01
Capture a thought
HUMAN
A first version, with a clear first-run experience.
# ideaLOCAL-FIRST · OPEN SOURCE
You think here. Agents act on it. Knowing and doing grow together.
A local knowledge workspace. Shared context for the agents you use.
- Open source
- Local Markdown
- Human in control
HUMAN + AGENT
HUMAN
You
AGENT
Agents
Shared context
Your direction
Next context
Your agents. One shared context.
Connect through MCP or Skills. Support varies by client.-
Cursor
-
Codex
-
Claude Code
-
Gemini CLI
-
Kimi Code
-
WorkBuddy
-
Doubao WorkConnection unverified
-
OpenCode
-
Windsurf
-
Cline
01 / THE WORKSPACE
From context
to action.
Start with a note. Put it to work.
Actual interface. Click to explore.
02 / FROM THOUGHT TO ACTION
One thought. A wiser next step.
You set the direction. Agents act. Review the result and carry the lesson forward.
Workflow illustration
THE CONTINUOUS WORKBENCH
02
Make it shared context
MindOS
.md
Projects / launch.md
The next release
A clear first-run experience
Review important changes
Profile
Projects
Workflows
03
Put it to work
Read the same project context
Claude Code
Review the first-run flow
Codex
Build the proposed changes
- Your review
- Keep the lesson
YOUR JUDGMENT
Set the direction.
Write down the goal, context and boundaries.
Goal: ship the first version.
Priority: the first-run experience.
Boundary: review key changes first.
Judgment becomes shared context.
AGENT ACTION
Move it forward.
Agents read the context and produce work you can review.
Read the project goal.
Review the first-run flow.
Explain changes and tradeoffs.
Context becomes action.
YOUR REVIEW
Keep the lesson.
Keep useful judgments. Correct the assumptions.
The shorter path is clearer.
Key choices still need to be visible.
Keep this lesson.
Action becomes experience.
CO-EVOLUTION
Start further ahead.
Reviewed lessons become context for the next task.
Make the next action clear.
Keep key choices visible.
Explain the reason.
Clearer judgment. Better agent context.
Workflow illustration. Knowledge updates do not retrain a model.
03 / THE DIFFERENCE
Same task.
A different starting point.
Spend less time rebuilding context.
Scattered context
Start the explanation again.
- Repeat the goal and constraints.
- Rebuild the current project state.
Shared with MindOS
Continue from shared notes.
- Connect the next agent to the project notes.
- Continue from the same written decisions.
A common starting point.
Scattered context
Find the history. Retell the story.
- Search old conversations for the decision.
- Piece together what changed.
Shared with MindOS
Bring the project into the session.
- Point the agent to reviewed project context.
- Check what it read before proceeding.
The context stays within reach.
Scattered context
A useful lesson stays in a chat.
- Repeat a mistake from the previous task.
- Write the same correction again.
Shared with MindOS
A reviewed lesson helps next time.
- Keep the lesson in a note or workflow.
- Use it as context for the next task.
Experience you can carry forward.
04 / YOUR CONTROL
Your knowledge.
Still yours.
Keep it local.
Markdown files you can edit, back up and take with you.
Review the work.
Inspect context and important changes.
Connect your way.
MCP and Skills connect agents. Cloud models may receive selected context.
Keep the lesson.
Take it further.
05 / START WITH ONE THOUGHT
Your next loop
starts here.
Get MindOS. Connect an agent. Bring your first note.
Desktop app
Latest desktop releaseOpen this page on your computer to install.
M-series Macs: Apple Silicon. Older Macs: Intel.
Choose either download route.
All packagesInstall in terminal
Requires Node.js 22.19+ and npm.
npm install -g @geminilight/mindos@latest mindos onboard
Let your agent install it
Help me install MindOS from https://github.com/GeminiLight/MindOS with MCP and Skills. Use the English template.
Bring one note.
Complete a task. Keep one reviewed lesson.
FAQ
Before you begin.
Is MindOS free?
MindOS is MIT open source. Models and third-party services may charge separately.
Does co-evolution train the model?
It improves judgment and shared context. It does not change model weights.
Can I use existing notes?
Yes. Test with a small Markdown folder and back up important files first.
Will agents remember everything?
No. Set up the connection. Tools, instructions and the task determine what is read.
Does all data stay local?
Files stay local. Cloud models or external services may receive selected context.