Plain-English questions. Verified answers. No SQL required.
Natural-language interface over your existing ERP, databases, and reporting tools, with 6 deterministic validation stages before any result reaches a user.
Your organisation has the data. Access requires a SQL developer and 2 days.
Most enterprise data is inaccessible to the people who need it most. Finance wants the number now. Operations wants the anomaly flagged this morning. Instead, they submit a request, wait for IT, get a dashboard that answers the previous question.
We build a natural-language layer over your existing data sources, ERP, databases, CRM, reporting tools, with a deterministic validation pipeline that ensures nothing hallucinated reaches the user. The system maps your schema once and maintains it. Your team asks questions in plain English.
Data access requires SQL expertise and schema knowledge. Reporting takes 2 days. IT is a bottleneck on every operational decision.
Any stakeholder asks in plain English. Schema-aware query generation. 6 deterministic validation stages. Result in seconds, cited to specific data records.
How we build it.
Every stage has a defined role and a quality gate before the next one runs, the same engineering spine under every Agentix system.
- 01
Schema Discovery
We map every table, field, relationship, and business rule in your data source. For Vista ERP this is 19,249 schema entries. This becomes the persistent knowledge base.
- 02
Disambiguation Layer
Schema-aware disambiguation resolves ambiguous column names, multi-company data, and complex join paths before any query is generated.
- 03
Query Generation
13-stage pipeline generates and validates SQL/SSRS queries. Stages 2–7 are deterministic, chromaDB semantic lookup, not LLM guessing.
- 04
Validation Pipeline
Syntax validation, schema validation, execution against a read-only replica, result reconciliation. Nothing reaches the user that has not passed all stages.
- 05
Reporting Integration
Results surfaced in Power BI, Excel, or custom dashboards. For existing Crystal Report corpora, VistaForge converts them to PBIR format.
- 06
Eval & Monitoring
Benchmark against known good queries. Langfuse trace every execution. Hallucination rate tracked per deployment, target: near-zero.
Delivered on
every engagement.
You own all of it, code, infrastructure and data, from day one. No licence fees, no hosted dependency.
- Schema manifest, structured map of every table, field, and relationship in your data source
- Multi-engine query generation with schema disambiguation (handles ambiguous column names)
- 6-stage deterministic validation pipeline, syntax, semantic, schema, execution, reconcile, citation
- Power BI REST API / Excel COM integration for existing reporting infrastructure
- LiteLLM abstraction for model independence, swap providers without schema work
- Full Langfuse observability, every query, every stage, every result traced
Not a demo. In production.
Open-source.
Self-hosted. Owned.
Built on components you can host, fork and extend. LiteLLM abstraction means you swap models without rewriting integrations.
Is this your
problem shape?
Describe what you're trying to do. We'll confirm whether this is the right system type, scope the build, and tell you what the eval harness looks like.