v1.0.5 macOS release iOS in App Store review AGPL-3.0

Watch your AI agents. Before the bill.

OpenBurnBar is a local-first developer tool that watches the AI coding agents you already use — Claude Code, Codex, Cursor, Copilot, Factory, MiniMax, and more — and gives you a live read on tokens burned, dollars spent, and quota left, across all of them, on your machine. Every token leaves a trail. BurnBar reads the fire.

MONITORED INTEGRATIONS
Claude Code Claude Code
Codex Codex
Cursor Cursor
OpenAI OpenAI
GitHub Copilot Copilot
Factory Factory
MiniMax MiniMax
No telemetry by default No account required Reads logs, not your API keys
OpenBurnBar
Updated 4m ago 11,377 sessions
QUOTAS
  • Anthropic
    5h
    69%
    7d
    83%
  • Codex
    5h
    57%
    7d
    34%
  • Claude Code
+3 more in Settings
LaHormigaDormida Not enough signal

LaHormigaDormida · not en…

Provisional · In Os
MissionNeeds approval
Burn$2,662.51
Tokens15.12B
Questions0
Followups2
MissionPlanned
RuntimeLocal inference
Review mission approval →
Faithful re-render · live data captured from OpenBurnBar.app on 2026-05-12
menu-bar popover · live data
23
Providers with real usage data today
3
Honest confidence labels — exact · estimated · unavailable
0
Telemetry, analytics, or crash reports out of the box
Works offline. The whole product, locally.
7·2
Upstream providers routed across two same-format pools (OpenAI · Anthropic)
Three commitments

A developer tool, not a SaaS dashboard.

OpenBurnBar is built for the kind of developer who has three agents open in parallel and only checks the bill at the end of the month. Three things shape every decision in the product.

  1. 01

    Local-first by default.

    The canonical store is a local SQLite database, run by a local daemon under your user account. Firebase is an optional replication plane, not the source of truth. The whole product works offline.

    How the privacy model works
  2. 02

    Honest about confidence.

    Every provider row carries one of three labels — exact, estimated, unavailable. We don't pretend to know what the vendor doesn't tell us.

    See the full provider matrix
  3. 03

    Built for multi-agent reality.

    A single daemon owns routing, quota, retrieval, projects, and missions. Every surface — macOS, iOS, iPad, Cursor, VS Code, CLI, widgets, smart displays — speaks to the same daemon. One source of truth, many windows in.

    Tour the product
Inside the app

What a calm console looks like.

The macOS app is dense by design. Today's burn, top providers, quota windows that matter, and an on-device assistant that knows your data — one keystroke away.

OpenBurnBar iPad app — Burn tab showing Fleet Quota 42% Pressured, per-provider rings (Claude Code 18%, Codex 51%, OpenAI 56%), $18.74 today, $126.40 over 7 days, $613.92 over 30 days, $6,264.44 all-time
Real iPadOS capture · same SwiftUI design renders on macOS
Companions

The phone, the tablet, the launcher.

The Mac records. The phone and tablet render. Companion apps stream the same daemon data; widgets and Live Activity surface the burn even without opening the app.

iOS · In App Store review

iPhone

OpenBurnBar iPhone app — Burn tab showing Fleet Quota 42% Pressured, per-provider rings (Claude Code 18%, Codex 51%, OpenAI 56%), $18.74 today, $126.40 over 7 days

Burn tab · fleet quota 42% pressured

  • Per-provider quota rings + cards
  • Live Activity + Dynamic Island
  • Siri shortcut: "What's my burn today?"
iPadOS · In review (same build)

iPad

OpenBurnBar iPadOS app — Pulse tab with adaptive split-view, real burn telemetry

Pulse · adaptive split-view · ⌘1–4 · ⌘R · ⌘H

  • Chart Studio — 10 Swift Charts kinds + Mermaid
  • Trend Atlas insights rotate across 9 rules
  • Hermes Realtime Relay (paid tier)
Android · In development

Android

OpenBurnBar mobile Pulse tab — burn 1.53B tokens, 14,276 requests, end-of-day forecast $1,232.36

Pulse · streaming live from your Mac

  • Material 3 + Jetpack Compose, edge-to-edge
  • Read-only Firestore consumer; Mac streams the data
  • 97 Kotlin files, JDK 21, targetSdk 35
  • No Play Store distribution path yet
Beyond the menu bar · Smart displays

Run BurnBar where you can see it.

