TokenAssemble
TokenAssemble is a web-based platform designed to help users determine whether their hardware—GPUs, Macs, or mini-PCs—can run local AI models using runtimes like Ollama, LM Studio, or llama.cpp. It provides detailed compatibility checks including VRAM fit, quantization levels, context length, and speed tiers. The service emphasizes transparency by showing sourced and stamped data and performing computations directly in the user's browser. The platform connects the five layers of a local AI setup: hardware, model, quantization, runtime, and software. Users can start from a hardware component and see which models fit, or start from a desired model and find suitable hardware. It also offers a stack builder that ranks complete setups within a given budget, a comparison tool for GPUs, Macs, and mini-PCs, and an ROI calculator to evaluate buying hardware versus using cloud APIs. TokenAssemble includes workflow templates for common local AI use cases such as a private coding assistant, a self-hosted chat server, a transcription and summary pipeline, and a local agent stack. These templates guide users through selecting compatible software, runtime, and model, then grade the hardware underneath. The tool avoids fabricating tokens-per-second figures, instead using speed tiers based on measured data. The site maintains a directory of GPUs, prebuilt systems, models, runtimes, and local-AI software, all with sourced and dated entries. It also provides an MCP (Model Context Protocol) connection for programmatic access. TokenAssemble is free to use and focuses exclusively on local AI hardware compatibility.
- Category
- Model Serving
- Pricing
- Free
Problem solved
TokenAssemble solves the problem of uncertainty when selecting hardware for local AI models. Users often struggle to know if a particular GPU or Mac has enough VRAM or unified memory for a model at a given quantization and context length, and whether the performance will be acceptable. TokenAssemble provides evidence-based compatibility verdicts, speed tiers, and runtime support information, eliminating guesswork and helping users avoid purchasing unsuitable hardware.
Use cases
- Check if a specific GPU or Mac can run an LLM with a given quantization and context length
- Build a complete local AI stack (hardware, model, runtime, software) within a budget
- Compare GPUs, Macs, and mini-PCs for local AI workloads
- Estimate break-even point for buying local hardware versus using cloud APIs
- Set up a private coding assistant with a locally served model
- Deploy a self-hosted chat server for a household or team
Audience
- Developers
- ML Engineers
- Data Scientists
- Researchers
- Hobbyists
Pros
- Transparent methodology with sourced and stamped data
- Computes feasibility in the browser without uploading user data
- Covers all five layers of a local AI setup from hardware to software
- Provides speed tiers instead of fabricated tokens-per-second numbers
- Offers practical workflow templates and ROI analysis
Cons
- Limited to local AI hardware compatibility, not general model serving
- No self-hosted version available; requires internet access
- Data snapshot is dated and may lag current hardware or model releases
- Dependent on manually curated data that could be incomplete or outdated
Official website: https://tokenassemble.com/
Canonical LambdaBase page: https://www.lambdabase.com/tools/tokenassemble