A web app that enforces hard dollar and token budgets per AI agent and kills runaway sessions before they blow the monthly spend

AI agents loop silently for hours and rack up thousands of dollars in token spend with no hard enforcement layer to stop them. Existing soft limits inside SDK params are advisory and bypassable; teams only discover the overage on the billing page. This app sits as a governance proxy between any agent runtime and the LLM API, evaluates per-agent dollar and token ceilings before each call completes, and terminates the session the instant a threshold is crossed.

Demand Breakdown

809

GitHub

168

OPENAI_FORUM

Social Proof 6 sources

Uber torches 2026 AI budget on Claude Code in four months

2026-06-14

402 HN

Are the costs of AI agents also rising exponentially? (2025)

@n/a · 2026-04-15

306 GH

dipampaul17/AgentGuard: Auto-kill AI agents before they burn through your budget

@dipampaul17 · 2025-08-20

168 HN

Show HN: AgentGuard - Auto-kill AI agents before they burn through your budget

@dipampaul17 · 2025-08-20

73 OA

AI Agents Burn 50x More Tokens Than Chats

@LeanOps · 2026-03-01

42 HN

Uber blows through its AI budget in 1 quarter

@n/a · 2026-05-26

Gap Assessment

CompetitiveMultiple tools exist but differentiation opportunities remain

4 tools exist (AgentGuard, AgentBudget, MintMCP, Anthropic token budgets (API param)) but gaps remain: No web UI, no multi-team budget policies, no per-agent audit logs, no pre-kill alert webhooks, no MCP server integration, no dashboard, single maintainer with no commercial support.; Early-stage, no team/org-level quota management, no observability layer, no API gateway mode, no commercial pricing tier..

Features7 agent-ready prompts

Per-agent hard budget policy engine

▶

Real-time spend ledger with per-call accounting

▶

Pre-kill webhook alerts and Slack/email notifications

▶

Multi-provider proxy with SDK-drop-in compatibility

▶

Team and project budget hierarchy

▶

Agent spend dashboard and history

▶

Automatic session replay and audit log

▶

Competitive LandscapeFREE

Product	Does	Missing
AgentGuard	OSS Python library that wraps an agent runtime and kills the process when a configured dollar or token threshold is crossed.	No web UI, no multi-team budget policies, no per-agent audit logs, no pre-kill alert webhooks, no MCP server integration, no dashboard, single maintainer with no commercial support.
AgentBudget	Real-time dollar budget enforcement for AI agents with per-agent hard caps via a lightweight SDK wrapper.	Early-stage, no team/org-level quota management, no observability layer, no API gateway mode, no commercial pricing tier.
MintMCP	MCP gateway that adds rate limiting and per-agent token quotas at the protocol level, blocking calls once the quota is hit.	Rate limiting only — no dollar-value budgets, no pre-kill webhooks, no spend forecasting, no team-level rollup dashboards, limited to MCP-compatible runtimes.
Anthropic token budgets (API param)	Per-session token budget parameter in the Anthropic API that signals the agent to stop when approaching the limit.	Advisory/soft only — the model can exceed it; enforced at model level not infrastructure level; no cross-provider support, no dollar-value enforcement, no kill mechanism at the network layer.

Notable VoicesFREE

@dipampaul17168 likes · 2025-08-20

"AgentGuard auto-kills AI agents before they burn through your budget — built this after watching a friend's side project rack up $800 in one night from a stuck loop."