Budget

What this page is for

The Budget page is your control surface for how much Exolvra is spending on LLM calls. Every token that goes through a provider is tracked, priced, and totaled here. Use this page to:

• Monitor total spend across time windows
• Break down cost by agent, project, model, or session
• Set hard caps that stop execution when limits are reached
• Watch rate-limit headroom across providers

If you’re running Exolvra self-hosted against your own provider API keys, this is where you avoid a surprise bill.

Layout

The page is split into sections from top to bottom:

1. Spend summary — today, week-to-date, month-to-date, and a sparkline graph of daily cost over the past 30 days
2. Budget caps — current daily, monthly, and per-session limits, each with a usage bar
3. Breakdown — a multi-axis table grouping spend by agent / project / model / session; click column headers to re-sort
4. Rate limits — live status of each configured LLM provider: recent token throughput, current limit, and headroom

Date range controls in the header let you scope all of the above to a custom window.

Common tasks

Set a daily budget cap

1. Click Settings in the budget page header (or go to Config → Budget & Security)
2. Enter a daily limit in US dollars
3. Save

When the day’s spend crosses the limit, Exolvra stops sending new LLM calls. In-flight calls complete; new ones are refused. The cap resets at midnight UTC (or your configured timezone).

Recommended starting values: $5/day for personal use, $50/day for small teams. Tune up as you learn your burn rate.

Set a monthly budget cap

Same page, separate field. The monthly cap is a rolling 30-day window, not a calendar month. When the cap is hit, Exolvra stops until the oldest day in the window ages out.

Enable auto-downgrade

When budget pressure rises, Exolvra can automatically route requests to cheaper models instead of refusing the call entirely. Enable Auto-downgrade in the budget settings and set the thresholds. With auto-downgrade on:

• At 70% of daily budget: requests to opus-class models route to sonnet-class models
• At 85% of daily budget: everything routes to haiku-class models
• At 95% of daily budget: hard stop

You can tune the thresholds. Auto-downgrade is off by default.

Configure session warnings

A single conversation can run up surprising cost if an agent gets stuck in a tool-call loop. Enable Session cost warnings and set a threshold (default $2). When a single session crosses the threshold, a banner appears in the dashboard and a warning is logged. The session continues — the warning is informational, not a stop signal.

Browse spend by agent

Scroll to the breakdown section, switch the grouping to Agent. You see one row per agent with total spend, token count, average cost per session, and average cost per issue. Sort by total spend to find the expensive ones.

Agents that are unexpectedly expensive are usually making too many tool calls per turn, or looping on a problem they can’t solve. Open their detail page from the link in the breakdown and investigate their recent sessions.

Browse spend by project

Switch the breakdown grouping to Project. Same columns, but grouped by the project the work belongs to. Useful when you’re trying to estimate what a specific initiative is costing.

Watch rate-limit headroom

Each configured provider shows a live rate-limit row: current tokens per minute, the provider’s declared limit, and the percentage remaining. When usage approaches the limit, Exolvra automatically throttles and may briefly cool down the provider before sending more requests.

If you see a provider consistently near 100%, you’re bumping against their rate limit — consider:

• Moving some agents to a different provider
• Spreading calls across multiple API keys (if you have them)
• Upgrading your tier with the provider

Per-session cost tracking

Every conversation and every issue execution has a cost-to-date visible inline in the dashboard (look at the header of any session or issue detail page). The total on the budget page is the sum of all those individual costs.

If a specific session surprises you with high cost, open it — the Traces view shows every tool call with its token usage, and you can usually spot the problem (an agent searching the web 20 times, or retrying a failed call in a loop).

Common pitfalls

No caps at all. The biggest mistake is running Exolvra against a paid provider without any budget cap. A single buggy agent in a tool loop can burn a month’s budget in an hour. Always set a daily cap, even if it’s generous.

Cap set too low. If you set the cap too tight, Exolvra stops mid-task and you’re left with half-done issues. If you’re routinely hitting the cap, raise it or enable auto-downgrade.

Ignoring per-session warnings. Session cost warnings exist for a reason. If you see them firing regularly on specific agents, that agent’s personality or tool set probably needs tuning — it’s doing more work than it should per turn.

Where to go next

• Agents — investigate expensive individual agents
• Admin overview — the full security + budget setup