Palantir proved that AI needs a typed model of the business to act on it. Their delivery path to one runs through forward-deployed engineers and six-figure contracts. Syntheka ships the same idea differently: a typed ontology your own agents reach over MCP, self-hosted, on your stack.
Palantir's documentation defines the Foundry Ontology in three layers: a semantic layer of objects and links, a kinetic layer of actions and functions, and a dynamic layer of models and simulations. Their architecture center positions it as the operational layer where humans and AI agents collaborate. Credit where due: this is the right abstraction. An agent reading raw tables guesses; an agent reading typed objects, relations, and actions reasons about your business the way your operators do.
The concept is documented publicly, in Palantir's own ontology overview and architecture-center materials. The idea is not proprietary. The delivery model around it is what keeps most companies away from it.
Third-party analysis of Palantir pricing puts commercial Foundry and AIP contracts at roughly $250,000 per year at entry, for a single narrow use case, and in the millions at enterprise scale. The reason is the delivery model: forward-deployed engineers embedded at your site for months, with discovery, integration, and delivery queued on the vendor's calendar and priced on the vendor's rate card.
The vendor staffs engineers on your premises. Your ontology takes shape on their timeline, extends through their bench, and lives in a platform you license by the contract. The FDE model earns its cost in complex, high-stakes industries: it produces real outcomes where the integration burden is genuinely heavy. But for a mid-market company, the entry ticket alone is the whole IT budget.
You model your objects, relations, and rules as typed definitions on the platform you self-host. Your own agent connects over MCP, discovers the tool catalog, and works the same semantic layer. No resident engineers. The judgment you build into the ontology is yours to keep.
Syntheka exposes its typed ontology (the semantic layer: objects and links) through an MCP server. Connect your agent, and it works the operational layer directly: reading typed aggregates, explaining rule decisions, staging actions that clear the approval DAG before they execute. The platform manual teaches your agent the surface; you keep the governance loop around it.
Your agent lists the tool catalog over MCP and learns what it can do without a workshop. Today the catalog holds five core tools and keeps growing.
Write-side work travels the same path as every agent action on the platform: staged as a proposal, evaluated against your rules, approved through the DAG, executed, audited.
Each agent connects through an MCP binding token, stored as a SHA-256 hash. Revocation takes effect immediately, and every binding event lands in the audit chain as event kind mcp_binding.
Five core tools ship on the MCP catalog now; the catalog is growing. Each name below is the live entry on GET /tools/catalog.
Ask why a policy evaluation came out the way it did. The agent gets the rule logic, not a shrug.
Group and aggregate typed ontology objects by domain, so the agent answers business questions instead of paging through rows.
Numeric profiling of a dataset on demand: the numbers your agent needs before it proposes anything.
Stage an action into the approval DAG. Nothing write-side executes without a human decision on the record.
The round-trip check your agent runs first, so a broken binding is a clean failure, not a mystery.
The surface extends with the platform. Tools land as typed, documented MCP entries, discoverable by the same catalog call.
Works with the agents your team already runs: Claude Code, Codex, Grok, and any other MCP-capable client.
The MCP tool catalog holds five core tools today and is expanding; it does not cover every domain yet. Complex industry domains are still served through a fixed-price pilot: you buy a scoped project, not a headcount with a bench.
And the direct comparison: Syntheka is not a replacement for all of Palantir. Foundry spans data ingestion, analytics, and operational applications at enterprise scale. What overlaps is the ontology layer that lets AI act on your business, and that layer is what the FDE model prices beyond most of the market. Syntheka is not a data warehouse replacement either; your data stays in your Postgres.
No. Foundry is a broad platform spanning data ingestion, analytics, and operational apps; Syntheka is not a data warehouse replacement. The overlap is one layer: the typed ontology that lets AI agents operate on a business as typed objects and governed actions, delivered self-hosted instead of through forward-deployed engagements.
Any MCP-capable client: Claude Code, Codex, Grok, and other agents your team already runs. The platform manual teaches your agent the tool surface; authentication is per-binding tokens.
Each agent connects through an MCP binding token. Tokens are stored as SHA-256 hashes, revocation takes effect immediately, and every binding event is recorded in the audit chain (event kind mcp_binding).
The catalog is growing, and complex industry domains are still covered through a fixed-price pilot: you buy a scoped project, not a headcount with a bench.
Early access starts with a self-hosted pilot: your Postgres, your model endpoints, your agent on the MCP server, one workflow that writes to a system of record. We'll be on the call, not in the loop.