Junction · Local AI

Give your local models a task.
Draft, review, compare, and keep the whole trail.

API spend: $0

New run

01 / CREATE

Project

Workflow

Primary model What should the model work on?

Context & limits

Instructions

Output tokens per step

Run token budget

Search project memory Allow calculator tool (direct workflow)

Run output

Ready when you are

↗

Room to think

Start a run or open one from your history.
Your answer and every step will appear here.

Execution trail

Recent runs

Models on your own machines.
Only explicitly allowed, private endpoints can be used.

Connect another model

Edit providers.json in the project folder and restart Junction. Add the exact model identifier to its allowlist. Public provider addresses and cloud model names are blocked.

For a VPS, connect to your model host through a private VPN or run Ollama on the VPS. A model on your computer is available only while that computer and its model server are running.

Keep useful reference material.
Notes belong to a project and stay within its access boundary.

Add a note

Find a note

Give each client the access it needs.
Scoped keys, revocation, and an audit trail.

Run workspace