Orient
See the first commitment, active exceptions, and the weather window before opening three separate apps.
Runs locally · bounded by design
Your information is not missing. It is scattered.
Recompose runs on a device you control. It reads bounded fields from sources you already trust—calendar, weather, notes, household systems—and shapes them around one decision horizon.
Early implementation prototype. This public experience is a sanitized static recreation with fictional data.
01 / The idea
Stop building another inbox.
Compose a point of view.
Most personal software stores information by source. Your day does not arrive by source. It arrives as a question: What needs attention now?
Recompose leaves authority where it belongs, reads only bounded evidence, and refreshes a persistent route for the task at hand. The presentation can change; the source records remain authoritative.
How it becomes yours
Name the horizon. Bind only what it needs.
The public demo starts at step four with a prebuilt Today horizon. In practice, you define the question and time boundary first; Recompose then binds read-only adapters and keeps their limits visible.
-
1
Choose a horizonWhat decision · for whom · until when
-
2
Bind existing sourcesRead-only adapters · selected fields · no omnibus copy
-
3
Compose within contractsFreshness · provenance · explicit failure rules
-
4
Use, inspect, expireThe route persists; its evidence refreshes or fails closed
Authority remains upstream. Raw records are not copied into a central Recompose store.
02 / In practice
A surface that changes with the moment.
Not a feed to maintain. A bounded route whose contents recompose as the clock, evidence, and decision change.
Reassess
The forecast has shifted. Recompose marks the source freshness and changes the recommendation without rewriting the source.
Look ahead
Tomorrow's constraints replace today's noise. The interface recomposes; the underlying systems remain independent.
03 / AI + data fabric
Grounded synthesis, not corpus roulette
The model reasons.
The wiki remembers.
Recompose sets the boundary.
LLM Wiki compiles source material into durable, interlinked pages with provenance, confidence, and contradiction markers. Recompose turns a decision question into a small retrieval contract, then gives the model only that evidence packet.
If a required page is absent, stale, or contested, the answer exposes the gap instead of asking the model to improvise around it.
Run the fictional AI demo72-second AI walkthrough
Watch the knowledge fabric become a cited decision surface.
Fictional wiki pages, an auditable contract fixture, and an illustrative synthesis that preserves missing or contested evidence.
04 / Hermes-operated end-to-end demo
Five minutes.
From question to accountable action.
Watch a real Hermes session invoke a read-only Recompose bridge, validate a fictional bounded packet, run an actual model synthesis, verify its citations, and skip inference when required evidence disappears.
05 / Pointed use cases
Same fabric. Different decision horizons.
The concept gets clearer when it has a job to do.
These fixtures exercise the decision loop: durable Wiki-shaped pages, bounded composition, cited questions, cards that reshape on demand, and visible refusal when required evidence becomes unsafe. Scheduled capture and Wiki filing are shown as explicit production contracts, not actions performed by the static demo.
01 · Career match loop
A resume becomes a daily opportunity decision.
Compare a Wiki-shaped resume with fresh job-search pages, rank the bounded matches, inspect how next actions would be filed, ask the selected match a question, or turn its card from evidence to gaps without touching the underlying knowledge.
02 · Living briefing room
A presentation stays attached to the evidence.
Compose an executive decision deck from four Wiki pages, recompose the entire page into a one-page brief or delivery war room, then ask the current page what blocks approval or reframe only the selected card for customer or technical audiences. When outcome evidence expires, every format withholds the dependent content.
06 / Trust model
A useful interface can still be honest.
If the evidence is unsafe, the answer is unavailable.
Read-only first
Recompose consumes existing sources without silently becoming the source of truth.
Bounded evidence
Every adapter has explicit horizons, limits, provenance, and freshness rules.
Independent failure
One stale domain degrades visibly instead of contaminating the entire surface.
Earned adoption
New horizons run beside existing tools until their usefulness and failure behavior are proven.
Recompose is
- A composition runtime on a device you control
- A contract between sources and surfaces
- A way to build five-second orientation views
Cloud services may remain upstream sources. Adapters request only contract-bounded fields; the composition and its state stay local unless you deliberately publish a surface.
Recompose is not
- Another universal database
- An AI agent with unrestricted access
- A replacement-everything migration
The practical question