Practice: CostGuard and where your data goes

Self-check

Six quick questions. Answer each in your head before opening the collapsible.

1. What is the CostGuard mental model in one sentence?

Show answer

CostGuard is a software safety net you configure once that watches your BYOK spending against a monthly cap and either warns or blocks at the threshold you pick, so a runaway agent loop or an experimental prompt cannot turn into a surprise on your card.

2. Which of the three usage types counts toward your CostGuard cap, and which two do not?

Show answer

Counts: pay-as-you-go API keys (Anthropic, OpenAI, Google AI, Groq, and most other providers; the dollar cost of every message adds up). Does not count: OAuth providers (Codex sessions show $0 because your ChatGPT subscription handles billing) and local models (no provider in the loop, no dollar cost to track).

3. You hit 100% of your cap and the next send is rejected. What are the two unblock paths?

Show answer

Raise the cap in Settings, Budget (which unblocks immediately on save), or wait for the period to reset (the default is a rolling 30-day window, so the oldest day’s spend ages out today; alternatively switch to calendar-month resets in Settings). The dock row indicator goes red at 100%; the in-chat banner reads “Monthly budget reached. Bump your cap in Settings to keep going.”

4. Trace the data path of a single message from your keyboard to the agent’s reply. How many servers does the conversation content touch?

Show answer

The path is: your keyboard, the Clawless app on your computer, the bundled OpenClaw engine (still on your computer), the AI provider over the internet, OpenClaw, Clawless, your screen. The conversation content touches one server: the AI provider’s. There is no Clawless server in the data path. The only server we operate is the license server, which receives license state, never message content.

5. What survives, what stops working, and what stays available offline if Clawless went out of business tomorrow?

Show answer

Survives: your conversations, memories, settings, agent definitions, and API keys, all on your computer disk. They are not in any Clawless cloud, so they do not disappear when the company does. Stops working: new commits to the app (no updates), and the license check after the grace period elapses. Stays available offline: local models. You can launch and chat with no internet and no Clawless services, which doubles as the answer for “what if I am on a plane” and “what if I need a setup that depends on no one else’s servers.”

6. You see a Codex session at $0 on the Usage dashboard. Bug or feature?

Show answer

Feature. Codex (the OAuth path from lesson 3) bills through your ChatGPT subscription rather than per-token on an API key, so the per-message cost tracked in Clawless really is zero. It also does not count toward your CostGuard cap. If most of your work is on Codex, the Usage dashboard will be quieter than you might expect.

Try it yourself: the 20-minute setup

The body lesson covered the dials and the data path. This is where you turn the dials yourself and look at what is on your dashboard. Twenty minutes if you take your time.

Side effects: no new charges (you are configuring caps, not spending). The settings take effect on the next message you send.

Steps:

Open Settings, Budget. Click the gear icon at the bottom of the navigation rail on the far left of the Clawless window. In the sticky sidebar on the left of the Settings page, click the Budget section.
Pick a monthly cap that feels comfortable. Start with something low if you are nervous (the $5 or $10 options are common first picks). You can raise the cap at any time, and raising it unblocks immediately if you have already hit it. If you really do want no cap, the $0 (unlimited) option does what it says; CostGuard tracks the numbers but does not block.
Choose your warn threshold. Default is 80%. The 50%, 90%, and off options are all reasonable depending on temperament. The 80% default leaves enough headroom to react without nagging you too often.
Choose your hard-stop behavior. Default is Block (the next send is rejected when the cap is hit). The alternative is Warn-and-allow (every send over the cap shows a warning but the message goes through). Block is the safe pick unless you have a specific reason to allow overage.
Save. CostGuard is now active. Pre-release-build note: the cap-enforcement layer goes live at production launch; pre-release builds show accurate usage numbers but do not yet enforce the block. If you are on a pre-release build, treat the cap as personal discipline until enforcement ships at launch.
Open the Usage dashboard. Click the Usage icon on the navigation rail. Look at the breakdown by provider, model, and conversation for the past week. If there is one session that ate disproportionate budget, that is the most important number on the screen; investigate it.
Find the privacy policy. In Settings or the About panel (depending on your build), find the link to the Clawless privacy policy. Skim it once now while the topic is fresh from the lesson body. The questions worth answering for yourself: who can read my conversations, what does the license server see, and what happens to my data if I uninstall.
Optional: try the calendar-month vs rolling-30-day toggle. In Settings, Budget, switch the period mode if you would rather think in calendar months than in a rolling window. Both are reasonable; pick whichever matches how you already mentally track recurring costs.

