Skip to Content
We are live but in Staging 🎉
Overview

Scriptum

Scriptum is Dodil’s workflow and agent engine. You write a workflow in a typed, purpose-built language — a .scriptum file — Scriptum compiles and type-checks it, then runs it as a thread: a durable, resumable execution in the cloud that calls tools, branches on LLM decisions, fans out in parallel, and streams results.

.scriptum (DSL) → compile → IR → thread (runs in the cloud) → result

One engine covers two things you’d normally reach for separately: deterministic step workflows (do this, then that, in parallel, with retries) and LLM agents (think → call a tool → observe → repeat). They’re primitives in the same language, so a mostly-deterministic workflow can drop in an agent step, or an agent can call a structured sub-workflow.

Why Scriptum

  • A typed language, checked before it runs. Inputs, outputs, and streamed fields are declared with types and constraints, and the compiler validates your workflow against each tool’s I/O schema — typos and bad field access are caught at compile time. Every script publishes a ScriptContract: a self-describing schema other systems (forms, K3 routers, LLM agents) validate against.
  • More than an agent loop. A composable set of primitives — do, each, decide, together, pipe, agent, ask, plan, emit, yield, wait, … — see Scriptum Language.
  • Durable cloud threads. A run is a thread with persistent per-step state. Threads are resumable: on an unhandled error a thread pauses so you can inspect and resume it instead of losing the run. Progress streams live.
  • Versioned like Ignite. Author in a draft → compile → test → publish an immutable numbered version; roll back anytime.

How it fits with Dodil

  • Inference runs on Ignite . Every model capability — llm, embed, rerank, transcribe, infer — dispatches to Ignite’s managed model service. Cloud tools run as Ignite serverless functions.
  • Data lives in K3 . Native tools read and write K3 tables, object storage, and vector collections.
  • K3 pipelines run Scriptum templates. A K3 pipeline names a Scriptum script as its processing step and validates inputs against that script’s ScriptContract.

In this section

CLI: dodil scriptum — see Install the CLI .


See also