping colleagues and subagents. | |
| 51 | | **Products** | Every product at price, from Usage: Agent, Sandboxes, AI Gateway, Deployments, Git & storage, Packages, Security & quality and Search. Workspace view only. | |
| 52 | | **Extensions** | Coming: what each installed extension runs, once extensions can be installed. | |
| 53 | |
| 54 | A slice is part of the page's address (`?by=channel`), so you can share it. |
| 55 | |
| 56 | ## Spend less, keep quality |
| 57 | |
| 58 | Every week, g1t checks each agent's finished sessions and asks one |
| 59 | question: would a cheaper [effort](/guides/agents/#effort) have done the |
| 60 | work as well? The answers are under **Spend less, keep quality**, on Spend |
| 61 | for every agent and on each agent's **Spend** tab for that agent. |
| 62 | |
| 63 | How a suggestion is made: |
| 64 | |
| 65 | 1. **What counts.** Each agent's own sessions that finished in the last 28 |
| 66 | days, at the effort level each recorded, with what each was charged as |
| 67 | budgets count it: the model at the provider's price and the |
| 68 | [agent rate](/guides/usage-and-billing/#the-agent-rate). A session is |
| 69 | **accepted** when it finished with nobody having to step in: no one |
| 70 | steered it, and it was not stopped or failed. Helpers and subagents |
| 71 | count as part of the session that started them. |
| 72 | 2. **What is compared.** The level the agent runs at now (for Auto, the |
| 73 | level most of its sessions ran at) against the level below. |
| 74 | 3. **When it suggests.** Both levels have at least 10 sessions; the |
| 75 | cheaper level's sessions were accepted at least as often, give or take |
| 76 | 5 points, even on a cautious estimate; and a typical session at the |
| 77 | cheaper level cost at most 85% of one at the current level. |
| 78 | 4. **What it says.** The suggestion names the numbers it rests on, such as |
| 79 | *At Medium, 12 of its 12 sessions finished with nobody stepping in, |
| 80 | against 19 of 20 at High; a typical one cost $0.21 instead of $0.48*, |
| 81 | and about what it would save a month at the agent's recent pace. |
| 82 | |
| 83 | When either level has fewer than 10 sessions, Spend says **Not enough |
| 84 | history yet**, with how many it has, instead of suggesting anything. When |
| 85 | the cheaper level was measured and did worse, or would save little, |
| 86 | nothing is shown. |
| 87 | |
| 88 | To act on one, as an owner: |
| 89 | |
| 90 | 1. Open **Spend** and find **Spend less, keep quality**. |
| 91 | 2. Read the reason under the suggestion. |
| 92 | 3. Choose **Apply** to change the agent's effort (a new version of the |
| 93 | agent, recorded in the [audit log](/guides/audit-log/) with who applied |
| 94 | it), or **Dismiss** to put it away. A dismissed suggestion is not made |
| 95 | again for the same setting. |
| 96 | |
| 97 | Applied and dismissed suggestions stay listed under **Decided lately** for |
| 98 | 30 days. A suggestion made for a setting that has since changed is put |
| 99 | away at the next check. |
| 100 | |
| 101 | ## Budgets |
| 102 | |
| 103 | Budgets nest, from the widest to the narrowest. Before every agent reply |
| 104 | and session step, g1t checks each level, and the first one that is used |
| 105 | up stops the work. Each row on Spend shows the budget, what is spent |
| 106 | against it this month, and what happens at 100%. |
| 107 | |
| 108 | | Level | Set by | At 100% | |
| 109 | | --- | --- | --- | |
| 110 | | **The workspace** | Owners and billing managers, under [Billing → Your spend limit](/guides/usage-and-billing/#your-spend-limit). | New work stops until the month turns or an owner raises it: agents, workflows, builds and deploys. With **Pause usage at 100%** off, owners are alerted and work goes on. | |
| 111 | | **All agents together** | Owners, here or under **Agents → Budget**. | No agent takes new work until the 1st, or until an owner raises it. | |
| 112 | | **Each person** | Owners, here. A default for everyone, and a budget of their own for anyone. | Agents take no new work for that person until the 1st, and say so where they were asked. | |
| 113 | | **Each agent** | Owners, on the agent's **Profile**. A new agent starts with the default set here. | That agent takes no new work until the 1st, or the next day for a daily cap. | |
