PanSphaira documentation

Use this hub to find the right public document without treating roadmap text as shipped evidence. PanSphaira's current product category is an open, knowledge-driven operating system for governed, adaptable AI ecosystems. Its current shipped maturity is a released, open-source local PoC for governed and verifiable AI-agent actions across business systems; broad live-system and production realization remains planned and unproven.

The current architecture is Agent Sphere → governed Connections and Crossings → Gateway Sphere. The Agent is an untrusted proposer; the existing Runtime Isolation Boundary and Gateway mediate typed context, capability, Policy and Approval; stable capability contracts, Governed Templates, typed Adapters and Provider Bindings connect to providers. Sphere is terminology and visualization only, not a protocol, schema, API or runtime abstraction. Only the declared local-synthetic paths are currently evidenced.

Status labels used here:

• Released bytes — bytes are in the current regular release; inclusion does not activate or promote a capability, binding, runtime or product claim.
• Locally validated — reproducible local evidence exists, but it does not establish live-system or production fitness.
• Candidate / inactive — included or locally validated material is not an active route or promoted product capability without separate authorization and its required evidence.
• Planned — design or roadmap only; not executable product evidence.
• External evidence required — a claim needs a real provider, tenant, environment or independent operation that this repository cannot prove.

Start

• Released: Quickstart for the synthetic CRM → ERP PoC.
• Released: CRM → ERP approval and readback for the bounded intent-to-receipt walkthrough.
• Released: Root README for product orientation and the SAFE_GUIDED ten-second flow.
• Released: Release governance for Latest, manifests, hashes and anonymous readback.
• Contributing, support, and the private vulnerability route.
• When to use an alternative and the curated Now / Next / Later view.

Understand

• Canon: normative laws for agency, authority, effects and evidence.
• Architecture: current local reference design and trust boundaries.
• PanSphaira terminology and identity guardrails: display/Sphere vocabulary, historical boundaries, and stable-ID decisions.
• Operating Field Guide: profiles, adapters and evidence procedures.
• Agent runtime isolation contract: engineering contract for untrusted runtime crossings.
• OpenClaw bounded runtime and state contract: default-off scratch, managed mind-store, reset/recovery and rollback semantics.
• System Advisor Guide: reusable knowledge format.
• Governed Knowledge Harvest: provenance, applicability, promotion, invalidation and supersession lifecycle.

Verify

• Security Assurance: scoped claims, maturity, TCB, evidence and non-claims.
• SAFE_GUIDED secure-default proof: human claim matrix, closed machine manifest and one deterministic proof command.
• Known Limitations: explicit gaps and external gates.
• Supply-chain verification: offline declaration checks and their limits.
• Company-data validation: canonical synthetic data constraints.
• release/governance.json: machine-readable release and component evidence bindings.

Extend

• Released bytes / locally validated candidate / inactive: the finite inactive capability/action catalogue binds strict synthetic action contracts to exact versions and digests and exercises fail-closed Gateway/broker admission. Presence, installation, discovery, listing, local validation, byte inclusion and release do not activate an entry; activation is a separate exact maintainer authorization.
• Released / locally validated: Builder operator guide and Builder defaults cover typed, target-neutral contracts and two synthetic systems. No live-system builder or production UI is claimed.
• Planned: Connect your first system is a governed blueprint beyond the bundled demo; steps marked planned are not current executable instructions.
• Released / locally validated: HMI/harness contracts provide authority-free generation, discovery and explanation surfaces. The published contribution preflight prepares canonical, digest-bound local-synthetic bytes; it performs no submission, publication, external write, credential use or runtime activation. A production UI is absent.
• Released bytes / locally validated / inactive: Skill Bundle canonical contracts define strict manifest, lock, exact-file verification and compatibility identities for generation, analysis, installation and rollback consumers. Their inclusion does not install, activate, promote, publish, auto-update, grant capabilities or prove live registry/runtime safety.
• Released contract / external evidence required: the Entra profile and all five Power Platform read operations use exactly cm.discovery.read. cm.operator.read remains reserved for a future separate administrative-read profile. Live consent, import, tenant policy, provider compatibility and production identity behavior remain unproven.
• Released / locally validated: Extension assurance profiles compare closed synthetic evidence, require eight universal hard-fail gates, expire stale evidence, and route security-shaped findings privately. They do not scan, accept, install, activate, certify, or grant authority to an extension or connector.
• Released / locally validated: the minimized agent-work event contract classifies pseudonymous, consented, digest-bound synthetic records and their finite retention/deletion lifecycle. It adds no collector, telemetry, training feed, dashboard, ingestion path or runtime authority.
• Released contract / optional external subsystem: External BI service v2 documents CM's fail-closed, default-off client for separately released SBA v0.8.0 / contract 2.0.0. SBA alone owns BI logic and runtime; CM has no direct Superset/database route, credential, SQL, raw-row or mutation surface.
• Planned: generic live connectors, production reversible-write onboarding and broader resource-plane profiles.

Limitations and status

Capability	Status	Evidence boundary
Governed synthetic CRM → ERP effect	Released / locally validated	One pinned loopback demo with fictional data, brokered execution, readback and receipt
Verification Fabric	Released / locally validated	Deterministic contract and negative fixtures; no production deployment proof
Update/Migration/Doctor	Released / locally validated	Immutable six-axis locks, compatibility and read-only preview contracts; no discovery, repair, migration or update application
Builder contracts	Released / locally validated	Typed synthetic reuse and inactive target-neutral planning; no live-system builder or production UI
HMI/harness and contribution preflight	Released / locally validated	Authority-free interface and preparation contracts; no submission, publication or external write
Entra identity	Released / external evidence required	Closed cm.discovery.read identity contract; no live tenant, registration, consent or production identity proof
Power Platform five-read connector	Released / external evidence required	Five closed cm.discovery.read operations; no import, live Gateway/tenant, DLP or certification proof; cm.operator.read is future separate admin scope
Extension assurance profiles	Released / locally validated	Closed synthetic contract and eight hard-fail gates; no third-party scan, badge, acceptance, activation, authority, certification or production proof
Minimized agent-work event contract	Released / locally validated	Closed synthetic consent/retention/deletion record only; no collection, telemetry, training, ingestion or production proof
External KaleidoSphere service boundary	Released contract / optional external subsystem	CM-side configuration and readback contract only; BI runtime remains independently released and operated
Arbitrary-system onboarding and writes	Planned	Blueprint only beyond the bundled synthetic path
Knowledge-driven Operating System	Current product category; broad live realization planned	Released local-synthetic foundation only; arbitrary live adaptation and production maturity remain unproven

The root README owns orientation. Release Notes own increment detail, issues, pull requests, assets and test summaries. Security Assurance owns security claims and non-claims. Architecture owns trust boundaries. Guides own procedures. Known Limitations owns gaps; roadmap issues own planned work.

Visibility maintenance is documented in the discoverability baseline and channel/automation matrix. Public-Truth, security-boundary and broken-primary-action failures are release blockers; broad layout and discoverability findings remain review warnings unless they create one of those failures. Follow regular increments through the Releases Atom feed.