Docs

Start Here · 01.05

Your First Workflow, App, and Reach: The Paved-Road Walkthrough

Describe intent, build, validate, dry-run, run, and deliver — the paved road every Agentis workflow follows, walked end to end with a real example.

Paved roadWorkflowsCode mode

The paved road

From here, an operator agent builds by calling the SDK, not by improvising tool calls in whatever order seems reasonable. There's one recommended path from an idea to a proven result, and every settled tool call nudges the agent toward it with a compass.next hint — guidance that lives in results, not just in a wall of upfront prompting.

The Agentis paved road Six steps from intent to a delivered result: describe intent, build, validate, dry run, run, deliver. Step 1 Describe intent Step 2 Build Step 3 Validate Step 4 Dry run Step 5 Run Step 6 Deliver Every step is one agentis.* call. The engine — not the agent — remembers where it left off.
The paved road: describe intent, build, validate, dry-run, run, deliver. Each arrow is one agentis.* call; the engine remembers progress so an interrupted build resumes instead of restarting.

Walk it end to end

Paved-road build loop (code mode)
const built = await agentis.build_workflow({ intent: "Follow up new leads by email within 5m" });
await agentis.workflow.validate({ workflowId: built.result.workflowId });
await agentis.workflow.dry_run({ workflowId: built.result.workflowId });
const run = await agentis.workflow.run({ workflowId: built.result.workflowId });
await agentis.run.await({ runId: run.result.runId });   // blocks server-side, 0 tokens

Notice what's absent: no polling loop, no manually re-checking status every few seconds burning tokens on "are we done yet?" turns. run.await is event-driven — it subscribes to the run's room and returns the instant the run settles, terminal or waiting, at zero token cost. That single call is why agents built on Agentis don't need to be taught patience.

Every workflow is a product, automatically

A standalone workflow doesn't stay a bare graph for long — it auto-wraps into an App-of-one the moment it's built. That means a product shell, a typed datastore, and a live-operations view all show up for free, with no extra step. You'll see the full shape of this in Apps & Interfaces.

Where to go from here

You have a running Agentis instance, a bootstrapped agent, and the one build loop that everything else in this platform reuses. From here, three paths are worth taking in order:

Own Your Agents

Bring in the agents you already run elsewhere, and understand the seam that lets you swap models without losing memory.

 Talk to Your Agents

The conversational layer: chat, permission modes, and watching agents work together in real time.

 The Brain

How memory actually forms, gets recalled with citations, and compounds across every run.

Continue

Bootstrap Your First Agent

Wire a durable Agentis identity to a model runtime with one CLI command, agentis bootstrap, then verify it's alive from the dashboard.

BootstrapCLIAdapters
 Own the Agent, Rent the Model

The Agentis thesis: identity, memory, and skills are durable platform state you own; the model executing a turn is a swappable, disposable tenant.

ThesisOwnership