// ai / industrial automation / deterministic evidence

The deterministic layer between AI and industrial automation.

Give coding agents one structured route to IEC 61131-3 analysis, fast diagnostics and vendor verification. 2ot keeps the inner loop deterministic, then hands off to Siemens TIA Portal, Beckhoff TwinCAT or CODESYS when stronger evidence is needed.

Illustrative CLI families; use 2ot --list --json for installed routes
Example commands: 2ot check --fast, 2ot --list --json, 2ot --mcp-dump-schema. The installed executable is authoritative.
closed-loop engineering

AI proposes. 2ot checks. Vendor tooling verifies.

See the agent workflow →
01

Agent proposes

Generate or modify IEC 61131-3 code from the engineer's chosen harness.

02

2ot analyzes

Parse, type-check and return structured compiler and semantic context.

03

Agent repairs

Use source-aware context instead of reasoning from an isolated error string.

04

Vendor verifies

Escalate to Siemens TIA Portal, Beckhoff TwinCAT or CODESYS when vendor evidence is required.

evidence, not confidence

Know what actually established the claim

2ot keeps different verification authorities conceptually separate. Compiler acceptance is useful evidence; it is not the same claim as vendor acceptance, controller observation or physical behavior.

01 · Inferred
Agent generated or reasoned
No independent verification
02 · Analyzed
2ot compiler / semantic analysis
Deterministic tooling evidence
03 · Vendor verified
Siemens TIA Portal / Beckhoff TwinCAT / CODESYS
Engineering-environment evidence
04 · Runtime / physical
Controller or machine observation
Only when an integration actually observes it

Fast deterministic feedback

2ot check --fast keeps the inner repair loop in the compiler, without waiting for a vendor IDE to start.

Semantic context, not grep

Agent tooling can work with source-aware compiler and semantic context so repairs are grounded in the program rather than a bare error string.

Real vendor boundaries

Dedicated Siemens TIA Portal, Beckhoff TwinCAT and CODESYS runtimes stay isolated behind the stable router instead of being flattened into a fictional common IDE.

safety architecture

From read-only analysis to physical action, authority should be explicit.

The current gateway classifies actions and denies raw persistent mutation over MCP. Consequential CLI operations require a protected transaction-service boundary. Vendor-specific rollback, deployment and physical guarantees must still be established separately.

Agent safety →

Bring the harness you choose

CLI and MCP share reviewed operation contracts, but their exposed command sets differ intentionally. 2ot stays underneath the agent framework instead of becoming another agent framework.

For agents →

Keep vendor knowledge at the right boundary

Vendor runtimes and licensed Siemens-derived API knowledge stay isolated from the compiler core and public website while remaining usable in the engineer's local environment.

Architecture →