Agents That
Make It Past Pilot
Agents stall in pilot when nobody can predict or check what they'll do. Give yours reviewed business rules to call over MCP, REST or Python, on data where it lives: same input, same output, every result traced to its rule and rows.
Proven in production
Tool Calls,
Not Guesswork.
An agent with raw table access has to infer your business logic on every run. An agent on Prometheux calls logic you've already approved.
Same Rules,
Every Agent
No new platform to move your data into, and no agent you're forced to adopt. Prometheux runs in your cloud, on-premises or inside Databricks and Snowflake, and serves your rules to the agents you already use over open routes.
Remote MCP Server
Cloud-hosted and signed in with OAuth. Works with Claude, Genie Code on Databricks and other MCP clients.
Local MCP Server
Install prometheux-mcp with pipx and add it to Claude Desktop.
REST API
Language-agnostic HTTP endpoints with bearer-token auth, for any service or orchestrator.
Python SDK
prometheux-chain lets your code save your rules and run concepts in a couple of calls.
Keep the assistants and models your teams already use. Own the data, the logic and the deployment. No walled garden to climb out of later.
The Agent Plans.
The Engine Runs.
The agent decides which question to ask; the reasoning engine answers it by running concepts over your live sources. Concepts in Vadalog, SQL, Cypher and Python are deterministic, including recursive logic that rolls up totals at every level, which would take an agent a long chain of tool calls. A workflow that worked yesterday works the same way today.
Repeatable Steps
Every concept call returns the same result for the same data, so you test a step once and can replay any agent run to check it.
Scheduled Runs
Schedule concepts to run automatically, so agents read fresh results instead of recomputing them.
Runs on Data Change
Re-run a concept when its source data changes, or when a Prometheus or Grafana alert fires.
Your Own Frontend
Build custom agent apps on your infrastructure, with auth, streaming and server-side tokens handled.
Automate multi-step work without losing control. Replay any agent run and get the same answer. Show the rule trail behind every action.
From Rules to
Working Agent
Model the Logic
Write the concepts your agents need: eligibility, exposure, risk flags, whatever the workflow decides.
Connect the Agent
Add the Prometheux MCP server to your assistant, or call the REST API or Python SDK.
Let It Call Concepts
The agent discovers available concepts and runs them against your live sources.
Act on Proven Results
Each answer comes back deterministic and traceable, ready for the next step.
Agents Where
Mistakes Cost
Explained Credit Decisions
Banca Sella's system accesses its databases, processes a client's full history and explains the outcome on its own.
Read the case study →Agents in Databricks
Connect Genie Code to the remote MCP server and let it run the concepts you've defined.
Scheduled Agent Pipelines
Run concepts on a schedule or when source data changes, so agents always read fresh results.
Custom Agent Apps
Ship your own assistant on your infrastructure, backed by the same business logic as your reports.
Integration docs →“Prometheux enabled us to build a system that can access our databases, process all the client's historical variables, and provide an intelligent, understandable explanation autonomously.”
Massimiliano MalinHead of Credit Risk · Banca Sella
“Generative AI without ontologies is an engine without a road. Prometheux builds the road.”
Marco RomeiDeputy CIO · Banca Sella
Questions,
Answered
Which agents can connect to Prometheux?
Any MCP-compatible client through the remote MCP server with OAuth, including Claude and Genie Code on Databricks. Claude Desktop can also use the local prometheux-mcp server. Anything else can use the REST API or Python SDK.
Do we have to move our data into a new platform?
No. Prometheux connects to your databases, warehouses and files where they are, and deploys in your cloud, on-premises or natively on Databricks and Snowflake. Agents call concepts that run there.
Are we locked in to your agent or your model?
No. Use Claude, Genie Code, your own agents or any MCP client, with whichever model you choose. Ontologies can be exported through the API, and rules are readable code you keep in version control.
What can an agent do over MCP?
Discover the concepts available in a project and run them against your connected sources.
Why not let the agent query the database directly?
It would have to infer your business logic on every run, differently each time. Calling concepts means the logic is written once, reviewed, deterministic and traceable.
How do we test an agent before it goes live?
Test the rules, not only the prompts. A concept that returns the right result on test data returns the same result on the same data in production, with its derivation attached. LLM steps are labelled non-deterministic, so you know which parts still need evaluation.
Can't our team build this with an LLM and some SQL?
You can build a pilot. The difference is whether the logic stays governed, portable and understandable after the original builders move on. In Prometheux it lives as versioned concepts every agent shares.
Related Capabilities
Put Your Agents on
Solid Ground
Bring a workflow you'd like an agent to run. We'll connect it to your business rules and show every step it takes.
