Use case

# One big model bill, and no idea where it went. *See each request.*

The console shows what each agent request cost and saved, by session, project and agent. Cost is not split by person yet; Insight shows who is stuck, by person.

[Get started free](https://console.context-mode.com/login) [Read the docs](https://context-mode.com/docs/how-saving-works#where-you-see-it)

You ask

Where did it go? What did we save? This turn Codex By person?

The bill gives you one total.

What you see now

- One total on the bill
- No answer to: which request?

The Context Mode console

- Tokens, cost and saving per request
- By session, project and agent
- Over time, in Analytics

Analytics is in every plan, Free included.

Saving against what? Against going direct.

What you paid

- The usage the provider returned
- At the provider's price

What going direct would cost

- What Claude Code would have sent, going direct
- Counted with the provider's own token counter

The saving is the difference. A request gets a dollar figure only when its count is within 5% of the provider's. On Codex the saving is a lower bound.

What did my last turn save?

```
⏺ context-mode · context 67 KB lighter than your raw transcript · kept 71 KB out of what the model read
one line in the terminal, at the end of a turn
```

Live in the console shows running sessions and the saving per minute.

We use Codex on a ChatGPT login.

Cost

- OpenAI's token counts for each response
- At OpenAI's list price
- Shown as equivalent API cost

Saving

- Counted when both sides can be counted
- A lower bound

No paired Codex A/B yet, so we quote no Codex saving percentage.

Can I see cost per engineer?

The console

- Cost by session, project and agent
- Not by person yet

Insight, by person

- Blockers, risks and changes
- Retry waste and error spikes

Insight shows who is stuck, not what they spent. It reads 222 patterns over Engine events.

The problem

## You pay the bill. *You cannot explain it.*

**One number**

The bill shows a total. It does not show which session, project or agent spent it.

**No baseline**

You cannot say if a tool or a change saved money, because you do not know what the same work cost before.

**No early sign**

An agent stuck in a retry loop spends all day. You find out when the bill comes.

How it works

## One place on the request path. *Each request priced twice.*

1. 01 **Connect your agents** `npx @context-mode/cli` points each agent at the gateway. Your own Anthropic or OpenAI account still pays the model.
2. 02 **Every request is recorded** The cost log (the ledger) prices each request as sent, at the provider's price, and again for what going direct would have sent. The saving is the difference.
3. 03 **Read it where you work** A line in the terminal that starts with `context-mode ·`. Live and Analytics in the console, split by session, project and agent.

**Your agents**Claude Code and Codex, on every machine.

→

**Context Mode · the cost log**Records each request on its way to the model.

- Tokens and cost, from the provider's usage
- Saving against going direct
- Split by session, project and agent

→

**The model**Called with your own credential. Your provider bills you as before.

Every request passes the gateway once. The cost log, in blue, prices it there, so the console can show what each request cost and saved.

Proof

## The cost log counts what you pay. *We checked it against paired runs.*

**Within 5%** A request gets a dollar figure only when its count is within 5% of the provider's prompt count

**Every plan** Live and Analytics are in every plan, Free included

**65.5% lower cost** on a long agentic coding session, Claude Opus 5.5, $7.20 to $2.49, 4 of 4 pairs cheaper

### Questions about the bill

| Your question | Where you see it | What it shows |
| --- | --- | --- |
| Which project costs the most? | Analytics | Tokens, cost and saving over time, by project |
| Which agent costs more? | Analytics | The same figures by agent, so Claude Code and Codex sit side by side |
| Did a change save money? | Analytics | Cost and saving by session: put a session before the change next to one after it |
| What does Codex cost on a ChatGPT login? | The console | OpenAI's token counts at OpenAI's list price, shown as equivalent API cost |
| What does each engineer cost? | Not split by person yet | With one Pro plan per engineer, each person has their own bill and budget |
| Will our Context Mode bill run over? | E-mail | On Starter and Pro, with a budget set, an e-mail at 75% and 100% of it. The budget covers what you pay us, not the model bill |

Our tests, not your traffic. The cost log matched paired runs on 2026-09-26 (Claude Haiku 4.5). On Codex the saving is a lower bound. [Method and sources →](https://context-mode.com/docs/benchmarks)

Products used

## What this *uses.*

**[Live, Analytics and the cost log](https://context-mode.com/docs/how-saving-works#where-you-see-it)**

How each request is priced, and where you see the saving.

**[Insight](https://context-mode.com/insight)**

How your team works with agents, from free Engine events: blockers, retry waste and error spikes, answered by role.

**[Team analytics, compared](https://context-mode.com/docs/landscape#observability)**

What the console records, what it does not, and when another tool fits better.

**[Plans](https://context-mode.com/pricing)**

Live and Analytics are in every plan, Free included. On Starter and Pro, set a budget and get an e-mail at 75% and 100% of it.

Related: [For engineering leaders](https://context-mode.com/solutions/engineering-leaders) · [Agent visibility and audit](https://context-mode.com/solutions/agent-visibility) · [Cut agent spend](https://context-mode.com/solutions/cut-agent-spend)

## Stop guessing at the bill. *See each request.*

*npx @context-mode/cli* connects your agents. 1,000 requests free, no card.

[Get started free →](https://console.context-mode.com/login) [Read the docs →](https://context-mode.com/docs/how-saving-works#where-you-see-it)