Product
Five layers. One runtime.
Each layer can be used independently. Together they cover the full spectrum — from a single decision table to a full AI-augmented workflow.
Rules as Code
Versioned decision tables. Readable by analysts. Full test suite before deploy. Same input — same output, every time.
Versioned · testable · deterministic
16 decision tables with eval order, rule counts, and gap warnings
Goal-Directed Orchestration
Declare goals and skills — the system determines execution order and parallelizes automatically. Async and failure recovery built in.
Declare what, not how
Execution graph with skill dependencies and compensator contracts
AI Under Rules
AI extracts, scores, classifies — as a full participant. But everything AI writes goes through explicit rules before affecting the outcome.
AI participates, rules control
Full Observability
Every tick, decision, mutation, and outcome is traced. Inspect any decision path. Replay any session.
Complete audit trail
Phase 2 — consumed event, state snapshot, ticks waterfall
Versioning & Rollback
Every rule, workflow, and decision table is versioned. Every change has a diff, author, and timestamp. Roll back any version in seconds.
Git for business rules
Spectrum
Start where you need.
From a single decision table to a full AI-augmented workflow.
| Need | You describe | You get |
|---|---|---|
| Simple rules | If/then conditions in a table | Deterministic, reproducible answer |
| Linear process | Steps that depend on each other | Step-by-step execution with audit |
| Parallel process | Steps that don't depend on each other | Automatic parallelization |
| Async process | Steps that wait for external response | Pause → resume without losing context |
| Failure recovery | What to undo if something breaks | Automatic reverse-order rollback |
| AI at the input | Unstructured text → structured data | Text → data → decision |
| AI in the decision | AI scores context, rules decide | AI participates, rules control |
Deep dive
Agent anatomy
One agent, fully unpacked. How skills, rules, and contracts come together.
Design time
Skill registry — requirements, guarantees, async contracts, compensators
Dependency graph between decision tables — chain of inference
Loan eligibility rules — conditions, outcomes, human-readable notes
Runtime
Playground — test any scenario, see which rule fires and why
Guided tour
Session deep dive
One session from start to finish. Every phase, every tick, every decision.
1 / 8
Session list — statuses, errors, durations at a glance
2 / 8
Time-travel — select Phase 1, see the graph at that point
3 / 8
Live graph — request_approval submitted, waiting 42.8 hours
4 / 8
Click a node — full DMN reasoning chain, rule 5 matched
5 / 8
Phase 1 detail — state snapshot after validate_input + screen_fraud
6 / 8
Phase 2 detail — ScreenFraud event consumed, state updated
7 / 8
Expanded tick — skill selection table, commands issued, goals 5/8
8 / 8
Error case — FAILURE triggers compensation cascade, full audit trail
Architecture
Two layers of governance
Business Rules
Decide what outcome to produce
Eligibility, pricing, risk categories, scoring, routing. DMN decision tables evaluated every tick.
Execution Policies
Enforce whether an action is allowed
Compliance boundaries, rate limits, jurisdictional rules, access controls. A separate layer that gates every side effect.
Ready to take control?
We're shaping TIATON with a small group of companies. Get a sandbox, direct support, and influence on the roadmap.
Get Early Access