Context scatters
Decisions, references, drafts, and next steps get separated across tools and time.
What we are building
An adaptive workspace that helps complex projects return with their context.
Lumina is being built for work that cannot be captured by one file or one chat: art, research, software, teaching, design, writing, and long-running creative systems. The goal is simple to say and hard to build: when you come back, the work should come back with you.
The problem
Important projects rarely live in one clean place. They spread across documents, sketches, tabs, code, images, conversations, tools, unfinished decisions, and half-remembered next steps.
Return is expensive
People waste energy figuring out where they were before they can make real progress again.
AI help can reset
AI conversations can be useful, but the surrounding workspace often loses continuity unless the system preserves it.
The plain-language promise
Lumina should reduce the cost of returning.
Instead of asking you to reconstruct the whole project every time, Lumina is designed to restore the state of the work, show what changed, recommend likely next steps, and keep a record of what happened.
1. Return
Recover the project layout, known references, recent action, working stance, and likely tools.
2. Reflect
Review the current state before jumping into action, so the system notices ambiguity and context gaps.
3. Recommend
Suggest useful next steps from restored context while keeping the human in charge.
4. Govern and record
Use checkpoints, receipts, and boundaries so important actions are visible and reviewable.
What exists now
- Working runtime scaffold and local command path
- Project-return and workspace-host prototypes
- Dashboard and visible continuity surfaces
- Bounded self-guidance and reflection experiments
- Receipts, sea trials, and governance checks
Where this is going
- More usable local operator surfaces
- Clearer project restoration for non-developers
- Stronger dashboard visibility
- Eventually, an OS-shaped environment with files, tools, permissions, applications, and continuity living together
- Human-AI collaboration that keeps consent and boundaries visible
For builders: the deeper layer
The practical promise is supported by a governed runtime scaffold. This is the machinery that keeps the project from becoming only a beautiful idea.
Runtime spine
Coordinates session state, checkpoints, mode transitions, context bundles, governance logs, and capability exposure.
Project return
Connects the latest known project state to a bounded workspace surface: panels, references, tools, focus, and stance.
Diagnostics
Uses tests and signal-style metrics to notice drift, fading, feedback risk, and continuity recovery.
Self-guidance
Recommends likely next actions from restored context and advisory history without claiming authority.
Governance rails
Records important runtime events and preserves clear boundaries between suggestions, execution, and structural changes.
Sea trials
Stress-test restoration, guidance, symbolic independence, checkpoints, observation cycles, and runtime wiring.