Jini
CLI Guide
The small public command surface and the setup check that sits behind it.
Jini should feel small. Most people should learn one command, not a command tree.
You do not need to know Python to use Jini.
Front door
Start here
jini
That should be the normal entry. Jini should either continue the thing you were already working on, show Active work when several projects are in flight, or offer simple choices if nothing is active yet.
Inside Jini
Main in-shell actions
Help me plan this is the structured mode. Use it when the work is still fuzzy and you want Jini to turn it into goal, requirements, design, steps, and run.
Decision card
What Jini should show before new work starts
Route context
- Tool
- Provider
- How chosen
- Continuity when Jini stays on the current coding route
Model context
- Model
- Effort level
- Why this route
- Verification depth when that matters
Multimodal context
For screenshot, scanned PDF, and audio work, Jini should make the subtype-specific learning and route reason explicit instead of using one generic multimodal explanation.
Scriptable surface
Commands that stay available for automation and power users
jini check
Shows a calm work summary: what you are working on, what is ready, what is missing, and the next step.
jini open
Opens useful outputs like a follow-up, memo, or check instead of internal file paths.
jini provider doctor
Local setup check for Claude, Bedrock, Azure OpenAI, Local SLM, or local preview.
Setup check
What provider doctor is for
jini provider doctor
jini
Most people should still start by pasting the work they want finished. Use doctor when setup help is needed, when you need one strict route, or when you are debugging access.
Doctor reports what Jini will use, what auto mode resolved to, and what is missing. It does not print secret values.
Local SLM
What extra context doctor shows for local routing
Machine view
DEVICE_CLASSDEVICE_OSLOCAL_ACCELERATORLOCAL_RUNTIME_CLASS
Learning buckets
Jini keeps separate learning for screenshot work, scanned PDF work, and audio/transcript work.
Measured signals
When the local endpoint is ready, doctor can refresh benchmark summaries for load success, warm latency, cold start cost, token throughput, and structured-output reliability.
The public rule
The normal path is still small: install once, run jini, paste the work you want finished. The rest of the CLI exists to support that experience, not replace it.