Free
your own keys · any provider
€0with your keys
- Every feature unlocked
- Any provider key, incl. local LLMs
- 4-tier memory + reversible compaction
- No account required
pricing
Free to run. Pay only to skip setup.
The full CLI is free forever with your own API keys. Prefer not to manage keys? Buy a request pack or top up a balance — routed through our managed gateway. No subscription, no auto-renew.
Have an account? Sign in to track your balance, packs, and license keys.Sign in →
Starter
DeepSeek V3.2 · Qwen3 · GLM
€4.99one-time
- 300 requests
- 30-day validity
- Fast open models
- Fair-use token cap
- Falls back to your keys
Builder
Haiku 4.5 · GPT-5 mini · Gemini Flash
€9.99one-time
- 300 requests
- 30-day validity
- Balanced frontier-lab models
- Higher token cap
- Falls back to your keys
Frontier
Sonnet 4.6 · GPT-5.x
€19.99one-time
- 150 requests
- 30-day validity
- Top frontier models
- Highest token cap
- Opus-class joins after launch
pay as you go
Every model
Opus-class included
Top up a balance (from €5) and pay the actual model cost + 10% service fee per request. Balance never expires. Set a monthly spend limit and stop whenever. The card processing fee is added once at top-up — the full amount you choose lands as balance.
Packs and pay-as-you-go are in development — counts, prices, and fees may adjust at launch. Follow the changelog for the release.
how it counts
What is a request?
1 prompt = 1 request
One message you send to the agent counts as one request — regardless of how many tokens it takes, within the pack's fair-use token cap.
Your license key is the login
No provider account needed. The CLI authenticates to the gateway with your Anthracode license key and the balance counts down.
Expires, never renews
A pack lasts 30 days or until the requests run out — whichever comes first. No card on file, no surprise charge, ever. (Pay-as-you-go balance never expires.)
faq
Pricing and billing questions. The agent itself is always free with your own keys.
What's free, exactly?
The full TUI agent — every feature: 4-tier memory, reversible compaction with recall, PTY sessions, background processes, subagents, skills hub, native Windows, macOS, and Linux. Bring your own provider key (Anthropic, OpenAI, OpenRouter, local Ollama…) and pay that provider directly. Nothing is gated locally.Packs vs pay-as-you-go — which do I pick?
A pack is the simplest: one fixed price, a known number of requests, no card kept on file. Pay-as-you-go is for heavier or frontier+ work — you top up a balance and pay the real model cost plus a 10% service fee, so the priciest models (Opus-class) stay available without a risky flat price. Both are optional; the free tier with your own keys never goes away.Why packs instead of a subscription?
You probably already pay for enough subscriptions. A pack is a one-time purchase: it can't forget to be cancelled, can't bill you while you're on holiday, and costs nothing when you're not building. Buy one when you need it, walk away when you don't.Where do I see my balance?
On your billing page once you're signed in. Active packs, requests remaining, expiry, and any pay-as-you-go balance all render there — and the CLI shows the same numbers inline. While the gateway is in development the billing page states your tier honestly and nothing is charged.Is there a token limit per request?
Yes — each pack tier has a generous fair-use token cap per request, sized so normal coding work never hits it. It exists to keep pack prices honest: without it, one pathological request could cost more than the whole pack. Pay-as-you-go has no cap — you pay for exactly what you use.What happens when my pack runs out?
The CLI tells you the balance is empty and falls back to your own API keys if you have any configured — the free tier never breaks. Buy another pack whenever you want; nothing happens automatically.When can I buy one?
The gateway and checkout are in active development. The buy buttons activate the day it ships — watch the changelog or the GitHub repo for the announcement. Prices shown are launch targets.