flux://router
initializing routing engine
0%
Open source · AGPL-3.0 License

Save 40% on LLM costs. Automatically.

One import. Intelligent routing. Real savings. Flux picks the cheapest model for every request — no config needed.

0.23ms P50Routing latency
40%Average cost savings
98×Faster than LLM-based routers
req incoming claude-haiku-4.5 gemini-2.5-flash gpt-4.1-mini claude-sonnet-5 llama-4-scout
Resolved model
claude-haiku-4.5
Task type
short reply
Est. cost
$0.00014
The problem

You're wasting money on LLMs

Teams waste 40% of their LLM budget before they even know it.

cost

Using Opus for simple summarization

Most teams default to the most capable model for every task. $0.01 per request could be $0.001.

blind

No visibility into model performance

You don't know which models work best for your tasks — or which ones are draining your budget.

manual

Manual selection is slow and error-prone

Humans can't optimize model selection at scale, especially across multiple task types and customers.

How it works

Flux picks the right model. Every time.

A 13-step algorithm with pure heuristics — zero AI calls in the routing path.

01
Classify
Task type detection
02
Filter
Capability match
03
Score
Cost × quality
04
Pick
Best-fit model
05
Learn
Adaptive feedback
<5ms latency
Pure heuristics
Zero AI calls
Adaptive learning
Savings calculator

How much could you save?

Based on real routing optimizations across diverse task types.

$2.0k/month
At 30–50% savings, you could save
$600–$1.0k
per month
That's $7.2k–$12.0k/year back in your budget.
Start Saving Now →
Built for production

Everything you need to optimize your LLM infrastructure.

No black box — every routing decision is explainable, fast, and tunable per customer.

select

Automatic model selection

Routes every request to the optimal model — no config, no guesswork.

learn

Adaptive learning

Improves routing accuracy from every response. Gets smarter over time.

cost

Cost breakdown

Real-time analytics showing spend by model, task, and customer.

tune

Per-customer tuning

Set different routing preferences for each of your customers.

perf

Lightning fast

<5ms routing latency. Pure heuristics — zero AI calls in the path.

oss

Open source

AGPL-3.0 licensed. Self-host with full access to the source. Commercial licensing available — contact fluxllmdev@gmail.com

15,000 routing decisions. 95.8% completed in under 1ms. Zero errors.

From our internal benchmarks across diverse task types.

Teams building chatbots

Route simple replies to cheaper models, complex reasoning to Opus — automatically.

Document analysis

Summarization, extraction, classification — each task gets the right model.

Code generation tools

Balance quality vs cost per developer tier. Per-customer tuning built in.

Benchmark — 500 synthetic requests
measured live
0ms
P50 latency
0
Tests passing
0
Models supported
0
Providers
Early access

Pricing coming soon

Join the waitlist for early access and founding member pricing.

We're an early-stage startup building Flux in the open. For any queries, reach us at fluxllmdev@gmail.com

Quick start

Start in 30 seconds

Drop Flux into any Python project with a single import. No API key required for local routing.

No API key required
No signup for local use
Works with all major LLM providers
Read the Docs →
# Install
git clone https://github.com/vbc1406/flux
cd flux
pip install -r requirements.txt

from router.flux import make_flux

flux = make_flux()
decision = flux.route(
    prompt="Summarize this document...",
    task_type='summarization'
)

No API key. No signup. Just works.

Save money on
LLMs today

Intelligent routing. Automatic savings.