rAIson by Argument Theory

Workflows

Four ways to reach the same deployed agent.

Pick the route that fits how you work. Every one ends with a symbolic AI reasoning agent live on rAIson and reachable through the same API endpoint.

1
Developer workflow · build directly

Dialogue

Model the decision problem through a natural-language dialogue with the platform. No coding, no AI expertise — you work in your own application vocabulary.

  • Natural-language dialogue with rAIson
  • PAF code auto-generated
  • Deployed on the platform
  • Accessible via API
2
Text-scenario workflow

From text

Upload a policy or scenario described in plain language. Claude drafts the encoding by following a dedicated modelling guide of more than 120 pages; you review and validate it before deployment.

  • Upload text scenario
  • Claude generates PAF code
  • You validate (revise as needed)
  • Deployed · API
3
Tabular-data workflow

From data

Upload numerical or CSV data. Claude drafts the encoding from your columns using the same modelling guide; you review and validate before it goes live.

  • Upload CSV / numerical data
  • Claude generates PAF code
  • You validate (revise as needed)
  • Deployed · API
4
Customised development · by Argument Theory

Done for you

Describe the agent type and application domain. The Argument Theory team builds it for you using the same infrastructure behind workflows 2 and 3.

  • You describe agent & domain
  • AT team builds it
  • Deployed on the platform
  • Accessible via API

In workflows 2 and 3, a large language model (Claude) reads your text or data and maps its structure — defaults, exceptions, overrides — onto PAF rules, following a dedicated prompting guide of more than 120 pages. Every workflow compiles a PAF program to CNF for SAT solving underneath.

One endpoint

Different routes in. The same agent out.

Whichever workflow you choose, the result is identical in shape: a deterministic symbolic AI agent deployed on rAIson, exposed through one API, returning verdicts with full rule traces.

How the engine works →

Encode once · execute many

  • Encode once — the model is consulted per source, not per decision.
  • Execute many — every decision is a deterministic SAT call.
  • Discipline between — the modelling guide is the contract.

Not sure which route fits?

Tell us about your decision problem and we'll point you to the fastest path to a working agent.

Request access