# How a job spends AI credit

> What leaves your machine, how each step reserves its cost, and where a job always stops.

Status: Available · Level: Deeper · About 5 minutes · Canonical: https://paracrux.com/docs/how-a-job-spends-credit

## You need

- A paid seat

## You'll come away with

- Know the order credit is spent in
- Follow one job step by step
- Know the two limits that stop every job

## Step by step

1. **Know your balances.** Each paid seat includes £5 of managed AI credit per UTC month, spent first and refilled on the 1st. Top-ups of £10, £25 or £50 are spent next and carry forward. My account shows both as meters.
2. **Know what is sent.** Nothing is sent until you tick Allow sending design context to OpenAI when I press Send, in Settings, Agent API. It is off by default, and Send stays disabled until you allow it. Then your message goes with the design intent and instructions, reference names and fingerprints (never file contents), the latest worksheet inspection and the results of tools the Agent runs. To see the context, open the Instructions tab and choose Preview.
3. **Follow a step.** Each step makes at most one tool call. Before it runs, credit is held; afterwards the hold settles to actual usage and the rest is returned. The app lists its rates under Settings, Agent API: per million tokens, £40 input, £4 cached input, £50 cache writes and £200 output, including reasoning.
4. **Know where it stops.** A job stops at £2 or eight steps, whichever comes first. It also stops when an outcome is uncertain, and it never retries a CAD step on its own.

## Walkthrough

The limits and rates are the app's own. Token counts and held amounts are examples, so each cost is calculated from example usage, not measured.

- **You:** Find a 50 mm square tube, then make me a 100 × 50 × 25 mm block.
- **Balances before:** Included this month = £5.00; Top-ups = £0.00; Job cap = £2.00 or 8 steps
- Provenance: illustrative

- **Tool call:** `design_profile_search` {"family":"square_hollow_section","query":"50x50"}
- **Step 1:** Tokens (example) = 5,000 in · 300 out; Cost at the app's rates = £0.20 + £0.06 = £0.26; Held first (example) = £0.60; Returned = £0.34
- Provenance: calculated

- **Tool call:** `solidworks_create_box` {"width":0.1,"depth":0.05,"height":0.025}
- **Step 2:** Tokens (example) = 6,000 in · 400 out; Cost at the app's rates = £0.24 + £0.08 = £0.32; Held first (example) = £0.60; Returned = £0.28
- Provenance: calculated

- **Agent:** Done in two steps. Three sizes found; the block is saved and checked.
- **Job total:** Spent = £0.58; Left this month = £4.42; Job cap never reached = £2.00
- *Included credit is always spent before top-ups.*
- Provenance: calculated

## What it won't do

- Managed AI uses OpenAI with global processing; see the privacy notice.
- Unused included credit does not roll over; top-ups do.
- A cancelled step may still incur a charge if it was already running.

## Covers

- Managed AI credit (Available): Each paid seat includes £5 of AI credit per UTC month; optional £10, £25 and £50 top-ups carry forward. Where: My account → Credits; rates in the app under Settings → Agent API
- What leaves your workstation (Available): Model files stay local; design context is sent only after you allow it in Agent settings, and you can preview it first. Where: In the app, Settings → Agent API; Instructions → Preview
- Job limits (Available): One tool call per step, at most eight steps and £2 per job; credit is reserved per step and unused credit returned. Where: Every managed AI job