Getting started with io
io is Vectara's AI assistant for working with the platform. Describe an outcome, ask about your account or Vectara APIs, and review the changes io proposes. This page helps you choose an interface and prepare for your first task.
Choose how to use io
Console and the terminal are alternative starting points; you do not need to complete both guides.
| Start here | Use it for |
|---|---|
| Using io in Console | Work in the visual workspace, inspect proposed configurations, and test an agent in Preview. |
| Using io from the terminal | Work from a shell, provide local files, or run io from scripts. |
| Working effectively with io | Improve requests, review updates, handle incomplete work, and diagnose an agent session. Read this after choosing an interface, or when you need help with a specific task. |
Both interfaces use the public Vectara REST API v2, but their controls and approval handling differ. Follow the guide for the interface you are using rather than assuming a Console button has a terminal equivalent.
What you can ask io
These are examples of requests, not complete tutorials or a guarantee that a generated configuration is ready for production.
| Task | Example request |
|---|---|
| Inspect your account | List the agents, corpora, or tools available to me. |
| Plan content ingestion and search | Help me get these documents into a corpus and query the content. |
| Build or review an agent | Propose an agent for this task, including its steps, tools, and an enabled LLM to evaluate. |
| Configure external tools | Help me connect an agent to this API using a credential reference. |
| Plan a pipeline | Help me keep a corpus updated from this external source. |
| Organize independent work | Help me structure independent tool calls or sub-agent work in parallel. |
| Investigate behavior and performance | Diagnose this agent session, or show usage and latency for this agent. |
| Evaluate answers | Help me check whether answers are grounded in retrieved evidence. |
| Connect an agent to Slack | Help me configure a Slack connector for this agent. |
| Learn the API or Python SDK | Explain the relevant API operation or help me build this workflow with the Python SDK. |
Use the platform guides for the underlying concepts and the API reference for exact operations and fields. io guidance does not replace either. A corpus configuration, for example, is not the same as indexed content that an agent can search.
Before your first task
- Identify the outcome and the resource you want to work with. If the resource already exists, have its name or key available.
- Use an account or API key with permissions appropriate to the task. Do not grant wider access merely to avoid a permissions error.
- For an agent that answers over your content, identify the corpus and whether its data has been indexed. If ingestion is part of the task, track it as required work before testing the agent. For an external service, identify the credentials and permissions it needs.
- Review proposed configurations and test the result. A proposal, an approval, and a successful API request are different stages.
The Console guide uses agent creation as a first example. The terminal guide covers installation, API-key scope, and provisioning before automation.
Deployment and LLM requirements
io depends on an LLM that can drive tool calls and follow its guidance reliably. A registered LLM is not, by itself, evidence that a generated configuration will be correct. Validate proposed agents against representative inputs.
On a self-managed or air-gapped deployment, confirm with your administrator that io is enabled, that a suitable LLM is configured, and that platform services can reach the deployment's API hostname. io reads and writes through the REST API and needs that network access.
Terminal-specific requirements, including the Node.js version, are in Using io from the terminal.
FAQs
Do I need to learn the terminal before using Console?
No. Choose the interface that fits your task. Start with Console for the visual workflow or the terminal guide for shell and script use. Their approval controls differ.
Do I have to create an agent before I can ask io a question?
You can start with account inspection or API and product questions rather than creating your own agent. The CLI may still need to provision its own io assistant before it can run; that setup is covered by the terminal's approval and provisioning rules.
Does creating a corpus also make my documents available to an agent?
A corpus configuration alone does not provide indexed content. Check that the intended documents have been ingested and that the agent uses the correct corpus. See Data ingestion.
Can I assume io is available on a self-managed deployment?
No. Confirm availability, model configuration, and API connectivity with your administrator. The requirements above are prerequisites, not a promise that io is enabled on every deployment.