---
title: AI spend
description: See every daily AI limit in your organisation on one page, with a bar for what each product spent today, and change any of them in place if you are an owner or admin.
section: cloud
order: 29
audience: [user, operator, agent]
sources: ["services/nodejs-cloud/app/(console)/(shell)/spend/page.tsx", services/nodejs-cloud/components/spend, services/nodejs-cloud/lib/spend, services/nodejs-cloud/app/api/admin/v1/spend, services/nodejs-cloud/lib/orgs/mcp-spend.ts]
features: [cloud.console-spend, cloud.api-admin-spend]
---

Several Venom products call your models on their own: People reads news about your contacts, Notes suggests notes, Chat agents answer in channels, Outreach drafts messages, Support drafts replies. Each has its own daily limit so one of them cannot spend the whole day's budget. The **AI spend** page (`/spend`, under **Usage & cost** in the console's sidebar) shows all of those limits together: one bar per limit with what was spent this period against it, grouped by product.

Any member can open the page. Owners and admins can change each limit on its row. The organisation's monthly spend cap is on the [Usage page](/docs/cloud/console-usage.md) and per-member, per-key and per-gear budgets are on [Budgets](/docs/cloud/console-budgets.md); the page links to both and does not repeat them.

## Raise a limit {#example}

People's overall limit stopped photo capture today and you want it to keep working:

1. Open `/spend`. The **People** card shows *People AI (overall)* with **Reached** next to its figures.
2. Type `5` in its **Limit a day (USD)** field and press **Save**.
3. The bar redraws against $5. Capture works again at once.

From an agent connected to the Admin MCP server, the same change is:

```json
{ "name": "admin_spendlimit_set", "arguments": { "limit": "people", "value": 5 } }
```

## What the page shows {#limits}

Each row has a bar, the figures (spent of limit), when it starts again, and a badge when it needs attention: **Reached** (spent is at or past the limit), **Limit is 0: nothing runs**, or **Off** (the feature is switched off). The badge carries the state in words, not only the bar's colour.

| Product | Limit | Unit | Counted from | Also stops at | Default | Changed |
|---|---|---|---|---|---|---|
| People | People AI (overall) | USD a UTC day | every People AI call | | $1 | here, People → Settings |
| People | Signals | USD a UTC day | signals' calls | People AI (overall) | $0.50 | here, People → Settings |
| People | Classification | USD a UTC day | classification's calls | People AI (overall) | $1 | here, People → Settings |
| People | Enrichment | USD a UTC day | paid PDL and Apollo lookups (not AI) | | $1 | here, People → Settings |
| Notes and Forms | AI in Notes | USD a UTC day | note suggestions, imports, entities and Forms drafts | | $2 | here, Notes settings |
| Agents and bots | Chat agents | USD a UTC day | Venom's and Tasks agents' channel runs | | $1 | here |
| Agents and bots | Agents on computers | USD a UTC day, each | each agent's computer runs | | the Chat agents limit | on each agent |
| Agents and bots | Bots | USD a UTC day, each | each bot's runs | | $2 | on each bot |
| Outreach | Outreach drafts | USD a UTC day | the drafter's calls | | $2 | here |
| Tasks | Pulse summaries | USD a UTC day | the weekly digest's summary | | $1 | here |
| Tasks | AI triage | USD a UTC day | triage suggestions | | $1 | here |
| Support | Support drafts | USD a UTC day | drafted replies | | $1 | here |
| Routing and quality | Model quality checks | USD a month | quality runs and autopilot checks | | none (off) | here, [Model quality](/docs/cloud/console-quality.md) |
| Routing and quality | Shadow tests | USD a UTC day, each | each running test's samples | | set per test | on each test |
| Safety and debugging | Guardrail drafting | drafts in 24 hours, per person | your own drafts | | 20 | here, [Guardrails](/docs/cloud/console-guardrails.md) |
| Safety and debugging | Injection shield classifier | checks in 24 hours | the shield's model checks | | 200 | here, [Guardrails](/docs/cloud/console-guardrails.md) |
| Safety and debugging | Failure classifier | sessions a UTC day | classified sessions | | 50 | here, [Failures](/docs/cloud/console-failures.md) |

