Workflows & agents

An agent in Tln is an automation the LLM writes once and the runtime then runs forever — deterministically, with no model in the loop at run time. A user describes a task in chat (“watch stock item ABC-123; when it drops below 10, open a refill ticket”), the LLM authors it as Tln source, and a plugin stores it and runs it autonomously.

That split is the whole point (see the deep dive in Deterministic Where It Matters):

• Authoring is probabilistic — the LLM is great at turning a fuzzy request into a small Tln program. This happens once.
• Execution is deterministic — from then on the runtime evaluates that program on every tick, over the facts of the moment. No model call, no sampling, no re-deciding. Same facts in, same decision out.

A workflow fired by a trigger

Two blocks make an agent: an on trigger that watches the facts, and a workflow it fires. A workflow is a sequence of steps, each calling a tool (see Plugins). This is the real stock-watcher from opentalon-agents:

// Fire ONCE on the downward crossing below 10 (prev >= 10, new < 10) --
// not every tick while it stays low.
on change attr "current_stock" {
  when prev_value >= 10 and new_value < 10
  workflow "Refill stock"
}

workflow "Refill stock" {
  step "ticket" {
    tool "tickets" "create" {
      title "Refill needed for ABC-123"
      item  step("trigger").result.entity
      qty   50
    }
  }
}

The on change block is edge-triggered: it fires on the moment stock crosses below 10, not on every tick while it stays low. The workflow then opens a ticket through the tickets tool — step("trigger").result.entity threads the item that crossed the threshold into the call.

How it runs — opentalon-agents

opentalon-agents is the OpenTalon plugin that owns the agent lifecycle: it stores the LLM-authored Tln source and its triggers, maps incoming data to facts, keeps the fact snapshot, and records every run — all in its own store. It runs no model at run time and no scheduler of its own; it rides the host’s periodic tick and evaluates the stored source reactively against the current facts.

  user (chat) ── create ──►  opentalon-agents ──► stores Tln source + trigger
  host tick (every 1m) ───►  opentalon-agents ──► evaluate source over facts
                                                    │  on-block fires?
                                                    ▼
                                             workflow steps ──► tools (MCP / io)

Authoring stays with the LLM; the decision — every time, forever — is Tln’s. That’s a deterministic agent: reproducible, auditable, and cheap to run.

See it in production in OpenTalon, and the full argument in Deterministic Where It Matters.