1. Laptop or workstation evaluation
Use the TUI to filter for coding, chat, reasoning, or multimodal tasks and compare which models fit the machine you already own.
llmfit
llmfit recommend --json --use-case coding --limit 5Use Cases
Where LLMFit becomes operationally useful.
The best use case for LLMFit is any workflow where local AI decisions need to be faster, less wasteful, and easier to defend. The tool is especially effective when the same question repeats across many machines or projects.
2. Fleet standardization
Run serve mode on each node and let a scheduler or inventory service query a normalized set of top runnable models.
llmfit serve --host 0.0.0.0 --port 8787
curl http://node:8787/api/v1/models/top?min_fit=good3. Hardware planning
Start with a target model and ask what hardware is required, rather than buying hardware first and discovering the model is impractical later.
llmfit plan "Qwen/Qwen3-4B-MLX-4bit" --context 8192 --target-tps 254. Advisory or consulting work
Use LLMFit to turn “what should we run on this hardware?” into a repeatable recommendation that can be reviewed with a client or internal team.
Audience detail
Professional user profiles.
Solo builder
Needs a practical answer fast: which model should run on a laptop or workstation without wasting storage and setup time.
- Prefer the TUI and CLI
- Filter by coding or general-purpose use cases
- Compare a handful of realistic candidates
Platform engineer
Needs a consistent way to expose node-local model availability into a larger internal platform or scheduler.
- Prefer `llmfit serve`
- Poll `/api/v1/system` and `/api/v1/models/top`
- Use `min_fit=good` as a conservative baseline
Solution consultant
Needs a defensible recommendation process for customer hardware rather than ad hoc personal preference.
- Use plan mode to show upgrade paths
- Capture JSON results for documentation
- Explain tradeoffs in fit, speed, and context
Homelab operator
Needs to squeeze useful local AI behavior out of mixed CPUs, smaller GPUs, and edge hardware.
- Prioritize runnability over leaderboard prestige
- Use context limits and quantization awareness
- Review marginal vs perfect fit behavior carefully
Workflow examples
Three repeatable operating patterns.
Pre-download decision
Use the TUI or CLI before pulling a model. This is the easiest way to prevent wasted downloads on underpowered machines.
Node-local recommendation service
Run `llmfit serve` on each machine, then let a separate control plane aggregate and decide across many nodes.
Upgrade planning
Use plan mode when you already know the target model family and want to size the hardware path needed to get there.
Next step