breviadev 0.1.1

Save 55-93% on Claude API tokens. Same quality, fraction of the cost.

Brevia

Use Claude smarter. Save 55-93% on tokens without losing quality.

Brevia is a local proxy that makes your Claude API calls cheaper and often better. It runs between your tools and Anthropic, automatically trimming unnecessary context so Claude focuses on what actually matters.

Works with Claude Code, Cursor, Continue, aider, and anything that uses the Anthropic SDK. No code changes needed.

---

Quick Start

pip install breviadev
brevia login          # Opens browser — sign in with GitHub or Google
brevia serve          # Starts on localhost:8420

Then add to your shell profile (~/.zshrc, ~/.bashrc):

export ANTHROPIC_BASE_URL=http://localhost:8420

That's it. Everything works exactly as before — but cheaper and often better.

---

How It Works

Your Tool (Claude Code, Cursor, etc.)
    │
    │  ANTHROPIC_BASE_URL=http://localhost:8420
    ▼
┌─────────────────────────────────┐
│           Brevia                │  ← Runs on your machine
│                                 │
│  Reads your request, figures    │
│  out what's relevant, removes   │
│  the noise, and enhances the    │
│  prompt for better results.     │
└─────────────────────────────────┘
    │
    │  Smaller, focused payload
    ▼
┌─────────────────────────────────┐
│      api.anthropic.com          │  ← Your API key
│                                 │     Your account
│  Claude gets less noise,        │     You pay less
│  gives better answers.          │
└─────────────────────────────────┘
    │
    │  Response streams back
    ▼
Your Tool (unchanged behavior)

Why it works: When Claude gets 2k tokens of the right code instead of 95k tokens of everything, it gives more precise answers. Less noise in, better signal out.

---

Benchmarks

Tested on real codebases (Django, FastAPI, psf/requests) with Claude Opus:

Context Size	Cost Savings	Quality
< 2k tokens	0% (passthrough)	Same
10k tokens	~55%	Same
50k tokens	76-93%	Same
95k tokens	76%	Better (more precise)

Real-World Example

Setup	Total Cost	Compared to Direct
Direct Claude (full context)	$0.836	—
With Brevia	$0.278	67% cheaper

Large Data (50k token input)

Metric	Value
Token reduction	97%
Cost savings	93%
Accuracy	Perfect (found all issues)

Full benchmark details: benchmarks/BENCHMARKS.md

---

Commands

Command	What it does
brevia login	Sign in (opens browser)
brevia serve	Start Brevia
brevia serve -p 9000	Start on a different port
brevia stats	See how much you've saved
brevia stats -d 30	See last 30 days
brevia logout	Sign out

---

What You'll See

When brevia serve is running:

╭─ Brevia ─────────────────────────────────────╮
│ Brevia is running                             │
│                                               │
│   Address:  http://127.0.0.1:8420             │
│   User:     @yourname                         │
│   Status:   Active                            │
│                                               │
│   Add to your shell:                          │
│   export ANTHROPIC_BASE_URL=http://127.0.0.1:8420 │
╰───────────────────────────────────────────────╯

Check your savings anytime with brevia stats:

╭─ Brevia Stats ────────────────────────────────╮
│                                               │
│   Days active:    12                          │
│   Total requests: 847                         │
│   Tokens saved:   4,230,000                   │
│   Avg reduction:  71%                         │
│   Est. $ saved:   $63.45                      │
╰───────────────────────────────────────────────╯

---

Where It Helps Most

• Big contexts (50k+ tokens): The more noise, the more Brevia saves
• Multi-file projects: Keeps only the files that matter for your question
• Repetitive code: Strips boilerplate so Claude focuses on the real problem

Where It Doesn't Help

• Short prompts (< 2k tokens): Already small — Brevia passes these through unchanged
• Already focused: If you're manually sending only relevant code, there's nothing to trim

---

Privacy & Security

• Your API key stays yours — Brevia passes it through, never stores it
• Your code stays local — nothing leaves your machine
• We only collect usage stats — token counts and savings, never content
• Credentials are stored locally in ~/.brevia/ with restricted file permissions

---

Platforms

• macOS (Intel + Apple Silicon)
• Linux (x86_64 + ARM64)
• Windows 10+

Requires Python 3.10+.

---

Enterprise

Need team-wide deployment, custom rules, or dedicated support?

Contact us: [e-mail]

---

How It Compares

	Brevia	Prompt Caching	Manual Trimming
Setup	One command	Built into SDK	You do it yourself
Effort	Zero — automatic	Zero — automatic	High — manual work
Savings	55-93%	Varies (cache hits only)	Depends on you
Quality	Same or better	Same	Risk of cutting too much
Works with	Any Anthropic tool	SDK only	Your code only

Brevia works alongside prompt caching — use both for maximum savings.

---

License

Proprietary. Free for individual use. See LICENSE for details.

---

Built for developers who'd rather spend money on building, not on sending noise to an API.