§ 01 — Platform
From building production distributed systems and cloud-native infrastructure, we’re now driving harness engineering to help enterprises unlock new AI capabilities.
The harness is everything except the model: eval suites, CI gates, drift detectors, compliance artifacts, operational telemetry. We build it. We run it. We keep it current when any model ships a new version. Eval Cloud live. Different substrate. Same discipline. Hadoop. Kubernetes. Now AI. Published OSS at github.com/VISystems.
Every decade, a new substrate becomes the next decade of enterprise infrastructure. VISystems has shipped at every inflection since 2007. The technology changes. The engineering discipline — eval-first, telemetered, audit-trailed, open-source — does not.
2007–2015
Built production Apache Big Data platforms when the stack was still in incubation. Contributed upstream, ran the infrastructure at F1000 scale, and learned a discipline: if you can’t evaluate it, you can’t ship it.
Built with: Apache Hadoop, Spark, Cassandra, HBase, ZooKeeper
2016–2023
Designed multi-cluster Kubernetes platforms, service-mesh topologies, and GitOps CI/CD for regulated financial services and large ISVs. Shipped internal tooling that became the engineering playbook for the next substrate.
Built with: Kubernetes, Istio, Argo, Flux, Prometheus, OPA
2024 → present
Current era. Building the platform that keeps AI models reliable in production — eval execution, migration safety, drift detection, cost intelligence, compliance evidence. 7 OSS repos published. Works with Anthropic, OpenAI, Google, AWS Bedrock, and Azure OpenAI.
Shipping: Eval Cloud (SaaS), eval-runner, eval-as-ci-action, model-migration-kit + 4 more OSS repos
We work with enterprises whose compliance and procurement constraints mean we can’t name them here. The list below is anonymized to the level their legal teams are comfortable with.
▌ Why no logos
“We don’t publish customer logos. Their procurement team prefers it that way.”
§ 04Platform capabilities
Eval Cloud covers the full AI model lifecycle — from wiring up your first eval suite to producing the audit evidence that satisfies a model risk officer. Five core capabilities, four operational dimensions, multi-provider from the start.
§04.1
Eval suites, context architecture, tools, MCP servers, agents. Built test-first against a published spec, version-controlled in your repos, harness-engineered from day one.
§04.2
Production traces, observability, drift detection, capacity planning, eval-as-CI gates. The day-two work that determines whether agents stay in production a year from now.
§04.3
Automated migration safety on every model release. Audit ledger by default. SR 11-7 alignment documentation. The work that survives a model risk officer's review.
§04.4
Automated reporting. Cross-provider benchmarking. Your reliability posture improves every quarter — the platform surfaces what to optimize next.
§ 05Tiers — platform plans
Every tier ships every product feature — SSO, SCIM, the eval harness, MCP registry, audit ledger, all CI integrations, full data export. Tiers differ on eval volume, retention, and isolation — never on whether you can do something. No SSO tax. No SCIM tax. No audit-ledger tax.
| Feature | §05.1 Free Get started | §05.2 Team Production scale | §05.3 Business Enterprise-ready · SR 11-7-aligned | §05.4 Enterprise Custom contracts · dedicated tenancy |
|---|---|---|---|---|
| Eval runs / month | 100 | 5,000 | Unlimited | Unlimited |
| Models monitored | 1 | 5 | Unlimited | Unlimited |
| Providers | 1 | 3 | All | All |
| Migration lead time | Standard | Priority | Next business day | Same day SLA |
| CS cadence | Quarterly | Quarterly + monthly | + named CSM | Dedicated CSM |
| Trace retention * | 12 mo | 24 mo | 84 months · SR 11-7 | 84 months · SR 11-7 |
| DB tenancy * | Shared | Shared | Dedicated Neon project | Dedicated · custom region |
| Specific capacity for the Enterprise tier is scoped on request. Early-access subscribers who join before public beta lock preferential pricing and free historical-telemetry migration at GA. | ||||
* Eval Cloud platform features — currently in private beta. Available on every tier at GA.
→ Full pricing detail at /pricing
§ 06Platform capabilities · five core modules
Eval Cloud ships every capability to every subscriber. Eval execution, drift detection, migration safety, cost intelligence, and compliance evidence are platform features — not add-ons, not tier-gated modules. You connect your providers, configure your gates, and the platform runs continuously.
BUILD · DEFINE SPECS, CONNECT PROVIDERS, GATE CI
RUN · REAL-TIME MODEL BEHAVIOR MONITORING
RELY · AUTO-REBASELINE ON EVERY MODEL RELEASE
RUN · CROSS-PROVIDER COST TRACKING AND OPTIMIZATION
RELY · AUDIT TRAILS, SR 11-7 DOCUMENTATION
→ Full platform documentation with coverage details and integration guides at /offerings
§ 07 Eval Cloud · the operational layer
▌ Built because we needed it · available on every tier
Every AI deployment needs evals, model-upgrade review, and evidence that survives audit. Eval Cloud turns eval and migration history into marketplace-procured model-risk evidence: run suites, detect provider releases, re-baseline behavior, flag regressions, and produce the audit trail regulated teams need.
Currently in PRIVATE BETA. Public beta Q3 2026. v1 GA target Q1 2027. Early-access subscribers receive preferential pricing and free historical-telemetry migration at GA cutover.
▌ Waiting list — open
Early-access subscribers who join before public beta lock preferential pricing and free historical-telemetry migration at GA cutover.
No marketing emails. We respond within 48 hours with early-access details and next steps.
§ 08 Three differentiators
We bet our brand on three load-bearing differentiators.
▌ Differentiator 01
Eval Cloud instruments token spend, latency cost-curves, and model substitution opportunities across Anthropic, OpenAI, Google, AWS Bedrock, and Azure OpenAI from the moment you connect. We publish the metrics — anonymised and aggregated — in the VISystems Impact Report so subscribers can benchmark their own cost trajectory against the cohort. If your cost-per-eval is drifting, the platform flags it before you notice it in billing.
→ VISystems Impact Report · published quarterly to subscribers
▌ Differentiator 02
7 Apache 2.0 repos, all built by the same engineers who build the platform. There is no separate OSS team — the engineers who build the platform write the OSS. The repos are the proof that the discipline we ship is the discipline we practice.
→ Q1 · eval-runner · live
▌ Differentiator 03
Published platform with documented capabilities. Every feature has documented coverage and integration guides. No ambiguity about what you get. Start free, upgrade when your workload grows, export your data anytime.
→ Self-serve · 5 CI integrations · data portability