bindwidth
Evidence-aware on-prem LLM inference sizing and TCO calculator
What is it?
What it is
A browser‑only calculator that sizes private local AI deployments by evaluating memory, serving, and runtime constraints.
Why it exists
Most online sizing tools mis‑identify the binding constraint, compare against non‑comparable options, or ignore the cheap metal cost. Bindwidth makes the assumptions explicit, identifies the true binding limit, and transparently compares owned hardware, sovereign rentals, and enterprise subscriptions so users can avoid buying against slack and understand the real cost of local AI.
Who should use it
Data scientists, ML engineers, and IT procurement teams looking to size private local AI deployments; architects evaluating hardware and cost trade‑offs for on‑prem or sovereign hosting; organizations needing a transparent, constraint‑driven sizing tool for LLM workloads.
Who should avoid it
End‑users or hobbyists who just want to run a model locally without detailed sizing; teams that prefer commercial cloud APIs without local deployment concerns; developers needing a full production framework rather than a sizing calculator.
How it works
A quick walkthrough in plain English
How bindwidth works in 4 steps
Step 1 of 4
Something triggers the flow
A schedule, webhook, or manual click tells bindwidth to start.
Features
Advantages
- Accurately identifies binding constraints to avoid oversized deployments
- Transparency in comparing deployment options with capability/service differences
- Confidence levels in data ensure reliability of estimates
- Exportable reports enable documentation and scenario reuse
- Guided setup reduces initial configuration complexity
Disadvantages
- Does not evaluate whether premium costs are justified for specific use cases
- Cannot compare quality, latency, or context between local and subscription models
- Data governance decisions require external legal/risk analysis
- Requires manual input for precise benchmarks and configurations
- Frontend-only design may limit advanced server-side processing needs
Installation
native
git clone https://github.com/juxhinr/bindwidth.git cd bindwidth python3 -m http.server 8080 # open http://localhost:8080
FAQ
What is the primary purpose of Bindwidth?
Bindwidth is a sizing calculator designed to help users determine the technical minimum requirements for running private, local AI. It identifies which specific constraint (KV cache memory, serving capacity, or runtime session ceiling) 'binds' your deployment, preventing over-investment in unnecessary hardware.
Is my data secure when using the calculator?
Yes. The tool is frontend-only with no build step, no server, no database, and no account requirements. Everything runs entirely within your browser tab, and nothing you type is transmitted to any external server.
How does the tool handle different types of workloads?
The tool treats interactive users and autonomous agents as structurally different loads. It accounts for the 'peak factor' of human users (who produce violent bursts of activity) versus the flat, high-cache-hit profile of autonomous agents running around the clock.
Can I export my results for later use?
Yes. You can export a decision record as a Markdown file or export the full scenario as a JSON file, which can be reloaded into the calculator at a later time.
How can I request a pricing quote for a specific configuration?
In the calculator, you can use the 'Request quote' feature. This copies your scenario summary and opens a Google Form so the team can provide pricing and next steps for your specific configuration.
How are the hardware and model profiles sourced?
The tool uses a combination of measured data, published recipes, and estimates. It distinguishes between 'easured' (verifiable first-party runs) and 'estimated' (derived via arithmetic) to provide a clear confidence level for every profile used in a calculation.
Featured in Videos
YouTube tutorials and walkthroughs for bindwidth
Alternatives
Similar projects ranked by category, topics, and text overlap.