| 114 | | **Each task** | Owners, here: the cap a session starts with. The plan's caps on one run and one issue are under [Billing → Caps](/guides/usage-and-billing/#caps). | The session stops at **Needs approval**, and an owner decides whether it goes on. | |
| 115 | |
| 116 | A team can also have a budget for what its agents spend together, set on |
| 117 | the team's own settings and checked after a person's budget. See |
| 118 | [lead, channel and budget](/guides/teams/#lead-channel-and-budget). |
| 119 | |
| 120 | Work already running finishes, so spend can go slightly past a budget. |
| 121 | |
| 122 | ### A person's budget |
| 123 | |
| 124 | A person's budget counts what agents spend on work that person asks for: |
| 125 | the replies agents write them and the sessions they start, with |
| 126 | everything those sessions bring in. Each person can see their own on |
| 127 | **You**, and in the top bar. |
| 128 | |
| 129 | To set budgets, as an owner: |
| 130 | |
| 131 | 1. Open **Spend** and find **Budgets**. |
| 132 | 2. Choose **Change** on **All agents together**, **Each person** or |
| 133 | **Each task**, fill in the amounts in dollars, and choose **Save |
| 134 | budgets**. An empty budget means none. |
| 135 | 3. To give one person their own budget, choose **Set** on their row, or |
| 136 | **Give someone their own budget**, and enter it. `0` means no budget at |
| 137 | all for them; empty puts them back on the default. |
| 138 | |
| 139 | ## Receipts |
| 140 | |
| 141 | **Costliest tasks** lists the period's most expensive sessions. Open one |
| 142 | for its receipt, at `g1t.sh/<workspace>/-/spend/receipts/<session>`: |
| 143 | |
| 144 | - Each session of its tree: the agent's own, and every helper and subagent |
| 145 | it brought in, with their tokens, the model at the provider's price, and |
| 146 | what each was charged. |
| 147 | - Each session's model and the [effort](/guides/agents/#effort) it ran at. |
| 148 | - **Models, at the provider's price**: every session's model answers |
| 149 | together. |
| 150 | - **g1t's agent rate**: the rest, at the rate in the price book. |
| 151 | - **Total, as budgets count it**. |
| 152 | |
| 153 | On your own model key, your provider bills the model and only the agent |
| 154 | rate is charged; the receipt says so. A session from a conversation you |
| 155 | aren't in shows what it cost, not what it was about. |
| 156 | |
| 157 | ## How it's priced |
| 158 | |
| 159 | The page's last section reads the [price book](/guides/usage-and-billing/#how-prices-are-set): |
| 160 | |
| 161 | | | Price | |
| 162 | | --- | --- | |
| 163 | | **Models** | What the model provider charges, with no markup. | |
| 164 | | **Agent rate** | Per million tokens an agent's run uses, for what g1t runs around every model call: the model gateway, secrets, routing, context and pass-through to your own provider. | |
| 165 | | **Everything else g1t runs** | Sandboxes, builds, hosting, storage, search and security scans, at cost plus 20%. | |
| 166 | | **Your own model keys** | Your provider bills the model; the agent rate still applies. Local models the same. | |
| 167 | | **Your own runners** | $0. | |
| 168 | | **People** | No seats. | |
| 169 | | **Spending less** | [Effort](/guides/agents/#effort) on each agent, and the weekly suggestions above. | |
| 170 | |
| 171 | ## In the top bar |
| 172 | |
| 173 | Your spend this month sits in the page's header, beside **Ask g1t**: what |
| 174 | agents did for you against your budget. Choose it for where it went, by |
| 175 | kind of work and by agent, and a link to Spend. Owners and billing |
| 176 | managers can switch it to **Workspace**: what the workspace was charged |
| 177 | this month against its spend limit, and its agents' spend. The choice is |
| 178 | remembered on this browser. On a phone, open **Spend** from the Workspace |
| 179 | sidebar instead. |
| 180 | |
| 181 | ## Next |
| 182 | |
| 183 | - [Effort](/guides/agents/#effort): how hard each agent works, and what each level costs. |
| 184 | - [Agent budgets and spend](/guides/agent-budgets/): who pays for a session tree. |
| 185 | - [Usage and billing](/guides/usage-and-billing/): the spend limit, caps, AI credit and rates. |
| 186 | - [Sessions](/guides/agent-sessions/): caps and approving more spend. |