Reality infrastructure

Reality infrastructure for autonomous systems.

Know exactly what your agent relied on, where it came from, whether it is still fresh, and what changed. AIality is the horizontal Reality layer — the product built on top for AI-commerce merchants is Agentomics.

Evidence topology — illustrative example
S2 NOT EXTRACTED
S1 MATERIALIZED
S3 ABSTAINED
Reality Resolved
Live Reality proof Loading

Live Reality proof is not connected in this deployment. This card is wired to the real, unauthenticated GET /receipts/{publication_record_id} surface — it does not fall back to a hardcoded value. When a receipts endpoint is deployed and configured, this card will show the real current published state for AIality's one production-grounded fact today: Anthropic's data retention maximum-duration policy.

Read the Resolve Reality contract →

The problem

Agents act on a version of the world that nobody versioned.

An agent that reasons about a vendor's policy, a regulation, or a partner's terms is reasoning about something that changes — silently, on someone else's schedule, with no durable record of what it said when the agent last checked. Re-scraping on demand answers "what does this say right now"; it cannot answer "what did this say when my agent decided," or "has anything changed since."

How AIality works

Sources become versioned, provable Reality.

Sources(vendor policies, docs, APIs)
AIality Reality Core(capture · verify · classify · lock)
Versioned Reality(immutable FactStateVersions)
REST / MCP Gateway(authenticated)
Agents / Systems(resolve · lock · act)

AIality continuously observes real source surfaces, turns evidence into an immutable, versioned Reality, and serves exact states — with provenance, freshness and authority attached — over REST and MCP. The hero above shows the shape of it: many governed evidence paths, one resolved Reality.

The Reality trust stack

Every exact state carries its own chain of custody.

  1. 01
    Source
    A concrete, audited surface — a vendor's published policy page, API, or document.
  2. 02
    Observation
    Evidence is captured and stored immutably, ontology-agnostic enough to reprocess later.
  3. 03
    Verification
    Extraction proposes a CandidateFact. An LLM alone can never assign VERIFIED or high assurance.
  4. 04
    Immutable Reality
    A new FactStateVersion is appended — history is never rewritten, only extended.
  5. 05
    Authority
    An explicit, versioned binding of this state to the source surface that authorizes it.
  6. 06
    Freshness
    A frozen-at-creation horizon determines whether a state now reads FRESH or STALE.
  7. 07
    Change
    A deterministic classification of what changed between two exact states, and how.
  8. 08
    Impact
    Registered dependents are attributed against the change that affects them.
  9. 09
    Reality Lock
    An immutable decision basis an agent can act on — and later re-check.
Agent integration

Built for agents that need to act, then re-check.

An agent resolves an exact Reality state, checks its freshness and authority, and can lock that basis before acting. Later, it can ask whether the basis it locked has changed — the same question a careful human reviewer would ask.

Sources
AIality Reality Core
Versioned Reality
REST / MCP
Agents / Systems
Implemented today
  • REST Gateway — /v1/reality, /v1/states/{id}, freshness, change, impact, locks
  • MCP Gateway — resolve_reality, get_freshness, get_authority, create_reality_lock, assess_reality_lock
  • Python SDK 0.1.0 — pilot release, not yet on PyPI (docs)
  • Bearer API key authentication on every integration route
MCP quickstart →
Autonomous maintenance

Reality that maintains itself.

An approved Reality state carries a finite Freshness horizon. When it expires, revalidation is obligatory — new evidence is captured, verified, and appended as a new immutable state, without an operator re-running anything by hand.