Your menu bar is one surface. Your Nest Hub on the kitchen counter and your Ulanzi pixel clock on the desk are two more — already shipping. The mockups below are faithful re-renders of the in-app preview code, hydrated from today's live router rundown.

Shipping (per-device QA matrix)

Google Nest Hub

Google · 7-inch smart display · Cast V2

BurnBar casts a live provider dashboard to the Nest Hub on your kitchen counter. Refresh, brightness, theme, and provider filter are all controlled from the Mac app.

  • Cast V2 + Home Assistant blueprints — no third-party server
  • Provider rail, big-total, ambient, photo-blend layouts ship today
  • Acceptance probe before "healthy" — `docs/SMART_DISPLAY_DEVICE_QA.md`
Shipping (per-device QA matrix)

ULANZI TC001

Pixel clock · 32×8 LED matrix · AWTRIX HTTP

A faithful 32×8 LED matrix render — same glyph tables, same palette, same per-pixel glow blur the BurnBar daemon paints to AWTRIX firmware.

  • AWTRIX HTTP — works on stock or community firmware
  • Four layouts · provider dashboard, quota carousel, burn status, alerts
  • Ember & whimsy palette by default · five palettes ship
Providers

The list of vendors it actually understands.

23 providers ship with real usage data today. 8 more are detected at install but don't expose data — so we show "Installed / Not installed" and stop there.

Selected · see Providers for the full matrix

Provider Where the data comes from Cost Quota Credential
Claude Code Anthropic's CLI agent — reads ~/.claude/projects/**/*.jsonl
Local JSONL + statusline bridge exact yes none (local)
Codex (OpenAI CLI) ChatGPT's coding CLI — reads ~/.codex/sessions/rollout-*.jsonl
Local rollout JSONL exact yes none (local) or ~/.codex/auth.json for hosted refresh
OpenAI Organization-wide usage from the admin API
/v1/organization/usage/completions exact partial Org admin key (sk-…)
GitHub Copilot Per-seat premium-interaction + chat limits
api.github.com/copilot_internal/user estimated yes GitHub OAuth or PAT (read:user)
Cursor Plan usage in USD straight from Cursor's web API
cursor.com/api/usage-summary exact yes Workos session token (auto-extracted)
Factory (Droid) Plan tier + rolling 5h/7d/30d windows, lane-aware
factory.ai org subscription + local session settings exact partial WorkOS browser session captured by FactoryLoginHelper
MiniMax Coding Plan remaining quota per model
minimax.io coding-plan endpoint exact yes Coding Plan key sk-cp-…
Warp Request credits, refresh windows, bonus grants
app.warp.dev GraphQL exact yes wk-… API key
Where it shows up

Eight surfaces, one daemon.

Every screen you care about — and a few you didn't expect — talks to the same local daemon. Multi-agent work needs multi-surface visibility.

Shipping macOS 14+

macOS menu bar app

The home base. Lives in the menu bar, reads local agent logs, surfaces cost, tokens, quota, sessions, and insights.

  • LSUIElement — no Dock icon, no windows stealing focus
  • Dashboard, providers, models, sessions, projects, streams, search
  • Hermes chat panel sits inside the dashboard
  • Daily digest, smart insights, quota refresh, daemon-backed control plane
  • Optional Firebase sync, optional iCloud session mirror, optional Sentry diagnostics
Download for macOS
In App Store review iOS 17+

iPhone & iPad companion

Your whole burn picture, in your pocket. The Mac does the work; the phone and tablet show it — and reach back to it.

  • See spend the moment it happens — charts drawn from your own numbers
  • Live on the lock screen and Dynamic Island, plus a “What's my burn today?” Siri shortcut
  • Floo — see, steer, and unlock your Mac right from your phone
  • Watch an agent work, or approve its moves, on the go
  • Ask your Mac's assistant from anywhere — privately (paid)
iOS launch status
Source-only beta Cursor / VS Code 1.95+

Cursor & VS Code extension

An activity-bar panel that hooks into the daemon over the local UNIX socket. Build locally and load unpacked — no marketplace listing yet.

