Philotas installs on the platform you already run, in your cloud, on-premises, or air-gapped.
Philotas is not a per-seat subscription to a vendor's data plane. It is installed on your infrastructure, typically in your own cloud account, and delivered with an implementation plan.
The plan starts with a fit assessment that maps your data: what is open, what is licensed, what you already subscribe to. A proof of value runs on your data. Deployment and integration connect your sources to the picture, build the workflows unique to how you operate, train the operators, and take it live. After go-live, a new source or a new region is configuration.
It sits on top of the lakehouse you already paid for. A connector is one file, against an API or a lake query, and from there it flows into the cache, the snapshot store, the ontology and the knowledge graph without further wiring. For networks without internet access, the Enterprise tier packages Philotas for air-gapped deployment.
The data platform's identity and permission plane is the source of truth. Philotas adopts it automatically, so a deployment typically needs no separate security model; it uses the one you already operate. Where the lake does not already enforce access, Philotas adds its own layer, enforced server-side in SQL: roles from viewer to administrator, classification labels gating visibility, and workspace-scoped access to the corpus.
Row and column-level security across the operating picture is delivered where a deployment requires it beyond what the lake provides. The audit trail records authentication, operator actions, rule changes and workspace access. The adjudicating model runs against an OpenAI-compatible endpoint, which can be one on your own network.
Palantir is the benchmark for fusing data into an operational picture, and worth being measured against on the points that decide procurements.
| Palantir | Philotas | |
|---|---|---|
| The data model | Hand-built by forward-deployed engineers over months | Induced from the sources, with each proposed link carrying a confidence score and an audit trail for a person to confirm |
| Your platform | A closed data plane you buy into | Sits on the lakehouse you already own. A connector is one file, against an API or a lake query |
| Time to value | Consultant-heavy deployment programmes | A bounded implementation plan on the platform you own, then a new source or city is configuration |
| Engagement | Enterprise licence plus services, consultant-heavy | Implementation plan on your infrastructure, then a managed or private deployment |
| Model hosting | Agents over a human-built ontology | The model that adjudicates the ontology can run on your own network, on your own hardware |
| Open sources | Investigation tooling over the data you bring it | Reads the public record around every entity continuously, and surfaces connections that exist in no database you own, each traceable to the documents that produced it |
| Model governance | Not separately described | Generated text is checked before it is shown: deterministic rules, then an isolated review pass that never sees the source material. Speculation is stored apart from findings and labelled |
| Cost | Enterprise licence plus services | A fraction of it, deployed privately or managed |
To discuss a deployment, email [email protected]. See pricing for the tiers.
