Skip to content
Viberia Docs

Anthropic Claude (CLI)

Anthropic’s claude command-line tool is the default engine behind every agent in Viberia, including the Chief of Staff.

Setup

Section titled “Setup”

You install the claude CLI on your computer, sign it in to your Anthropic account once, and then point Viberia at the binary.

Prerequisites

  • An Anthropic account with API access (or a Claude subscription that authorizes the CLI).
  • A network connection — the CLI calls Anthropic’s API on every turn.
  • Permission to install a global binary on your machine (or to place one on your PATH).

Install the CLI

Follow Anthropic’s macOS instructions — install via the official installer or npm install -g @anthropic-ai/claude-cli (verify the current package name in Anthropic’s docs).

Once installed, run it once from a normal terminal to sign in:

Terminal window
claude

Complete the auth flow. The CLI persists your credentials so Viberia doesn’t need to handle them.

Point Viberia at the binary

Open Settings → CLI Configuration. The Claude CLI path field (a CliPathField row) is where Viberia stores the path. If the binary is on your PATH you can paste claude; otherwise paste the absolute path. To find it:

Terminal window
which claude

Click the Test button next to the field. A green check means Viberia can invoke the binary. A red error means the path is wrong or the CLI isn’t installed.

Note: If you change the path while agents are running, restart Viberia so every agent picks up the new binary.

Configuration

Section titled “Configuration”

Two layers control which Claude model runs for any given agent: a system-wide default per role tier, and a per-agent override.

Role-tier defaults

Open Settings → Agent Defaults. Roles are grouped into tiers (for example, a coordinator tier covering the Chief of Staff and Deans, an engineering tier covering Planner / Developer / Reviewer, and so on). For each tier, pick the model you want new agents in that tier to use by default. Existing agents are not retroactively changed — only newly spawned agents inherit the new default.

Per-agent override

Open any agent’s chat window and use the Model & Tools cell above the composer. The model dropdown lists every Claude model Viberia currently supports. The change takes effect on the agent’s next turn and persists across restarts. You can also flip the agent’s thinking mode here.

SettingWhereDefaultNotes
Claude CLI pathSettings → CLI Configurationplatform default (e.g. claude on PATH)Restart agents after changing.
Default model per role tierSettings → Agent Defaultsprovider’s recommended balanceAffects newly created agents only.
Per-agent modelAgent chat → Model & Toolsinherited from role tierTakes effect next turn.
Per-agent thinking modeAgent chat → Model & ToolsregularOff / regular / extended.

Spawning agents on this provider

Section titled “Spawning agents on this provider”

You don’t “spawn on Claude” explicitly — every agent uses Claude by default. To make an agent use Claude when its role tier was overridden to another provider, open the agent’s Model & Tools cell and pick a Claude model from the dropdown. To make all new agents in a tier default to Claude, set the tier default in Settings → Agent Defaults.

When you create a new building with the Building Creation Wizard, the agents it adds inherit the role-tier default at creation time. You can change any of them afterward.

Features

Section titled “Features”
  • Model selection — pick between the supported Claude models (Sonnet 4.6 as the everyday default, Opus 4.7 for harder reasoning, Haiku 4.5 for cheap quick turns). The exact list reflects whatever is currently shipped in your Viberia build.
  • Thinking modes — three settings per agent: off (fastest, no extended reasoning), regular (balanced), extended (deepest reasoning, slowest, highest token usage).
  • Skills (slash commands) — Claude agents can invoke every installed skill via /<name>.
  • MCP — Claude agents can use every MCP server configured at the user, building, or agent scope.
  • Tool streaming — tool calls render in the agent chat as they execute, including diff previews and permission prompts.
  • Memory — agents maintain a file-based memory directory and read it themselves when relevant.

Limitations

Section titled “Limitations”
  • Account required. Every Claude agent call goes to Anthropic’s API. You need an account in good standing and a valid sign-in on the claude CLI.
  • Network required. No offline mode. If your connection drops, the agent’s next turn fails until the network returns.
  • API rate limits and quotas apply. Heavy use across many agents in parallel can hit Anthropic’s per-account rate limits. The error surfaces in the agent’s chat as a failed turn; retry after a moment.
  • Token cost is your responsibility. Viberia’s Token Tracker (toggle with Cmd+Shift+U) shows running usage per provider so you can keep an eye on it.
  • Model availability tracks Anthropic. When Anthropic deprecates a model, it disappears from Viberia’s model dropdowns at the next app update. Existing per-agent overrides may need to be reset.
  • Mid-session opt-off has limits. Disabling analytics mid-session closes the JS telemetry client immediately, but other consent-gated changes (Rust shell, gateway) apply on the next restart.
Section titled “Related”