Local-first AI workspace

Sage

An AI workspace that runs on your machine by default, keeps your conversations in your control, and helps you move from a question to a verified next step.

Run Sage locally See the privacy model

Default setup: local Ollama model, no API key. Hosted providers are optional and can carry provider charges when you configure them.

Sage workspace Local

PROJECT WORKFLOW

Map this project

2m ago

Where is chat handled, and what should I verify first?

Inspecting the local project

S Sage

Chat streams from app.py. The local-agent path lives in sage/local_agent.py.

Start with the project map, then run the focused selftests.

+ Proposal readyReview a change before it is applied. +18 -2

Ask Sage about this workspaceSend

Illustrative local workspace. Sage stores user data on the machine running it.

Own the runtime. Start local. Decide when anything leaves your machine.

Keep a trail. Conversations, tool activity, proposals, and verification stay inspectable.

Change with approval. Sage can propose code changes; people approve what gets applied.

Start where your data is

Local by default. Explicit by choice.

Sage starts with a local Ollama model and local storage for conversations, memory, and opt-in browser location. You can export or delete that data from the app.

Optional hosted providers are clearly labeled in Settings, where every provider is marked as free and local or as a paid API before you can use it. Nothing calls a paid provider unless you add and enable its credentials yourself.

From question to work

Workflows keep useful work findable.

Explore

Map a project, inspect files, and keep tool activity attached to the conversation that used it.

Build

Use the dedicated code workspace to review context, diffs, and proposed changes before they touch source.

Verify

Run the right checks, keep the result visible, and promote improvements only when measured evidence supports them.

Mac local setup

Install without turning your machine into a cloud account.

1
Get Sage from a trusted release
Use the release folder or archive you received. The installer keeps developer history, keys, and existing local Sage data out of the installed copy.
2
Open install.command
Sage creates a clean copy in your Applications folder and prepares its local Python environment.
3
Open Sage
The launcher checks for Ollama, builds the local Sage model when it is missing, and opens the app in your browser.

What you need: a current Mac, Python 3, Ollama, and enough free disk space and memory for the local model. Sage explains missing setup pieces instead of silently falling back to a paid provider.

Privacy model

Your conversations are not the product.

Local chat: Default conversations and memories live on the machine running Sage.
Internet tools: Web search and weather reach their named external sources only when you request those tools.
Hosted providers: Optional providers receive prompts only after you explicitly configure and select them. Their billing and retention rules are their own.
Backups: Export a versioned local data bundle when you choose. Provider credentials and app configuration are excluded.

Measured, not mythologized

Sage is a product agent, not a frontier-model claim.

Local models have real limits. Sage publishes raw and repair-assisted coding measurements separately, tests tool behavior, and keeps a human approval step around source changes.

ALX also runs a separate research experiment that trains a small original model from scratch. That experiment is not part of this release, and nothing Sage ships depends on it.

Local latency targetsMeasured on the reference Mac

Tool behavior checksRun before a release candidate

Approval-gated changesNo silent source edits

Support path

Start local. Keep control. Ask for help with evidence.

For setup help, begin with Sage's built-in health message and local installer notes. Before sharing any diagnostic details, remove private conversation content and never include provider keys.