AI Cost Gate sits between your agents and your LLM providers, tracking every request by project, agent, and model. Budgets that auto-stop runaway spend. SQLite or Postgres — pick your storage. Your prompts never leave your machine.
Tracks every provider — OpenAI, Anthropic, Google, OpenRouter, and more. No vendor lock-in.
Your agents will think before they spend.
Watch spend climb against a $30 budget. The gate blocks the next request before it blows past the limit.
Every cent attributed by agent and provider — all in a single local .db file.
SQLite database: acg.db (single file, ~240 KB) tables: requests, agents, projects, budgets, alerts rows: 5 (demo)
Only metadata is logged. Prompts never leave your machine.
⚠️ These examples are for feature illustration only — the real product's pages may look and behave differently.
Every LLM call tracked with project, agent, and model tags. Know exactly where every cent goes.
Slice cost data by any dimension. See which project spends most, which agent is the heavy hitter, which model costs the most per request.
Set per-project or per-agent budgets. When spend hits the threshold, the gate blocks the next request automatically.
Get notified before budgets blow. Configurable thresholds let you intervene early, not after the invoice arrives.
Works with OpenAI, Anthropic, Google, OpenRouter, and any OpenAI-compatible API. No vendor lock-in.
Pick SQLite for zero-setup single-host, or Postgres for concurrent multi-agent and team deployments. Same code, same migrations — switch via one env var.
Only metadata is logged — tokens, cost, model, timestamp. Your prompts and responses never touch the gate's storage.
Agents query costs and budgets via a simple REST endpoint. Ask 'this costs $3, continue?' before every expensive call.
Real workflows where local-first cost control pays off.
Claude Code session burnt $200 in an afternoon.
Per-project budget cap auto-stops at the threshold you set.
Need cost tracking inside our existing platform.
`docker compose up`, attribution flowing in 10 minutes.
Sponsor bots and CI agents all hit paid APIs in parallel.
Per-agent budget caps keep each bot under its own ceiling.
Grant-funded API credits disappearing faster than papers ship.
Per-grant budget + alerts before the credit pool runs dry.
Claude Code session burnt $200 in an afternoon.
Per-project budget cap auto-stops at the threshold you set.
Need cost tracking inside our existing platform.
`docker compose up`, attribution flowing in 10 minutes.
Sponsor bots and CI agents all hit paid APIs in parallel.
Per-agent budget caps keep each bot under its own ceiling.
Grant-funded API credits disappearing faster than papers ship.
Per-grant budget + alerts before the credit pool runs dry.
Deploy AI Cost Gate in 10 minutes. One SQLite file. Full control.
