Version: 1.2.0 Status: AUTHORITATIVE
XYPH is a causal computer: a time-travelable, decentralized work + runtime substrate built on WARP graphs — with Git used only as a settlement/object-store layer.
XYPH exists to solve the Agentic Coordination Problem and then outgrow it.
As we move from human-led projects to agent-driven autonomous workflows, we need a substrate that ensures:
- Deterministic Provenance (history is the source of truth)
- Human Sovereignty (laws, consent, and cryptographic authority)
- Speculative Collaboration (parallel worldlines without merge-conflict rituals)
XYPH rejects "state-first" thinking. State is a projection; history is the artifact.
- The system can time-travel data, files, and the runtime itself back to prior ticks.
- "Branches" are an implementation detail of legacy tooling; Shadow Working Sets (SWS) are the native unit of isolation.
- Counterfactuals are first-class: forks are kept, indexed, and queryable.
XYPH treats the roadmap not as a document, but as a program to be compiled.
- Source Code: Human intent, NL prompts, and formal specs.
- IR (Intermediate Representation): The WARP graph.
- Targets: Verified artifacts — code, docs, deployments, test receipts, audit receipts.
But XYPH also compiles execution itself: jobs, automation, orchestration, and review are all graph-native.
XYPH may use Git today, but Git is not "the product."
- Ontology: XYPH is sovereign. Worldlines, observation, comparison, collapse, lawful transformation, and authority are XYPH concepts.
- Settlement / Object Store (Today): Git CAS as a ubiquitous, battle-tested content-addressed store.
- Causal Substrate: WARP graph + multi-writer convergence.
- Runtime: JIT — Just-In-Time graph execution (SWS, promotion, collapse, receipts).
- Experience Layer: XYPH CLI/TUI and observer views (status, diff, slice, provenance).
Swapping the settlement layer is allowed long-term; the invariants are not. Alfred-derived components may help at the edges, but Alfred is not XYPH's public ontology.
XYPH adopts the Digital Guild model to govern humans and agents:
- Genealogy of Intent: Every
Questtraces to a human-signedIntent. - Ceremonies with receipts: State transitions are legal actions, not UI toggles.
- Guild Scrolls: Outputs are signed artifacts stored in the graph.
- Consensual Labor: Work is volunteered, not assigned.
XYPH replaces "Issues / PRs / Actions" with a single principle:
The graph is the shared environment; coordination emerges from what's written there.
Early scaffolding (temporary names):
- Submission / Patchset / Review: append-only proposed changes and feedback.
- Merge: a terminal act that emits a signed receipt and seals the quest.
- Automation: jobs executed from graph triggers, producing receipts (tests, builds, deploys).
Long-term: "PR" disappears into continuous, local, graph-native review and policy enforcement.
A world where:
- "Launching a product" is as deterministic as compiling a binary.
- You can seek the system to any prior tick, fork reality, and replay forward.
- Trust, tests, and deployment safety are cryptographically provable — not vibes.