Using io in Console
Use io in the Vectara Console to describe a task, review a proposed configuration, and check the result in the visual workspace. This guide uses creating an agent as the introductory example.
For the broader task overview and prerequisites, start with Getting started with io. For shell or script use, follow Using io from the terminal.
Describe the task
Name the outcome and the resources you want to use. For example, if you already have an indexed corpus named product-docs, you could ask:
Create a support Q&A agent over the product-docs corpus. It should cite answers and say when it does not know.
Replace the example corpus name with your own. If you do not know which corpus to use, ask io to inspect your available corpora first. Creating an agent configuration does not ingest missing documents.
When editing an existing resource, open it in Console or supply its name or key. Check that the proposed change targets that resource. See Write prompts that reduce follow-up questions for more examples.
Review what io creates
Expand View proposed body on the confirmation card to read the request io drafted. Choose Review and edit to inspect the draft in the linked editor, Accept to approve the request without editing it, or Cancel to decline it.
Review the relevant sections under Configure before accepting:
| Section | What to verify |
|---|---|
| General | Confirm the Name, Key, and Description. The key identifies the agent in API paths and cannot be changed after the agent is created. |
| Model | Review the Model name and JSON Parameters against your quality, latency, and cost requirements. |
| Tools and skills | Confirm that tools point to the intended resources. For a retrieval agent, inspect the Corpora Search tool and its selected corpora. |
| Steps and instructions | Check the behavior for citations, missing information, escalation, and any source policy the agent must follow. |
| Reminders | Add reminders only when the agent needs guidance reinforced during the session. |
| Advanced | Review metadata, compaction, structured output, and other settings relevant to your workflow. |
You can also use Edit JSON to inspect the agent definition. For existing agents, review the change-set and text-replacement guidance before approving an update.
If io asks a clarifying question, an option may have a Recommended badge. Console does not preselect that option: click an option to answer, or choose Write your own answer to respond in your own words.
Accept and test the agent
For a draft opened through Review and edit, choose Accept in the linked editor to approve its configuration, or Cancel to decline it. These controls replace Discard and Save while the proposal is pending.
Approval does not establish that the API request succeeded. Check that io reports the created agent and its key, then open that agent and inspect its configuration before using Preview. If the outcome is unclear, ask io to check the resource rather than assuming it exists.
Check that the agent uses the intended corpus and tools, follows the requested behavior, handles missing information, and returns citations when required. For subsequent manual edits outside an io-linked review, return to Configure, change the relevant settings, and use Save before retesting.
If creation fails or a multi-resource task stops partway through, check failure and progress guidance before continuing.
Finish the required setup
| Tab | Use it to |
|---|---|
| Schedules | Create recurring agent runs. |
| Connectors | Connect the agent to supported external systems. |
| Secrets | Add encrypted credentials for tools. |
Tools reference an agent secret from argument_override with {"$ref": "agent.secrets.<name>"}. Add and verify the actual secret after the agent exists; a reference does not supply a credential. See Secrets.
Terminal usage
For terminal setup, approval, automation, and local files, use Using io from the terminal.
Terminal topics
Scope the CLI API key
See CLI permissions and provisioning before supplying a key.
Review a proposed change
See terminal approval and the shared agent-update table.
Answer clarifying questions
See terminal question controls, which differ from Console's question cards.
Running io unattended
See unattended execution and its approval requirements.
Working with local files
See file and screenshot inputs, including image-size limitations.
Design a web app
See web-app design and shared credential considerations.
Previewing an agent in a browser
See local browser preview and its local-only restrictions.
Deployment and LLM requirements
FAQs
Does clicking Accept mean the agent was created successfully?
No. It approves the configuration and resumes io. Check the API outcome and resulting agent before testing or continuing a larger task.
Why is Review and edit missing from a proposal?
Not every card supports opening an editor. Lambda-tool update cards and generic API-call cards do not offer that action. Inspect View proposed body; cancel an incorrect proposal and ask io for a revision. See editor availability.
Should I use Save or Accept?
Use Accept for the outstanding proposal in an io-linked review. Use Save for subsequent manual changes in the standalone editor. The linked review and ordinary editing are different workflows.
Does io supply the credentials needed by my agent's tools?
No. Add the required secret values after the agent exists, and verify that the tools reference them correctly. Do not mistake a generated credential reference for working authentication.