Jini
Install
Install Jini once, then use `jini`.
The normal path should be simple: install once, run jini, paste the work you want finished.
Recommended
Install once from any terminal
curl -fsSL https://raw.githubusercontent.com/maridlabsai/jini/main/install.sh | bash
The installer first tries to install a matching release binary. If needed, it falls back to a source-backed runtime, verifies that the command launches, and prints one PATH fix only when your shell still cannot see jini.
jini
If the installer prints a PATH line, run it once in the current shell and add it to your shell profile later.
First run
What should happen next
Run jini
Do not start with provider jargon unless Jini tells you setup is missing.
Paste the work
Start from the notes, draft, screenshot, transcript, or rough ask you already have.
Use Auto if needed
If setup is missing, type Use Auto and let Jini help you connect the best available route.
Choose your path
Only use the strict setup blocks when you actually need them
Use Auto
Best for most people. Jini chooses the cheapest suitable route by default and escalates only when the request clearly needs deeper work.
Connect Claude
Best when you already use Anthropic directly and want Claude as the strict route.
Connect Bedrock
Best when AWS policy or existing Bedrock access decides the route.
Connect Azure OpenAI
Best when company policy requires Azure OpenAI only.
Connect Local SLM
Best when you already run an OpenAI-compatible local model server and want cheap-first local routing.
Copy-paste paths
Strict route setup
<h3>Claude direct</h3> ```bash export JINI_PROVIDER=claude export ANTHROPIC_API_KEY="paste-your-key-here" export JINI_MODEL=sonnet jini provider doctor jini ```
<p>Use this when your team already gave you a direct Anthropic key and Claude should be the fixed route.</p>
</div>
<div class="scenario-card" markdown="1">
<h3>Amazon Bedrock</h3> ```bash export JINI_PROVIDER=bedrock export AWS_REGION=us-east-1 export AWS_PROFILE="your-profile" export JINI_MODEL=sonnet-4.6 jini provider doctor jini ```
<p>Use this when AWS policy or Bedrock access already exists. If you know the exact model id, you can set <code>BEDROCK_MODEL_ID</code> instead.</p>
</div>
<div class="scenario-card" markdown="1">
<h3>Azure OpenAI</h3> ```bash export JINI_PROVIDER=azure-openai export AZURE_OPENAI_ENDPOINT="https://your-resource.openai.azure.com" export AZURE_OPENAI_API_KEY="paste-your-key-here" export AZURE_OPENAI_DEPLOYMENT="your-deployment-name" export AZURE_OPENAI_API_VERSION=2024-10-21 jini provider doctor jini ```
<p>Use this when your company requires Azure OpenAI only. On Azure, the deployment decides the actual model.</p>
</div>
<div class="scenario-card" markdown="1">
<h3>Auto route</h3> ```bash export JINI_TOOL=auto export JINI_PROVIDER=auto export JINI_MODEL=auto jini provider doctor jini ```
<p>Use this if you want Jini to choose the cheapest suitable route, model, and effort level for each request.</p>
</div>
<div class="scenario-card" markdown="1">
<h3>Local SLM</h3> ```bash export JINI_PROVIDER=local-slm export JINI_TOOL=auto export JINI_MODEL=auto export JINI_LOCAL_SLM_ENDPOINT="http://127.0.0.1:11434/v1" export JINI_LOCAL_SLM_MODEL="qwen3:8b" jini provider doctor jini ```
<p>Use this when you already run an OpenAI-compatible local model endpoint and want local workhorse routing.</p>
</div>
Setup check
What jini provider doctor actually tells you
What it does
- shows what Jini will use
- shows what auto mode resolved to
- shows what settings are still missing
- does not print secret values
What it does not prove
- your AWS auth is valid
- your Bedrock model access is enabled
- your Azure key is accepted by the service
- your Claude account is allowed to use a model
Local SLM extras
When local routing is active, doctor also shows the machine and runtime view Jini is using, including device class, local runtime class, and accelerator context.
Most people should stop here
Install Jini, run jini, and paste the work you want finished. Only drop to the strict route blocks when policy or debugging requires it.