- A limit under another ("Also stops at") is indented under it: Signals and classification stop at their own limit or at People's overall one, whichever comes first.
- Daily figures start again at midnight UTC. The guardrail classifiers count the last 24 hours, so they free up gradually. Quality checks start again on the first of the month.
- For agents, bots and shadow tests every item has its own limit. Their row shows how many there are, how many set their own, how many are at it, and the total spent; the bar is the busiest item's share of its limit. Change those on each agent, bot or test.
- An agent's or bot's run on a computer counts from where its model ran: on the daemon's own keys, what the daemon reported; on your organisation's cloud keys ([Where a daemon chat's model runs](/docs/agents/model-source.md)), what the cloud booked for the run, with a model nothing prices counted at the dearest model you serve. The figure here is the one the daily limit checks.
- A product your organisation does not have (Notes, Tasks, Computers, Agents) shows no row.

## Bounds {#bounds}

A value outside its bounds is brought inside them, as People's limit always was.

| Limit | From | To |
|---|---|---|
| People AI, Signals, Classification, Chat agents, Outreach, Pulse, AI triage, Support | $0 | $100 |
| Enrichment | $0 | $1,000 |
| AI in Notes | $0 | $200 |
| Model quality checks | $0 | $100,000 a month |
| Guardrail drafting | 1 | 500 |
| Injection shield classifier | 1 | 10,000 |
| Failure classifier | 1 | 500 |

Dollar limits keep cents; counts are whole numbers. A limit of $0 stops that feature until you raise it.

## Who can do what {#roles}

| Action | Who |
|---|---|
| See `/spend` and `admin_spendlimit_list` | every member (`usage:read`) |
| Change a limit | owners and admins (`org:manage`), in person |

An agent acting for a member, such as a quick chat or an agent's run on a computer, cannot change a limit, even for an owner. Every change goes through the product's own setting, so it is checked there again and written to the [audit log](/docs/cloud/console-audit.md) with the value before and after (`chat.ai.limit`, `outreach.ai.limit`, `support.ai.limit`, `tasks.triage.limit`, `people.ai.limit`, `globe.loop.settings` and the other products' own entries).

## API and MCP {#api}

| Call | What |
|---|---|
| `GET /api/admin/v1/spend/limits` | every limit with this period's spend, its state, the limit it also stops at, when it resets and whether you may change it |
| `PUT /api/admin/v1/spend/limits/{key}` with `{"value": 5}` | change one limit, then the whole table again |
| `admin_spendlimit_list` | the same table, as text or JSON (`format: "json"`) |
| `admin_spendlimit_set` with `limit` and `value` | change one limit |
| `GET /api/v1/admin/spend/limits`, `PUT /api/v1/admin/spend/limits/{limit}` | the tools' REST twins, with a personal API key or an OAuth token holding `admin:read` (and `admin:write` to change) |

The keys of the limits you can change:

```text
people  people_signals  people_classify  people_enrichment  globe  chat_agents  outreach
tasks_pulse  tasks_triage_ai  support  quality  guardrail_drafting  injection_shield  failure_classifier
```

The per-item rows (`computer_agents`, `bots`, `shadow`) are read-only here. The tools are in the Admin MCP server's `keys-and-providers` set ([Workspace MCP](/docs/cloud/workspace-mcp.md)). The page reads figures and limits only: never a prompt, a transcript, a key, or the name of someone's bot or agent.

## Troubleshooting {#troubleshooting}

| Message | Cause and fix |
|---|---|
| Only an owner or admin of the organisation changes its AI limits. | Ask an owner or admin. Your role can read the page. |
| An agent cannot change the organisation's AI limits | An agent made the call. Change it yourself on `/spend`. |
| This limit is set on each item | Agents, bots and shadow tests each have their own limit: open the linked page. |
| A product's work stops with "daily AI limit … is reached" | The product's bar shows **Reached**. Raise the limit here or wait for midnight UTC. |
