AI & Automation

DrToller.Framework is designed so that DRTML is both human-authorable and machine-generatable. Primitives are human-extensible; pipelines are machine-composable.

Humans extend the Framework. AI and humans compose the Framework’s primitives into pipelines.

The Framework does not contain an LLM. External agents (or other software) discover contracts, draft DRTML, and repair it from structured validation errors.

Two-sided model

Human developers
      │ implement
      ▼
Framework primitives
      │ registered + described
      ▼
Machine-readable capability catalog
      │
      ├───────────────┐
      ▼               ▼
    Human             AI (external)
      │               │
      └───────┬───────┘
              ▼
            DRTML
              ▼
          validation
              ▼
        compiled plan
              ▼
           execution

Humans expand what the Framework can do. External AI expands what users can build with those capabilities.

Available today

Capability	API / CLI
Discover primitives	`build_catalog()`, `drtoller inspect`
Read contracts	`get_contract(id)`, `drtoller inspect <id>`
Composition	`can_compose`, `compatible_consumers`
Validate DRTML	`validate_manifest_path()`, `drtoller validate --format json`
Compile / plan	`compile_step_manifest` / `drtoller plan`
Bootstrap project	`drtoller init`

Worked example (no LLM): examples/agent-pipeline-loop.
AI quickstart: AI quickstart.

What the Framework does not do

It does not call LLMs or ship OpenAI/Anthropic clients.
It does not turn natural language into DRTML inside the runtime.
It does not replace DRTML with a second config language.
It is not an AI application framework (see Compare for LangChain-class layers).

Framework surface: machine-readable contracts, catalog, composition, structured validation.
External consumer: coding agent / IDE / pipeline builder that generates and repairs DRTML.

Next pages