Honest line in the sand
What AgenticObjects is not. And what is next.
Positioning is clearer when the fences are explicit. Everything on this site that is not on this page is in the product today. This page is the single deliberate exception: we tell you what is not there, so you do not have to discover it yourself.
Six fences
What it is not
- Not a second semantic or BI engine. The semantic model is authored in Axoria Data Studio and imported as a versioned
.axrfile. This product imports, validates, versions and exports models. It does not author them. - Not a writer to your systems. The gateway accepts read-only grammar. Today an agentic object is something a person decides on, not something the platform executes. The evidence layer has to be trustworthy before anything is permitted to act on it.
- Not a general-purpose knowledge graph. Three record types, always populated correctly, rather than thirty populated occasionally.
- Not an LLM chat wrapper. The model narrates and orchestrates. It never produces the number and never persists knowledge without passing the gates.
- Not a claim of perfection. We do not claim zero error, we do not claim every number is necessarily right, and we do not claim equal accuracy in every language. The honest claim describes the mechanism and its refusal behaviour.
- Not a benchmark. Every figure on this site comes from our own operating record: three agents, sixteen scheduled runs, three and a half weeks. It shows the mechanisms firing under real conditions. It is not a customer result.
Status
Shipping, next, later
| Capability | Status |
|---|---|
| Ask the agent in plain language; answer with evidence card and addresses | SHIPPING |
| Open a finding and carry its proven evidence into the conversation | SHIPPING |
| Each scheduled run starts knowing what this agent already published | SHIPPING |
| A conversation started cold carrying the agent’s whole back catalogue as context | NEXT |
| Scheduled discovery: findings, recommendations, briefs | SHIPPING |
| Eight publication gates, Guard Ledger, withheld values | SHIPPING |
| Row-level security, named permissions, frozen budgets, append-only audit | SHIPPING |
| SQL warehouse and SSAS cube gateways, read-only | SHIPPING |
| Interpretive badge on causal / superlative wording | SHIPPING · Interpretation as its own record type: NEXT |
| Absence published as a data-state object | SHIPPING · Absence with an owner and a lifecycle: NEXT |
| Deterministic question matching tuned for Turkish; English and other languages via the model path | SHIPPING · English language pack for deterministic matching: NEXT |
| Brief distribution by e-mail | NEXT |
| Charting layer | NEXT |
| Excel / CSV as a source | NEXT |
| Question objects — the agent asks a named person when the data cannot answer | NEXT |
| Human input as a record with the author’s name | NEXT |
| Endpoint and MCP invocation modes; background Watch delivery | NEXT |
| Situation, Investigation, Decision — the macro layer | LATER |
In the schema, not yet exercised
Four states that exist but have not produced a live record
The urgency badge on recommendations, trust tiers B and C (every object so far has come out A), the overflow state (no run has yet exceeded its type cap), and the lesson-capture path. They are listed so you find them here rather than in a demo.
If it is not on the screen, it does not get a label.