Trusted Agents for
the Public Sector
Agencies, tax authorities, central banks and regulators are asked to do more with less, and to explain every decision. Prometheux runs AI on the data you already hold and the policy you already approve, on your own infrastructure: same case, same answer, with the rule and records behind every result.
Proven in production
Pilots Everywhere.
Little in Service.
Every public body is being asked to use AI. Most get stuck at the same points: data split across departments, answers no one can defend, and legacy systems too risky to move. Prometheux works on the data and policy you already have.
Where Public Sector Teams
Start
Pick the area you run. Each use case runs as reviewed rules on your own data, with the record behind every result.
Citizen Services
Faster, consistent decisions on benefits, grants and licences, explained to the citizen, the caseworker and the appeal panel.
Tax & Revenue
Collect what is due with less effort: check returns against the data you already hold and focus officers where it matters.
Fraud, Error & Waste
Stop overpayments, fraud and waste earlier, with evidence an investigator can act on.
Supervision & Regulation
For central banks and regulators: validate returns faster, see where risk concentrates, and trace every finding.
The Questions Other
Tools Can't Answer
Linked Cases at Any Depth
Resolve the same person or company across registers, then follow shared addresses, phones or accounts hop after hop to surface linked cases, without a graph database. See graph traversal.
One Person Across Departments
Match records from tax, benefits and housing that differ in spelling or format, with the rule and evidence behind every match. See entity resolution.
Exposure Along a Chain
Recursion with aggregation totals exposure at every hop between counterparties, so risk that only appears across several firms is visible. See recursion.
The Record Behind a Decision
Trace any eligibility, tax or supervisory result back through the rules that fired to the source rows, fact by fact. See data lineage.
One Rule Layer Across
Every Department
Typical government data platforms copy departmental data into one central store before anyone can use it. Prometheux sits between your registers and the officials, services and agents that act on them, and runs the policy where the data already lives.
Analytics
Case dashboards, graph and BI
Integrations
Back into the databases behind case, tax and supervisory systems
Agents
Caseworker agents, deterministic answers
Models
Define the ontology, read answers, and power analytics, integrations and agents
Your registers and returns
Registers, case systems, tax records, supervisory returns and mainframe extracts, connected in place. Nothing to migrate.
-
Public Services and Oversight
- Eligibility, tax, fraud-and-error and supervisory checks on one set of reviewed rules
- Agents call approved rules over MCP and API, with the rule and records behind every answer
- Results written back to the databases behind case, tax and supervisory systems, on a schedule or when data changes
-
The Policy Rules Engine
- One shared model of people, businesses, cases and schemes across departments
- Recursion with aggregation and graph algorithms: linked cases at any depth, exposure along a chain, central parties in a network
- Same input, same output, with lineage to the rules and source rows
-
Cloud Agnostic
- Run on your own servers, private cloud or any cloud, under your own controls
- Scale horizontally on clusters as registers and returns grow, with optional GPU acceleration
- 1-click deployment anywhere: EKS, AKS, GKE, Databricks Clusters, Snowpark Containers
-
Register Agnostic
- Read and write registers, case files, returns and ledgers in any database, format and schema
- Connect warehouses, legacy databases, mainframe extracts, files and APIs across clouds and on-premises
- No new central store to migrate departmental data into and keep in sync
Same Layers.
Better Built.
Map a typical government data and analytics stack to Prometheux and the difference is where the work happens: in place, in rules your officials own, with an answer you can defend.
Your Servers.
Your Rules.
Install the whole platform inside your own network, keep data under the controls you already run, and give auditors a record of which rule produced which result from which data.
Sovereign Deployment
Run on your own Linux servers or Kubernetes, in your private cloud or on-premises, under your change control. See on-premises.
Lineage Per Decision
Provenance links every result to the rule that fired and the source records it used, ready for appeal, audit or inquiry.
Policy You Can Read
Rules are plain text in your version control. When policy changes, you change one rule and every service follows.
Your Access Controls
Data is read in place under existing permissions, and credentials stay in saved connections. See AI governance.
Answer an appeal with the rule and the record. Keep citizen data inside your own walls. Change policy once, apply it everywhere.
One Decision.
Then the Next.
Connect in Place
Point Prometheux at the registers, case systems, tax records or returns behind one decision. Nothing is copied out. See in-place data processing.
Codify the Policy
Your policy and legal teams, our forward deployed engineers and the AI assistant write the criteria, thresholds and checks as rules officials can read, review and version.
Run One Use Case
An eligibility check, a return validation or a fraud-and-error review, run on your data for caseworkers, analysts and agents alike, with lineage on every result. See data lineage.
Expand Service by Service
Each new service reuses the rules and definitions already written, so policy changes once and every service follows.
Questions,
Answered
Can Prometheux run entirely on our own infrastructure?
Yes. The whole platform installs as containers on your own Linux servers, and heavy workloads can run on your own Kubernetes cluster. Data is read where it already lives and results are written where you choose.
How do we explain a decision to an appeal panel or auditor?
Each result carries provenance: the rule that produced it and the source records it used. The engine can turn that into a readable, fact-by-fact explanation.
Is the whole system deterministic?
The rules are: same data, same result. Language-model steps, such as summarising a case, are labelled as non-deterministic, so you know exactly which steps could vary and keep decisions in the rules.
Which language model does it use?
The LLM provider and model are settings you control, separate from the rules. Your agents, including a model you already run, can call the rules through MCP, the REST API or the Python SDK.
Is this only for benefits, or also for tax and supervision?
The same engine runs eligibility, tax, fraud-and-error and supervisory rules. Agencies, tax authorities, central banks and regulators each start with the decision they most need to defend, and reuse the same definitions as they add more.
Do departments have to pool their data first?
No. Prometheux connects to each department's existing databases, warehouses and files and joins them at query time, under the access controls already in place. Network and recursive rules run directly over tables, with no graph database to build.
More Industries
Bring One Decision
That Must Hold Up.
Tell us the eligibility rule, tax check, fraud-and-error review or supervisory report you need to defend. We'll run it as rules on your own data and infrastructure, with the record behind every result.
