Multi-agent work needs multi-agent visibility.
You don't run one model anymore. You run Claude in one tab, Codex in another, Cursor and Copilot in the editor, and Factory or Forge on the side. Spend is fractured across vendors and time zones, quotas don't share a clock, and "why did the bill explode" arrives on the first of the month.
- B1
Avoid bill shock.
See today's burn the moment you sit down. See this week's pattern. See where it's going before the invoice rounds the corner.
- B2
Understand agent spend.
Cost broken down by provider, by model, by session, by project. Which model burned the most cache? Which agent doubled this week? OpenBurnBar answers — locally.
- B3
Know quota before workflows fail.
Five-hour windows. Weekly windows. Plan-tier ceilings. Lane-aware counters. OpenBurnBar reads them all and tells you which one is about to roll before your agent hits a wall.
- B4
Recover context fast.
Sessions, projects, streams, tool calls, retrieval — all indexed locally. When something goes sideways, you can replay the run, inspect the tool trace, and pick up where the agent left off.
- B5
Keep core data local.
Your sessions never have to leave your machine. The cloud is an opt-in convenience, not a requirement. Disconnect the network and the product keeps working.
- B6
Use the cloud only when it helps.
When you want hosted quota refresh from a phone or a backed-up conversation across devices, the cloud is there — gated by Apple-verified entitlement and end-to-end encryption.
- B7
Operate agents responsibly.
Whether you're a solo developer or a small team, OpenBurnBar makes the cost of every agent legible. The dashboard is the operating layer that lets you keep spending sane.
- B8
Trust by inspection.
MIT-licensed. The threat model is in the repo. Provider claims cite the file and line of the adapter. You can read every promise in code, not just in copy.
The bill is going to find you anyway.
Read it on your own terms.