See what AI actually costs

The honest
AI chat. Brewed,
not metered in secret.

One OpenRouter key unlocks 400+ models. Every reply is stamped with the exact tokens and cents it cost. Your chats and keys live on your phone — there is no account, no server, and nothing to cancel.

$0 to start — bring your own key 400+ models 0% markup, ever
100% local keys & chats never leave the phone
The meter is the point

Most apps hide the spend.
This one stamps it on every reply.

Tokens and dollars from the provider's own usage block — not a guess. Spend lives in an append-only ledger that even survives deleting your chats.

Typical chat app

Flat fee, hidden meter

  • $20/month whether you use it or not
  • Never know what a message really cost
  • One provider's models, their rate limits
  • Chats stored on someone else's servers
  • Stop button is decorative
HonesTea

Pay the provider, not the middleman

  • Real usage stamped on every reply
  • Auto picks the cheapest capable model
  • 400+ models behind one key
  • Chats + keys stay on your phone
  • Stop actually aborts — partial turns cost what streamed
Features

A harness, not a wrapper

Auto routing

Cheapest model that can handle each message — vision for images, tools for web search, enough context for the whole conversation. Graceful fallbacks included.

📈

Honest cost ledger

Per-message USD from real usage, per-model lifetime spend, regeneration roll-up, and a savings card vs. flagship pricing.

Stop that works

Abort any stream. The partial reply is kept, finalized, and billed only for what streamed.

📚

Context compaction

Long chats get summarized at 80% of the model's window — visible as a divider, never silent truncation.

🌐

Web tools

Search, page fetch, and date-time tools with a live activity panel and citation chips on the replies that used them.

🔄

Regenerate history

Flip between versions of any reply with < n/total > pagination. Superseded turns still count — they were real spend.

📤

Files & voice

Images and PDFs with per-model capability gating, voice dictation, and read-aloud on every reply.

🔒

Local by default

SQLite on device, keys in the system keystore, FTS search across chats, offline history. No backend exists to leak.

Curated lineup

Seven good defaults.
400+ more in the browser.

Live prices below; the app fetches the whole catalog and its real pricing on launch.

FlagshipClaude Fable 5.1Anthropic's frontier — best-in-class writing, 1M context.$10.00 in / $50.00 out per M tokens
FlagshipGPT-6 AstraOpenAI's flagship multimodal — top-tier reasoning, 1M+ context.$10.00 in / $50.00 out per M tokens
FlagshipKimi K3Open-weights flagship — frontier-class agentic work.$3.00 in / $15.00 out per M tokens
WorkhorseGLM 5.3The daily driver — 1.3M context, strong reasoning. App default.$1.40 in / $4.40 out per M tokens
WorkhorseDeepSeek V4 ProHard problems per dollar — reasoning and code.$0.90 in / $1.80 out per M tokens
BasicDeepSeek V4 FlashNear-free replies for high-volume chat.$0.09 in / $0.17 out per M tokens
BasicGLM 5.3 FlashThe speed tier of GLM 5.3 — sub-cent turns, huge context.$0.07 in / $0.25 out per M tokens
Get started

Three steps to your first brew

01

Get an OpenRouter key

Pay-as-you-go at openrouter.ai/keys. No subscription — most chats cost fractions of a cent.

02

Install the app

Android: grab the APK from Releases. iOS or dev: run it with Expo Go from source.

03

Paste the key, talk

Default is a balanced workhorse. Pick any model in the header — or choose Auto and let the harness route.

$ git clone https://github.com/BobbyNooby/honestea && cd honestea
$ pnpm install
$ pnpm dev:app        # scan the QR with Expo Go
FAQ

Questions, answered.

What does HonesTea cost?+

The app is free and open-source-by-inspection. You pay OpenRouter directly for tokens at cost — the app applies 0% markup and shows you every charge. A typical GLM-flash chat costs a few hundredths of a cent.

Do I need an account?+

No. There is no account, no sign-up, and no server. Paste an OpenRouter key on first launch and you're done. Delete the app and everything is gone — by design, not policy.

Where does my data live?+

On your phone. Chats are in a local SQLite database, keys are in the iOS Keychain / Android Keystore, and requests go straight from your device to OpenRouter over TLS. This repo contains no server code at all.

Which models can I use?+

Everything OpenRouter carries — 400+ models from Anthropic, OpenAI, Z.ai, DeepSeek, Moonshot, Google, Meta, Qwen and more. The picker curates seven good defaults; the model browser shows the full catalog with live pricing.

What does "Auto" do?+

It routes each message to the cheapest model that meets that turn's requirements — vision if you attached an image, tool support if web search is on, enough context for the projected prompt. If nothing fits perfectly it relaxes requirements instead of failing.

Is there an iOS build?+

The app is iOS-ready but prebuilts are Android-first right now. On iOS you can run it from source with Expo Go: clone, pnpm dev:app, scan the QR.

Stop renting your chat history.

Bring your own key, see every cent, and keep every word on your own phone.

Download the latest APK