Storylines: stateful, graph-orchestrated agents that deliver a process
A storyline is a directed graph an agent follows with each end user — every node is a step (its own task, knowledge and tools), exits carry conditions, and each person's progress, profile and notes are saved and resumed across sessions and channels. It turns an agent from an assistant that does point-tasks into one that can deliver a multi-step service itself.
Most agents do point-tasks: answer a question, book a slot, collect a detail. They sit around a business and make it faster. The teaching itself, the coaching itself, the consultation itself — the core of what the business sells — still needs a person. A storyline is what lets an agent take on that core: a multi-step, stateful process it runs with each individual, remembering where they are and adapting as they go. A language school's storyline doesn't just book the class; it teaches it.
A storyline is not "an agent with more skills"
This is the first thing to get straight, because the two look similar and are not. A skill is a capability the model can reach for; a storyline is an order things happen in.
| An agent with several skills | An agent running a storyline | |
|---|---|---|
| Sequence | Any skill, any time — no order | "You can't reach B until A is done" |
| State | Doesn't remember where you were | Each person's progress is saved and resumed |
| What drives it | The model decides, each turn | The graph and its conditions decide |
| Determinism | The order lives in the model's head | The order is drawn as a graph — versioned, measurable |
Skills answer what an agent can do. A storyline governs what happens in what order, and where this particular person is right now. The two aren't rivals — a storyline is a layer above skills: each node still hangs its own skills, knowledge and tools. And because the sequence lives in an explicit graph rather than the model's working memory, a storyline stays reliable as the process gets long — the model only has to handle the current step, not remember a twelve-step procedure without slipping.
The pieces
- Nodes are steps. Each node reconfigures the same agent for that step: its own task prompt, knowledge bases, skills and tools (added on top of the agent's base, or replacing it). The agent's identity and safety boundaries never change between nodes.
- Exits are how a node advances, and each carries a condition: a rule on the profile
(
listening ≥ 70), the learner picking a button, an AI judgement of a fuzzy criterion ("when they can form the sentence unprompted"), a signal from an outside system, or a jump to another storyline. - A profile travels with the person — dimensions you define per subject (listening, speaking, reading, writing for a language; something else entirely for history), accumulated across storylines, plus a free-form scratchpad of handoff notes passed from one node to the next.
- Chaining links storylines end to end — grade 1 flows into grade 2 — so you can grow a large journey without building it all at once, and branch to a different storyline on the outcome (a learner who doesn't pass the exam moves into the remedial line, not the next grade).
Progress is per person, not per chat, so someone can leave mid-way and pick up exactly where they were — on the web, in Telegram or WhatsApp, wherever they return.
Bringing in outside systems
A node can hand off to a third party and act on what comes back: a graded assessment, a KYC check, an embedded mini-game. That result — a score, a verdict, a payment confirmation — can update the profile or satisfy an exit. Two shapes: a backend hand-off (data flows out to a tool and a result flows back, invisibly) and a redirect-and-wait hand-off (the conversation shows a link, locks, and resumes when the outside step finishes — the way a payment flow does).
Where it fits
Storylines suit any business whose core service is a structured, staged, per-person conversation: tutoring and language practice, coaching and accountability programs, staged advisory and guided selling, application intake and pre-qualification, certification and training. Anywhere the value is today capped by a person's hours, a storyline lets one agent run that process for thousands of people at once — while the parts that genuinely need a human stay with the human.
Storylines are a Pro capability, switched on per tenant. They are authored as a graph in the console (or over the API, so a wizard or an assistant can build one), versioned and published, with in-flight learners pinned to the version they started on.