Skip to main content
Vibex uses the Agent Client Protocol (ACP) to connect to online Agents. The Agent (who does the work) and the model provider configuration (which model, over which route) are managed separately. You work with both in the Config Center → Agent tab.

​
How the Agent tab is organized

When you open Config Center → Agent, the Agent list is on the left and the details area for the selected Agent is on the right. The details area has three cards, from top to bottom:
Model providers are not a separate tab. They hang off each Agent, because different Agents support different communication protocols and model catalogs.

​
Add an Agent

The list on the left is grouped by status, and the action on each entry depends on its current state: Vibex ships four built-in presets: Claude Code, Codex, ZCode, and OpenCode. The ACP catalog also offers integrations such as Google Antigravity, Cline, Codebuddy Code, GitHub Copilot, Cursor, DeepSeek Harness, Devin CLI, Gemini CLI, Grok, Hermes, Kimi Code CLI, and Pi.
Install downloads and verifies the runtime over the network and cannot be canceled. Uninstall shows a confirmation dialog and deletes that Agent’s managed runtime and cached versions; session history is kept.

​
Add a custom ACP Agent

Click Add custom ACP Agent at the bottom of the list and fill in: Custom Agents can only be created, not edited. Turning the switch off deletes one, but session history is kept.

​
Sign in to an Agent account

The Native credentials card manages the Agent’s own account. It shares the same credentials as the external CLI—the card states this explicitly: “Agent-native account · shared with the external Agent CLI”.
  • The available sign-in methods are up to the Agent. Common ones are Sign in, Save and sign in, and Open sign-in terminal.
  • The card provides Refresh methods, Verify account, Refresh models, Re-authenticate, and Sign out.
  • Re-authenticating affects Agent processes still using the old credentials; the UI warns you first.
  • Sign out affects both the external CLI and Vibex sessions, and shows how many sessions are affected before it runs.

​
Configure a model provider

In the Model provider configuration card, click Add config, or click an existing configuration to edit it. The editor has three parts:

​
Basics

​
Connection & credentials

The API key field is write-only: it is masked by default, with an eye icon to reveal it, and shows Saved when a value is stored. An untouched field never rewrites the stored value, and leaving it blank keeps the current one. To remove it, click Clear; the state becomes Will be cleared and the button becomes Keep stored key.

​
Models

The editor splits models into two columns: the Available models picker on the left and the selected model’s Model settings on the right. The picker is the only model list — click a row and the right column shows its editable fields instead of repeating the list. Vibex does not ship a model list:
  • Fetch models — requires an API key and address; loads the models available to that account into Available models. Fetching on its own configures nothing.
  • Available models — checking a row adds that model to the draft; unchecking releases it while keeping its declared capabilities, so re-checking restores them. Select all / Enable all handle the list in bulk; when there is nothing to fetch, type an id under Custom model and add it.
  • Model ID / Display name — edited in the right column. Selection follows the model id, so renaming keeps the column on the model you are editing.
  • Model API protocol — each model can specify its own protocol; by default it Inherits the provider setting.
  • Reasoning / Thinking levels — choose whether to reason, or declare thinking depth levels.
  • Context window / Max output tokens — declare model limits in the Advanced collapsible section.
When the provider does not offer a model, the right column offers Add this model, which adds it to the list. The available thinking depth levels come from the provider or Agent. Common values are off, minimal, low, medium, high, xhigh, and max.

​
Use them in a session

Before you send a message, choose the Agent, Model provider, Model, and Thinking depth from the cascading menus at the top of the Composer. A switch affects only later runs; existing timelines keep the Agent and model information from their original run.
Test connection and Fetch models make real requests to the provider’s network and may consume quota. read_only mobile devices cannot trigger these operations.

​
Troubleshooting

  • The Agent is not in the list: confirm the executable is on PATH, or check the custom command and arguments.
  • A model cannot be selected: run Refresh models first, then check whether that model is enabled.
  • Authentication fails: check the account state and sign in again; do not paste response headers or tokens into an issue.
  • Remote clients do not see the configuration: confirm the device permission level is full_control and which authoritative runtime you are connected to.
  • A managed Agent fails to install: check whether the network can reach GitHub and nodejs.org, then retry.