Skip to content

PointSav Documentation

The engineering library for the PointSav platform — operating systems and services for regulated businesses that own their data, their AI, and their record-keeping outright. Where the monorepo holds the code, this wiki holds the reasoning: architecture, services, security, and the governance commitments that bind future development.

PointSav Digital Systems

Revision as of · cb3da75f

Historical revision — this record as it stood on 5 September 2026, not the current version. View the current record →

Zero-container inference

Zero-container inference is the deployment pattern for the platform's Tier B GPU compute: native Linux binaries under systemd on GCE virtual machine instances, with no container runtime or orchestrator.

The economics close because idle detection ensures GPU billing stops when inference is not running. The Tier B inference pool that embodies this pattern has real, deployed infrastructure code; it is not a from-scratch build.

Why no containers

OCI container images imply a container registry: the registry becomes the durable artefact, and the operator must maintain image build chains, registry credentials, and CVE remediation for base images. For a one-shot inference VM that boots, runs for 30 minutes, and stops, the container layer adds operator surface without solving any problem the virtual machine approach does not address more directly. The single-binary, systemd-supervised approach is consistent with the Zero-container runtime structural commitment that governs all platform service deployments.

What is used instead

A native binary in the slm-yoyo GCE image family. A systemd unit supervises the binary; both a llama.cpp service and a vLLM service ship in the image, and which one serves live Tier B traffic is not yet settled (see Operational artefacts, below). OpenTofu handles VM provisioning and lifecycle management. Model weights are cached in Cloud Storage, so the cold-start path fetches locally rather than downloading from the upstream registry on each boot. nginx handles TLS termination, with the firewall restricted to port 9443. CUDA drivers are baked into the GCE image at build time.

Authentication does not use Secret Manager. The mechanism is a static bearer token passed via GCE instance metadata.

SMB economics

The GPU is an nvidia-l4 on a g2-standard-4 instance, running as a preemptible/spot instance. The economics close because idle detection is the load-bearing primitive: the instance stays up only while inference is running, not for operator convenience.

How idle detection works. A background task inside the Doorman server process polls the instance's health metrics every 5 minutes and deletes the instance — a real deletion, not a stop — once it has been idle past SLM_YOYO_IDLE_MINUTES (default 30). A separate, VM-local systemd unit is a dead-man's switch that powers the instance off at a metadata-set maximum lifetime — a different mechanism, for a different failure mode (a runaway or orphaned instance), not the routine idle-shutdown path.

Cold start

Startup budgets configured on the systemd units run in the minutes, not under two — cold start from a stopped instance to inference-ready takes several minutes, not seconds. For latency-critical workloads where a fast response is required, the deployment should extend SLM_YOYO_IDLE_MINUTES to keep the instance warm rather than assume a fast cold start. For nightly batch workloads — the primary use case for continued pretraining and large-scale corpus extraction — the cold-start cost is the price of zero idle cost and is a reasonable trade.

Operational artefacts

The deployment stack for a Tier B inference instance has four pieces. An OpenTofu module handles instance lifecycle management. The GCE image ships CUDA drivers, nginx, systemd units, and both a llama.cpp and a vLLM service — which one serves live traffic is not yet settled platform-wide. A bearer token in GCE instance metadata handles authentication. Cloud Logging points to the customer's own GCP project.

Defence-in-depth against runaway spend — a Cloud Billing budget with a kill-switch — is not yet built. The operator never interacts with the instance directly during inference; the systemd units and the Doorman-side idle monitor handle the lifecycle autonomously.

What this rules out

Managed container orchestration platforms, container runtime systems, multi-cloud abstraction frameworks, OCI image registries, layered container image build caching, and container build pipelines. These categories are not excluded because they are inferior in general; they are excluded because they introduce operator surface that is inconsistent with the Zero-container runtime structural commitment and the SMB economics case described above.

See also

Important Information

Corporate structure. PointSav Digital Systems ("PointSav") is currently a trade name of Woodfine Capital Projects Inc. ("Woodfine"), planned to become a wholly-owned Woodfine subsidiary upon incorporation. PointSav does not itself offer, sell, or solicit any security. Any securities offering associated with Woodfine's real-property direct-hold solutions is made exclusively by Woodfine, and only by means of the applicable Private Placement Memorandum.

No investment advice. This wiki's content is provided for engineering, operational, research, and development purposes. Nothing on this wiki constitutes investment advice or a solicitation to invest in any Woodfine partnership or direct-hold solution.

Intellectual property. The PointSav name, trade name, wordmark, and marks, together with all current and future PointSav- and Totebox-branded products, services, and offerings — and the software, source code, documentation, design system, and all related materials — are proprietary to Woodfine and its affiliates, except for components identified as open source. No rights are granted except as expressly set out in a written license or agreement. The full trademark notice appears in the footer of every page on this site.

Open source components. Portions of the platform are made available under permissive open-source licenses identified in the accompanying repository. Use of those components is governed by their respective license terms.

No warranty; informational use. Content on this wiki is provided for general informational purposes only and does not constitute a representation, warranty, or commitment with respect to product functionality, availability, pricing, or roadmap. Some articles describe planned or intended features, capabilities, and milestones — language such as "planned," "intended," "targeted," "may," and "expected" marks this forward-looking content, which is subject to change and does not constitute a commitment regarding future performance.

Confidentiality. Where an article describes an operational or deployment detail that is not intended for public disclosure, that article is not published on this wiki. Content here is general-purpose engineering documentation, not customer-specific configuration.

Jurisdiction. Woodfine Capital Projects Inc. is organized in British Columbia, Canada. References to the Sovereign Data Foundation on this wiki describe a planned or intended initiative only, not a current equity holder or active governance body.

Changes to this notice. PointSav may update this notice from time to time; the version posted on this page governs.

Not a filing system. This wiki is not a securities filing system, an electronic disclosure repository, or a substitute for SEDAR+ or any other regulatory filing system. Formal securities filings are made through the applicable regulatory filing system, not through this wiki.

Full disclaimer. This notice supplements, and does not replace, the full Disclaimers article. In the event of any conflict, the full Disclaimers article governs.

Read the full disclaimer →