  • Burn for the active workspace, scoped to the current Cursor agent run
  • Quota panel pulled from the same daemon the menu bar reads
  • Routed-provider gateway support — Z.ai, MiniMax, Ollama Cloud via Cloudflare tunnel
  • Open VSX / VS Marketplace publication is on the roadmap, not shipped
Editor setup
Shipping Embedded in macOS app

Local daemon

The control plane. A launchd-managed UNIX socket service every surface talks to.

  • Auth-token-gated JSON-RPC + HTTP gateway
  • Owns provider routing, quota refresh, retrieval, mission control
  • JSONL run journal — every agent invocation, every tool call, replayable
  • Repairs itself when launched from a stale plist or moved app bundle
Shipping macOS

Command-line interface

openburnbar — eight commands for scripting, agents, and operators. Same daemon, no extra account.

  • health · controller · questions · followups · missions
  • mission-approve · simulator-runs · simulator-replay
  • Pipes cleanly into git hooks, CI, and other agent scripts
Shipping (with iOS app) iOS / iPadOS

Widgets & Live Activity

Lock-screen quota, Dynamic Island countdown, home-screen cost sparkline. Siri shortcut: "What's my burn today?"

  • Hero small, cost sparkline medium, dashboard large
  • Live Activity on iOS 16.1+ with top provider + tokens + cost
  • App Intents for Spotlight + Siri
Shipping (per-device QA matrix) Nest Hub · Pixel Clock · Chromecast

Smart displays

One-click "Make display work" — casts a live OpenBurnBar dashboard, with proof of acceptance before marking healthy.

  • Google Nest Hub via Cast V2 + Home Assistant blueprints
  • ULANZI TC001 via AWTRIX HTTP or stock-firmware simulator
  • Per-device QA matrix gates support claims — see docs/SMART_DISPLAY_DEVICE_QA.md
See the live mockups
Feature-complete · Play Store pending Android 8+

Android companion

Everything the iPhone app does, now on Android — your burn, your sessions, the assistant, and Floo. Built and tested; not yet on the Play Store.

  • Full parity with the iPhone & iPad app, screen for screen
  • Floo media and the assistant, same as on iOS
  • Not yet distributed through the Play Store — that's the last step
BurnBar Router — same model, every time

Same model. Different account.

Failover swaps the account · the model id never moves

BurnBar's local gateway preserves the exact model you asked for. Your gpt-5.5 stays gpt-5.5; your claude-opus-4-7 stays itself. The router picks from accounts in your library that already carry that model — and if none survive, it returns a structured 503 instead of a silent substitute. Two modes ship: Provider-Family Failover (default, stay inside the selected provider family) and Exact Model Failover (opt-in, cross provider/account only after canonical model proof). Benchmarks are advisory; they never authorize a model swap.

  • Provider-Family Failover default · same model · multiple accounts
  • Exact Model Failover opt-in · same canonical model · fail closed
Locked-Model Relay — a model token labelled gpt-5.5 stays anchored at the top while five OpenAI-family accounts (OpenAI · personal, OpenAI · team, Codex · plus, OpenAI · agency, OpenAI · backup) cycle their availability below. The active routing pillar slides between healthy accounts every 2.4 seconds, demonstrating that failover swaps accounts without ever changing the model identity. Provider-Family Failover keeps the retry inside the selected provider family. Exact Model Failover can look outside that family only when the destination proves the same canonical model ID; every merely similar route is filtered out for a gpt-5.5 request.
  1. active route
  2. pressure
  3. rate-limited
  4. cooling · recovering
Hermes

A messenger for your own data.

Ask, in plain language, what your week looked like. What model burned the most cache. Which window is about to roll over. Hermes pulls context from your local SQLite — your sessions, providers, models, tools — and answers right there, in chat.

  • Conversation Atoms. Every entity Hermes mentions becomes a tappable chip — cost, session, provider, model, window — that opens the matching native view.
  • Chart Studio. On iPhone and iPad, Hermes streams JSON envelopes that render to 10 native Swift Charts kinds and sandboxed Mermaid diagrams.
  • Two backends. Local Index for stateless per-turn answers via the CLI, or the local Hermes Gateway for multi-turn, OpenAI-shaped chat with server-side tool calls.
  • Hermes Remote Relay. Reach the Mac's Hermes from anywhere over an end-to-end-encrypted WebSocket. Paid-tier; relay never sees plaintext.
Hermes Gateway · BurnBar Cloud

Connect Hermes through BurnBar Cloud

Choose BurnBar in Hermes the same way you'd choose Telegram or Signal. Approve a short code, and Hermes can receive your BurnBar messages and send replies back through your Cloud account.

