Skip to content

Self-Budget

IDself_budget
CategorySystem
Features(none)
Tools(none)
Included inGeneric harness (default)
DependenciesNone

Teaches the agent how to self-manage an indicative budget that the user mentions in conversation — for example, “you have $7” or “keep this under 20k tokens”. The capability contributes prompt text only; it adds no tools and performs no enforcement.

For platform-enforced budgets (authoritative limits that pause or stop sessions automatically), use the separate budgeting capability.

self_budget is prompt-only. When the capability is enabled the agent’s system prompt gets a “Self-Managed Budget” section that explains:

  • The self-budget is an agent-managed soft target, not a hard limit.
  • Session usage metadata (exposed via get_session_info) is the source of truth for current spend.
  • The agent decides when to start tracking, when to re-check, and when to stop.
  • As the target tightens, the agent should adapt — shorter outputs, fewer retries, narrower exploration, fewer redundant tool calls.
  • The agent avoids claiming exact cost certainty when only token counts or partial pricing are available.
  • The agent distinguishes between platform-enforced budgets and user-requested indicative budgets when reporting progress.

There is no self_budget tool. Usage data comes from get_session_info, which is provided by the session capability (bundled by default in the Generic harness).

Aspectself_budgetbudgeting
What it isAgent-managed soft targetPlatform-enforced limit
ToolsNonecheck_budget
EnforcementNone (prompt guidance only)Session is paused/stopped automatically
Data sourceget_session_info cumulative usageBudgets table / ledger
Use caseUser says “you have $7” in chatOrg/session has a configured budget in the platform

The two capabilities are non-conflicting and can run together. The Generic harness includes both.

  • User-set budget in prompt — “Spend at most $3 on this investigation.” The agent tracks usage via get_session_info, adapts scope, and reports progress.
  • Coaching the agent to be frugal — Even without an exact dollar target, the prompt guidance nudges the agent toward shorter outputs and fewer redundant tool calls.
  • Budget-aware estimation — The agent qualifies cost estimates (“roughly”, “on the order of”) rather than over-stating certainty.
User: You have about $2 of compute for this task. Analyze the three largest
dependencies in package.json and tell me which one to remove.
Agent: [calls get_session_info — current usage $0.14]
Agent: Got it. I'll focus on the three largest dependencies and keep this tight.
[... scoped analysis ...]
Agent: [calls get_session_info — current usage $1.62]
Agent: Roughly $1.60 spent so far. Based on size and use count, `moment` is the
best removal candidate — replacement path is straightforward. I'll stop
here to stay under your $2 target.
  • Budgeting — platform-enforced budgets with the check_budget tool
  • Session — provides get_session_info, the usage data source
  • Budgets — full budgeting system documentation