AI Credits and Algo Agent Depth

Basic or Professional

AI chat and Fix with AI use AI credits. Editing, saving, and reviewing a staged diff do not send a model request by themselves.

Use automatic routing

Algo Agent chooses the underlying capability for the current action. Routine inspection, extraction, summarization, formatting, and deterministic checks use an efficient route. Research design and interpretation use a reasoning route. Frontier review is reserved for explicit conflict, repeated failures, difficult debugging, risk audits, low-confidence high-impact conclusions, and final review.

Tool execution alone does not trigger frontier review, and account settings do not override routing. This keeps model policy consistent and auditable as provider catalogs change.

The composer offers two product-level experiences:

Experience	Use it for
Algo Agent	Most explanations, edits, and research turns; routing and budgets adapt automatically
Deep Analysis	A difficult or high-impact decision that warrants broader expert review

What affects credit use

Pine editing requests are metered from the total tokens they use. Algo Agent research uses a newer provider-cost-aware meter: cached input costs fewer credits than fresh input, output costs more than input, and model prices are reflected in the charge. Requests whose input crosses a provider long-context pricing threshold also carry that provider’s multiplier. Larger prompts, longer conversation history, attached scripts or reports, longer responses, tool output, and Deep Analysis can therefore increase credit use. Do not rely on a fixed per-request estimate from a documentation page.

Algo Agent conversations compact their server-side context when the rendered window grows large. Compaction reduces repeated context while preserving the normal conversation continuation flow. It can reduce future credit use, but the effect depends on the task and how much context the agent still needs.

To keep a request bounded:

attach only the script tab, report, image, or error needed for the task;
remove stale context before changing topics;
ask for one coherent change rather than several unrelated changes;
start a separate chat when the previous history no longer matters; and
inspect the resulting diff before asking for another full rewrite.

Check balance and usage

Open Account → Billing. The page separates available credit buckets when present and shows usage for the previous 30 days.

Current account behavior:

monthly credits reset each billing period;
purchased credits are tracked separately and roll over;
new accounts receive 50 starter credits;
Basic includes 400 monthly credits and Professional includes 1,000;
top-ups are 250 credits for $10, 500 for $20, or 1,000 for $40;
paid plans can purchase top-ups when checkout is enabled; and
beta accounts request additional credits from the admin team.

The editor also surfaces an action when a request cannot run because the balance is insufficient.

Choose depth without overspending

Start with Algo Agent, save the current script, and record the prompt and attachments.
Ask for a bounded result—for example, explain one validation error or propose one diff.
Review factual accuracy, Pine Script validity, diff size, and credit usage.
Use Deep Analysis only if the standard result leaves a consequential question unresolved or evidence remains in conflict.
Review the new evidence and Billing usage. More depth should buy a clearer decision, not merely a longer answer.

When a request is blocked

If the editor reports insufficient credits:

confirm the balance in Account → Billing;
reduce unnecessary attachments or old conversation context;
switch from Deep Analysis to the standard Algo Agent experience when the extra research budget is unnecessary;
purchase a top-up or request additional beta credits when that control is available; and
retry the request. Your code should be reviewed from the editor state, not reconstructed from a partial model response.