Expected outcome: you finish with an actively enforced (or about-to-be-enforced, on pre-release builds) monthly cap that matches your tolerance for surprise, a real sense of where your money went the past week, and a current understanding of where your data lives and who can see it.

If something went sideways: the five first-day surprises at the end of the lesson body cover the most common confusions, especially the OAuth-shows-$0 and the data-path ones.

Flashcards

Ten cards. Review once a day for a week, then on the intervals your spaced-repetition tool suggests.

Q. What is the CostGuard mental model in one sentence?

A software safety net that tracks your BYOK spending against a monthly cap and either warns or blocks at the threshold you pick, so a runaway agent loop or an experimental prompt cannot become a surprise on your card.

Q. What three settings do you pick in Settings, Budget?

Monthly cap (dropdown, defaults include $5, $10, $25, $50, $100, $0 for unlimited, or custom). Warn threshold (50%, 80% default, 90%, or off). Hard-stop behavior (Block new messages by default; alternative is Warn-and-allow which passes messages but shows a warning).

Q. Which usage types count toward the CostGuard cap, and which do not?

Pay-as-you-go API keys count (Anthropic, OpenAI, Google AI, Groq, most others). OAuth (Codex sessions through your ChatGPT subscription) does not count (shows $0 because your subscription handles billing). Local models do not count (no provider, no dollar cost).

Q. Describe the two visual cues that signal where you are against the cap.

The dock row indicator turns yellow at the warn threshold (80% by default), with a notification shown once. It turns red at 100%; with Block on (the default), the next send is rejected and an in-chat banner appears (“Monthly budget reached. Bump your cap in Settings to keep going.”).

Q. What are the two unblock paths once you hit the hard cap?

Raise the cap in Settings, Budget (unblocks immediately on save) or wait for the period to reset. The default reset is a rolling 30-day window so the oldest day’s spend ages out today; calendar-month resets are an alternative in Settings.

Q. Trace the data path of one message from your keyboard to the agent's reply.

Your keyboard, the Clawless app on your computer, the bundled OpenClaw engine (still your computer), the AI provider over the internet, OpenClaw, Clawless, your screen. The conversation content touches one server: the AI provider’s. No Clawless server in the data path.

Q. What is the one thing our server holds, and what it does not hold?

The Clawless license server holds license state (trial, active, expired). It never holds conversation content, message bodies, or memories. They never leave your computer, so they never reach us.

Q. What survives if Clawless went away tomorrow?

Your conversations, memories, settings, agent definitions, and API keys all live on your computer disk, not in any Clawless cloud. They stay where they are. The app keeps working as long as your local install runs and the AI providers you use stay online. The license check has a grace period built in.

Q. What is local-models mode and when is it relevant?

A mode where Clawless runs models entirely on your own computer with no cloud provider in the loop. You can launch and chat completely offline. Doubles as the answer for “what if I am on a plane” and “what if I need a setup that depends on no one else’s servers.”

Q. What four parties do you trust when you use Clawless?

Us (to ship a desktop app that does what it claims and protects local data to the standards your operating system supports), the AI provider (with the contents of your messages and any memories that travel with them), OpenClaw (the open-source intermediary engine; auditable code), and any tool, skill, or integration you install (with the permissions it asks for; same model as browser extensions).