Included with BurnBar Cloud & Cloud Pro Read the gateway docs
Hermes messaging platform
BurnBar Cloud your account

1

Choose BurnBar in Hermes

In Hermes' platform setup, pick BurnBar Cloud — right next to Telegram, Signal, and Slack.

2

Approve the code

Sign in to BurnBar and approve the short pairing code Hermes shows you. Takes seconds.

3

Send & receive from BurnBar

Hermes receives your BurnBar messages and sends replies back through your Cloud account.

Privacy & trust

By default, OpenBurnBar collects nothing.

All processing happens on your device. No telemetry, no analytics, no crash reports — nothing is transmitted unless you explicitly opt in. And when you do opt into cloud, your content leaves sealed — what stays readable is labeled operational metadata, domain by domain, in a ledger you can audit.

  • No API keys read for local-only tracking.
  • No account required for the core product.
  • Optional Firebase sync after sign-in — metadata only by default.
  • Optional iCloud mirror uses your Apple ID, separate from OpenBurnBar's servers.
  • Hosted credentials, when used, go to Google Cloud Secret Manager — not Firestore.
  • Claude Code is self-hosted only, by design.

Where the seal holds — the relay sees ciphertext only. The full map →

Trust boundary map: your devices and the untrusted relay Two islands marked Your Mac and Your Phone sit inside solid trust boundaries with pinned keys. Between them, a dashed cloud marked untrusted relay carries a sealed envelope that crosses without being opened. Where the lane meets the device boundary, a gate marked fails closed refuses anything that tries to downgrade. YOUR MAC YOUR PHONE UNTRUSTED RELAY FAILS CLOSED
Pricing

Free where it should be. Paid where it costs us money.

Free · Local

OpenBurnBar Local

$0 forever
no account

Local-first cost and quota tracking. No account. No cloud.

  • Local SQLite tracking, zero telemetry
  • Cost & token rollups — today / week / month / all-time
  • Quota windows for 11 providers
  • Daemon, CLI, Cursor / VS Code panel
  • iOS, iPadOS & Android companions
Included
  • Smart insights & configurable daily digest
  • The Wand: 1 local worker per Wand cast
  • Self-hosted quota runner — Claude Code & Codex
  • Hermes assistant on local backends
  • Optional Firebase sync — metadata only
Get OpenBurnBar
For power users

Cloud Pro

$24.99/mo $249/yr $20.75/mo · save 17% $249/yr available

Use your Mac from your phone and let agents work under your grant.

  • Everything in BurnBar Cloud
  • Floo — see & use your Mac from your phone
  • Agent Control — agents work under your grant
  • The Wand opens 8 agents in parallel
  • Live screen, files, calls & shared clipboard
  • Tamper-proof action record · per-task grants
Included each month
  • 500 hosted Agent Control actions
  • 50 relay-accounting GB
  • Prepaid overage only · BYOK never spends credits
Choose Cloud Pro
For your whole second brain

Ultra

$59.99/mo $599/yr $49.92/mo · save 17% $599/yr available

Everything in Cloud Pro, plus 10× your private agent memory — 100 sources, 500,000 sealed chunks, and 10 GB your agents can recall.

  • Everything in BurnBar Cloud Pro
  • 100 Pensieve knowledge sources
  • 500,000 encrypted memory chunks · 10 GB
  • Repo docs, notes & chat memories your agents recall
  • The Wand opens 16 agents in parallel
  • Sealed on-device · search runs over encrypted structures (some patterns stay visible)
  • Same hosted Agent Control & relay allowance as Pro
Private semantic memory
  • Cloaked vectors + sealed text — hosted recall is opt-in; structural patterns remain visible
  • 500 hosted actions · 50 relay-accounting GB each month
  • Prepaid overage only · BYOK never spends credits
Choose Ultra

Don't get burned. Keep building.

OpenBurnBar is AGPL-licensed, signed and notarized for macOS, and ready to run today.