Choose how you run Flux
Flux OS is free and you install it yourself. Flux Custom Plans, Enterprise Solutions and Managed SLM Deployment are sold under a separate signed agreement. There is no self-serve checkout.
| Feature | Flux OS | Flux Custom Plans | Enterprise Solutions | Managed SLM Deployment |
|---|---|---|---|---|
| What it is | A router that picks the cheapest model that clears a quality bar and reroutes when a call fails | The full Flux gateway with run-level budget governance | Flux for larger deployments, set up around your workflow | An open-license small language model fine-tuned on your data |
| License | MIT, open source | Proprietary | Proprietary | The base model's own license; our work under contract |
| Where it runs | Your machine or servers | Your infrastructure | Your infrastructure | Your server, on-prem or private cloud |
| Budgets, dashboard, spend tracking | No. Flux OS also has no telemetry | Yes | Yes | Not applicable |
| How you get it | Install it yourself | Written agreement | Written contract | Written contract |
| Price | Free | Contact us for pricing | Contact us for pricing | Setup fee plus maintenance fee. Contact us for pricing |
Flux OS: free and open source
Flux OS is a free, open-source router released under the MIT license. It reads each request, sets a quality bar, and sends it to the cheapest model that clears it. If that model fails with a rate limit, outage or timeout, Flux OS reroutes the request to the next-best model, trying a different provider first.
It does routing and automatic fallback only. There are no budgets, no dashboard, no spend tracking and no telemetry; those come with Flux on a custom plan. You install and run it yourself.
flux-os route output, from the Flux OS README
# install and ask Flux OS what it would do (no API keys needed)
pip install "git+https://github.com/vbc1406/flux-os"
flux-os route --all "Prove that there are infinitely many primes"
model: gpt-5.6-luna (openai)
task: reasoning complexity 0.55 quality bar 0.84
est. cost: $0.001203
reroute to: gemini-3.1-flash-lite, gemini-3-flash-preview, gemini-3.8-flashIllustrative output. The model picked and the estimated cost depend on your configuration and current provider prices.
Flux Custom Plans
The full, proprietary Flux product. Each agent run gets one shared budget, Flux shifts to cheaper models as the run uses that budget, hierarchical per-agent caps limit what each agent can spend, and the local dashboard shows it all. It runs self-hosted in your infrastructure, and we set it up around your agents, providers and tenants.
Plans are priced by governed agent runs and tenants per month, with no revenue share, and are sold under a written agreement. The Flux documentation shows how the product is configured.
Enterprise Solutions
For larger deployments that need custom quality floors, dedicated support, or a setup tailored to their workflow. Enterprise also covers commercial licensing if you plan to embed Flux in your own product, offer it as a hosted service or otherwise redistribute it.
Enterprise work is sold under a written contract.
Managed SLM Deployment
We take an open-license small language model, for example Qwen, fine-tune it on your data, and deploy it on your own servers, on-prem or in your private cloud. You pay a setup fee, then an ongoing maintenance fee.
Some companies want the model and their data to stay in-house. This service is for them.
The base model's own license applies to the model we deliver, and we check that it allows your intended use before we start. You are responsible for having the rights to the data used for training. How well the model performs depends on your data and your use case, and we do not guarantee any performance beyond what the signed statement of work says. Contact us for pricing.
What you pay for on a custom plan
Custom plans follow the unit Flux governs: the agent run. Plans are set by how many governed runs and tenants you operate each month. Tokens and savings don't enter into it.
- runsGoverned agent runs. A run is every call that shares one
X-Flux-Run-Idheader, including retries and subagents. Each run gets one shared budget: Flux degrades to cheaper models from 70% of it, raises a warning at 90% and hard-stops with a 429 at 100%. The agentic run guide shows the setup. - tenantsTenants. A tenant is a customer or account you tag with
customer_id. Each tenant gets its own cost tracking, daily ceilings and, optionally, its own adaptive routing scores. See multi-tenant deployment in the docs. - modelsModel usage. Model calls are billed by the providers you configure (OpenAI, Anthropic, Google, Groq, Mistral) on your own API keys. Flux doesn't resell tokens. The savings benchmark methodology explains how estimated savings were measured.
Comparing options? Flux vs OpenRouter vs LiteLLM sets out how run-level budgets differ from per-key and per-team limits.
Not sure which one fits?
Tell us what you're running and what you need to control. Use the Flux contact form or email vandith@flux-llm.com.