Next steps

You have an agent, a tool, a workflow and a configured run. That is the whole surface you need for something real — everything past this point is depth.

Read this next

What runs without me writing it? — every framework hides something. The problem is not hiding it; it is not knowing what was hidden. This is the exact list.

Troubleshooting — the errors people actually hit on the first few runs, and what causes them.

Glossary — the terms used throughout these docs, in English and Portuguese, so the vocabulary stays fixed.

Things worth knowing exist

Each of these is a real, shipped feature. Until its page lands here, the README documents it in full.

Hooks	beforePrompt, beforeTool, afterTool, afterResponse, onError on the agent class — intercept without changing the framework
Injection	@input(), @context(), @state(), @memory() on tool and agent parameters, so order stops being a contract
Middleware and plugins	app.use({ name, tool, chat }) wraps every tool run and every model call — caching, rate limiting, metrics
Streaming	for await (const token of app.run({...}).textStream) — tokens as the model writes them
Cancellation	abort() reaches your tool's fetch through ctx.signal; stop() ends the run gracefully instead
Vector memory	@thenajs/qdrant-client plus @memory() for semantic recall
Live view	await app.use(thenaFlow()) from @thenajs/flow — the run as a graph in the browser
Nested runs	a tool that starts its own workflow, with its own history and its own budget

When something goes wrong

1. Turn on log: "verbose" — it shows the prompt and the tool I/O, which is usually enough.
2. Turn on report: true and open report/index.html — it lists every run, and each links to its own tree with durations and token cost.
3. Check Troubleshooting.
4. Still stuck: get support.

Contributing

The framework is at 0.9.x and moving. Issues, reproductions and pull requests are welcome at thenajs/ThenaJS; the docs live separately at thenajs/thenajs.github.io.