Estimate OpenClaw multi-channel API cost without double-counting

Quick answer

Count the requests each channel actually causes, then count shared agent work once. A second chat channel does not automatically double your scheduled heartbeats or sub-agent runs. Clawback's free cost calculator is useful for a one-channel scenario, but its channels control multiplies every category, including background work. If your channels have different usage, do the per-channel worksheet first.

Start with a billable-usage worksheet

Choose one period from your provider's usage or billing export. Separate interactive requests by the channel that triggered them when you have a reliable channel identifier. Keep work that runs regardless of channel, such as a shared schedule, in its own row. If you cannot attribute a request, keep an unattributed row instead of assigning it by guesswork.

• Channel A: billed input, cached input if separately charged, and output tokens from requests triggered by A.
• Channel B: the same categories for requests triggered by B. A quieter channel may be much smaller.
• Shared: heartbeat, scheduled work, and child-agent requests that would run even with either chat channel disabled.
• Unattributed: requests that cannot be joined confidently to a channel or shared job.

For each row, multiply tokens in each billed category by its applicable price per token and sum the charges. Check your provider's current rate sheet, account discounts, and invoice rules rather than borrowing an old price. Compare all rows with the invoice for the same time window; investigate any gap before claiming savings.

A simple example of the counting rule

This is a hypothetical workload, not Clawback customer data. Suppose channel A triggers 120 requests in a month and B triggers 30. A shared schedule triggers 60 other requests, whether one channel or both are connected. The request inventory is 120 + 30 + 60 = 210, not 2 × (120 + 60). Requests alone are not a bill: collect actual input, cached-input, and output tokens for each row before estimating a price. A lengthy B request can cost more than several short A requests.

Use the calculator without mistaking its multiplier for measurement

1. Open the Clawback calculator and set channels to one. Choose the primary, heartbeat, and sub-agent models and enter one channel's approximate message rate. Note the displayed monthly scenario, not a measured bill.
2. Try the channels control at two to see the tool's equal-workload scenario. The implementation multiplies main messages, heartbeats, and sub-agents by two. This is only appropriate if all three categories really occur independently for both channels.
3. Return to one channel for the shared-work scenario. Price channel-specific and shared rows from your provider export outside the tool, using actual billed token categories and the same period. Do not add two full calculator totals if both contain the same heartbeat or child job.

We inspected the calculator code for this walkthrough. It projects 30 identical days, adds 4,000 context tokens to each entered message, estimates output at 40% of entered message tokens, and assumes 6,000 input plus 200 output tokens per heartbeat and 8,000 input plus 4,000 output per sub-agent. These are tool assumptions, not observations about your agent. The tool does not import your invoice, model cache usage, or per-channel request records. Use your actual usage export to approve a budget.

What to do next

If channel B barely changes your measured total, keep shared workload out of its incremental cost. If B creates additional schedules or child runs, move those measured requests into its row. Recheck after adding a channel or changing a model. The calculator helps compare assumptions; ongoing spend tracking is not available through the free calculator. You can request early access to the planned spend dashboard by email. That page is not a self-service signup or a paid checkout.