ServicesIndex

PointSav's three-ring architecture assigns every service to a layer with defined authority and dependencies. Ring 1 services handle per-tenant boundary ingest — each accepts raw data from one external source and writes it to a durable ledger. Ring 2 services provide deterministic knowledge and processing: they read from Ring 1 and produce structured records, knowledge graphs, and search indexes. Ring 3 is a single service, service-slm, which reads from Ring 2 and never writes to it.

The platform functions fully across Rings 1 and 2 without AI compute — a deployment can exclude Ring 3 entirely, shrinking the attack surface and satisfying network-isolation requirements. Where Ring 3 is included, the compliance question of whether AI has touched the authoritative record is answered architecturally, not procedurally: Ring 2 services may call Ring 3 for extraction or classification proposals (service-extraction's corpus hand-off to service-content, which calls the Doorman for grammar-constrained entity extraction into the DataGraph, is one such path), but Ring 3 never writes to the knowledge graph, the ledger, or any structured record store. Every accepted proposal enters the record only through a Ring 2 write path with a human approval checkpoint.

Start here: service-fs — the WORM ledger backbone — the filesystem service every other Ring 1 service writes to, and the foundation of the WORM audit posture the rest of this category assumes.

Ring 1 — Boundary ingest

Per-tenant boundary services. Each runs as a separate process per tenant and exposes a Model Context Protocol server interface.

• service-fs — the WORM ledger backbone — The filesystem service: append-only WORM ledger, per-tenant storage root, the foundation every other Ring 1 service writes to — architecture, durability, and the SEC 17a-4(f)/eIDAS/SOC 2 compliance posture it enables by construction.
• Email ingest — Email ingest: SMTP and IMAP, sanitised payloads, append-only Maildir on local block storage.
• service-people — the identity ledger service — Identity ledger: person records, role assignments, and the Anchor-Claim-Socket data model that never overwrites state.
• service-input — reference-archive migration and calibration — Document intake at the Ring 1 boundary: parses PDF, Markdown, DOCX, and XLSX by format detection, normalizes to a ParsedDocument, and hands off to service-fs for WORM ledger commit.

Ring 2 — Knowledge and processing

Deterministic processing services. Each reads from Ring 1 and produces structured records — no AI variance enters the authoritative record.

• service-extraction — the DataGraph ingestion pipeline — The central Ring 2 traffic controller: strips proprietary formatting, constructs Entity Bundles, assigns transaction IDs, routes to deterministic services or to service-slm.
• service-content — entity extraction and knowledge-graph host — The Gravity Engine: reads raw payloads from a Totebox, runs them against an institutional taxonomy, generates the structured documents an organisation publishes.
• Full-text search — Full-text search on Tantivy: per-tenant sharding, microsecond retrieval, no active database process required.
• Egress service — Physical release valve: structured records leave the platform only through this service.
• Archetypes and chart of accounts — The institutional taxonomy: eleven archetypes and a Chart of Accounts that classify personnel and documents by structural position and functional role.

Ring 3 — AI gateway

One service spans Ring 3. It reads from Ring 2 and produces proposals a human reviews; it never writes to the knowledge graph or the ledger.

• AI inference service — The Doorman: AI routing across local, burst, and external compute tiers; audit ledger on every call; every API key held at this boundary.
• SLM and Yo-Yo operational state — Operational state of service-slm and the Yo-Yo GPU burst VM: Tier A/B configuration, apprenticeship brief queue, idle-shutdown cost ceiling.
• SLM as Totebox sysadmin — the plan — How service-slm becomes the operational assistant for Totebox deployments: ten operational task families, four-stage pipeline from corpus capture to per-tenant LoRA adapters.
• service-slm graph store rebuild — The DataGraph's live property graph: nightly LadybugDB rebuild via grammar-constrained entity extraction through the Doorman, gated by human approval before any write.
• Yo-Yo daily enrichment cycle — The Yo-Yo GPU burst VM's daily batch window: enriches the DataGraph and accumulates DPO training pairs under a fixed schedule and hard cost cap.

Specialist and domain services

Services built for specific platform capabilities.

• Business clustering — Turns raw retail data into commercial clusters: parent-child spatial schema, one commercial entity per site.
• Places filtering — Filters civic and institutional infrastructure to retain only regional-grade facilities for GIS tier rankings.
• Wallet settlement — the design — Wallet and direct payment settlement infrastructure.
• Message courier service — Headless web-automation engine bridging internal identity ledgers with external web portals.
• FS anchor emitter — Signed WORM ledger checkpoints at hourly cadence, anchored to Sigstore Rekor on a monthly schedule for external auditability.
• GIS data lake — Flat-file data lake for the GIS pipeline: raw geospatial points from open sources, no ETL step.
• Template ledger — Distributes approved email templates to the operator's mail environment; eliminates version drift between template design and execution.
• The proofreading pipeline's wire contract — Three-stage proofreading pipeline ordered by cost: deterministic banned-vocabulary scan, LanguageTool mechanical pass, then a generative rewrite routed through the inference layer.
• Private binary download endpoint for paying customers — The binary release server behind software.pointsav.com: verifies Ed25519 license tokens and streams compiled binaries, holding no payment records or signing keys.
• service-pointsav-link — planned PointSav fleet adapter — The hot-pluggable adapter connecting an os-* Subject node to a PointSav fleet: not installed by default, with a clean-severance failure mode.
• service-vm-fleet — the PPN VM fleet controller — The placement and registry service for the PPN VM resource pool: two-pass placement algorithm and heartbeat-driven node state.
• service-vm-tenant — the PPN VM tenant proxy — The customer-facing tenant proxy for the PPN VM resource pool: authentication, namespace isolation, quota enforcement, and an immutable audit trail.

See also

• Systems — the operating systems that services run within
• Architecture — the three-ring model and the invariants that govern ring interaction
• Infrastructure — fleet deployment and the physical layer services run on