Managed
CloudOpen waterHosted by PSYVAR in your chosen region. Fastest path to production.
- Regional residency
- SOC 2 controls
- Shared model pool
The control plane for every model your organisation runs.
Route, evaluate, observe and govern — across our models, open weights and any provider — from a single pane that your engineers, risk team and auditors can all read.
PSYVAR CoreV4
Workspace preview · synthetic data
- Requests
- [COUNT]
- P95 latency
- [MS]
- Eval pass rate
- [%]
- Spend
- [AMOUNT]
Route table
| ROUTE | PRIMARY | FALLBACK | POLICY |
|---|---|---|---|
| support/triage | psy-core-s | psy-core-l | policy_v9 |
| claims/extract | open-wt-8b | provider-a | policy_v9 |
| research/agent | psy-core-l | — | policy_hitl |
Four modules. Use one, or all of them.
Router
Declarative routes send each request to the cheapest model that meets your quality bar, with automatic fallback on degradation or outage.
Latency-aware · cost-capped · provider-neutralEvaluations
Run your own test sets on every deploy and on a schedule. Regressions block the rollout before anyone notices in production.
Custom suites · LLM + human grading · CI gatesTraces
Every prompt, tool call and decision is captured end to end. Replay any interaction, diff two model versions, export the record.
Replay · diff · retention policiesGuardrails
Policies as code: PII handling, topic limits, human-in-the-loop checkpoints and jailbreak detection, applied at the route level.
Policy-as-code · HITL · versioned
Runs where your data is allowed to be.
Same platform, same interface, three deployment models. Pricing on request.
Private
Your cloudYour harbourDeployed inside your AWS, Azure or GCP account. You hold the keys; we hold the pager.
- Your VPC, your IAM
- Dedicated inference
- Bring your own models
Air-gapped
On-premiseEnclosed basinNo outbound network required. For regulated, classified and sovereign environments.
- Offline updates
- Hardware sizing included
- On-site enablement