v0.1.0 — initial release

Python port of @mukundakatta/agentbudget. Same API surface, snake_case names. Token + dollar budget caps for AI agents — raises BudgetExceededError when an LLM call would push past the ceiling.

Install

pip install agentbudget-py

What's in v0.1.0

• Budget(max_input_tokens=, max_output_tokens=, max_total_tokens=, max_cost_usd=, pricing=, allow_unknown_pricing=)
• record_usage({model, input_tokens, output_tokens}) — tally + raise on over
• wrap(fn, extract_usage=) — auto-record from Anthropic / OpenAI shapes
• would_exceed / assert_can_spend — pre-flight checks (no mutation)
• remaining() / reset() — introspection
• DEFAULT_PRICING starter table (Claude + GPT, early-2026 rates)
• allow_unknown_pricing flag for preview models with no rate sheet yet
• BudgetExceededError carries cap, limit, attempted, overshoot, model
• 23/23 tests on Python 3.10+, zero runtime deps

Sibling libraries

Part of the @MukundaKatta agent reliability stack:

• agentsnap-py, agentguard-firewall, agentcast-py, agentfit-py, agentvet-py — and agentbudget-py (this lib)

JS sibling: @mukundakatta/agentbudget.