  • Reality maintenance LIVE
  • Finite Freshness ACTIVE
  • Autonomous revalidation LIVE
  • Restart recovery PROVEN
  • Manual bootstrap COMPLETE / ONE-SHOT
  • Natural 7-day autonomy cycle PENDING OBSERVATION
Real Reality history
  1. Historical FSV
    fresh_until = NULL — captured before finite-Freshness activation
  2. Finite-Freshness FSV
    value 30 · FRESH · 7-day horizon
  3. Next revalidation
    Natural 7-day cycle — pending observation

The bootstrap that moved Anthropic's retention Reality onto a finite horizon was a one-time, manual, operator-invoked step — by design, not a workaround. Every state created from here is discovered and revalidated autonomously. What hasn't happened yet: the first full natural 7-day expiry-to-revalidation cycle running end to end with no manual step at all — that's the one line above still marked pending, and it will flip the moment it's observed.

Built on AIality

Agentomics is AIality's first reference client.

Agentomics is an independent product for AI-commerce merchants, built separately from this codebase. It is not part of AIality, and AIality does not implement commerce logic — Agentomics owns that layer entirely. What it demonstrates is the target integration shape every AIality developer gets: submit and resolve exact Reality through one canonical contract, never a private database link.

Agentomics
Canonical AIality contract
Exact Reality · Freshness · Authority
Reality Lock · Impact
Integration status
  • Today Evidence ingress only — Agentomics submits observations through a private, narrow contract (ADR-017/018/019), not the public Gateway.
  • Planned Agentomics resolving Reality, Freshness, Authority and Reality Locks through the same public REST/MCP contract documented here.

AIality does not grant Agentomics — or any integrator — privileged database access. The target is one shared contract for every consumer.

Use cases

Where a versioned Reality layer changes the outcome.

Agents acting on provider policy

Problem
An agent depends on a vendor's data-retention or usage policy, but has no reliable way to know if what it 'knows' is current.
What the agent needs
An exact, versioned answer to "what does this source currently say" — with provenance and freshness.
What AIality supplies
Resolve Reality returns an exact FactStateVersion with Authority and Freshness attached, not a re-scraped guess.
What changes after AIality
The agent can lock its decision basis and later check whether that basis has changed before acting on stale grounds.

Compliance / governance automation

Problem
Governance teams need an auditable record of what a vendor's terms said at the moment a decision was made — not what they say today.
What the agent needs
Immutable historical state, not a live re-fetch that silently reflects today's page.
What AIality supplies
Publication history is immutable; a receipt references the publication delivered at that time, never "latest".
What changes after AIality
Audits point to an exact, unrepointable historical record instead of a screenshot or a re-crawl.

Vendor / procurement intelligence

Problem
Procurement and legal teams re-read vendor policy pages manually to catch material changes.
What the agent needs
A deterministic signal for whether a source materially changed, and along which dimension.
What AIality supplies
Reality Change classification (value / status / assurance / authority changed) instead of manual diffing.
What changes after AIality
Review effort concentrates on real semantic changes, not on re-reading unchanged pages.

Enterprise AI orchestration

Problem
Multi-agent systems share assumptions about the outside world with no shared, versioned source of truth for those assumptions.
What the agent needs
A common, queryable Reality layer every agent in the system can resolve against.
What AIality supplies
REST and MCP access to the same exact states, so every agent resolves the same Reality the same way.
What changes after AIality
Cross-agent disagreement caused by stale or divergent world-assumptions becomes visible and attributable.
AIality Index

A public, evidence-based record of AI vendor policy change.

The Index is designed around observable changes, not an opaque composite score. It is coverage-gated: we do not publish a comparative index at one vendor. See the methodology and current coverage status.

This is AIality's own horizontal Reality/Change Index — general AI-vendor policy change, not tied to any single product built on AIality. A separate, commerce-specific index is a possible future Agentomics asset; it is not this one, and the two will never be merged into one score.

View Index status →
For teams building directly on AIality

Talk to us about your Reality dependencies.

A 30-day infrastructure pilot: bring 5–20 critical external dependencies your systems rely on, and see them as continuously observed, versioned, provable Reality. This is for teams integrating directly against the REST/MCP contract — merchants looking for an AI-commerce product should look at Agentomics instead.

Request a pilot →