Claude Code

Tejus connects to the Claude Code CLI installed on your machine and drives it through Anthropic's Claude Agent SDK. Claude owns the account, model access, tools, and authentication; Tejus owns the task, working environment, transcript, approvals, review surfaces, and Git lifecycle around the session.

Install

Use Anthropic's current setup instructions. On Windows, install it either with npm:

npm install -g @anthropic-ai/claude-code

or with Anthropic's native installer from PowerShell:

irm https://claude.ai/install.ps1 | iex

The native installer places claude.exe in %USERPROFILE%\.local\bin; make sure that folder is on your PATH. Tejus looks for the claude executable on PATH.

Verify the installation from a new PowerShell window:

claude --version

Authenticate

In Tejus, open Settings → Agent providers, expand Claude under Provider tools, and click Sign in. Tejus runs claude auth login in a terminal window; choose your login method and complete the browser sign-in. You can also start Claude Code in a terminal:

claude

Claude Code supports Claude account subscriptions, Anthropic Console access, and enterprise backends such as Amazon Bedrock, Google Vertex AI, or Microsoft Foundry. Tejus passes the matching environment variables (for example CLAUDE_CODE_USE_BEDROCK and the cloud credentials) through to Claude.

Inspect authentication directly:

claude auth status

Tejus runs the same check to show the provider's status. It does not copy or replace Claude credentials.

Verify

Before opening Tejus, verify Claude Code from a new terminal in a test repository:

cd C:\path\to\test-repository
claude --version
claude

Ask a harmless read-only question, then exit. This checks the binary, authentication, repository access, and provider service independently of Tejus.

Connect to Tejus

Capabilities in Tejus

Claude Code sessions can expose:

• Model and effort selection
• Plan mode and the Ask for approval, Approve for me, and Full access permission modes
• Tool calls, file changes, and command output
• Approval requests and user questions
• Resumable sessions
• Claude subagents and tracked tasks where supported
• Context-usage information where available
• Native steering of a running turn

Native steering means a follow-up can be delivered to a live Claude session without first cancelling the active turn. The provider still decides when and how to act on it.

Approve for me (Auto mode) is available only for Codex and Claude Code. For Claude it needs a model that supports Auto mode and Claude Code 2.1.111 or newer.

Claude capabilities change with the installed CLI and account. Treat the controls shown in your current Tejus session as authoritative for that runtime.

Effort levels

Fable models have thinking always on: there is no thinking toggle, no fast mode, and no Ultrathink option. Ultracode runs Extra High effort with workflows. Model access still depends on the Claude account and CLI.

When Anthropic returns a model-scoped weekly allowance, Claude usage shows a separate Fable window. A missing Fable row means the account response did not include that limit; it does not by itself mean the model is unavailable.

Automatic compaction and explicit windows

The Auto-compact picker defaults to Auto (Claude Code). In this mode Tejus leaves the window unpinned so Claude Code can resolve its own model, account, settings, and environment behavior. Choose 200k or 1M when you want an explicit override; returning to Auto clears it. A model's context capacity and its automatic compaction threshold are different values, and an explicit window does not grant an account additional model access.

Cache review before resuming

Before an expensive resumption of a cached conversation, Tejus can hold your message for review. Choose Continue with full context, Compact, then send, or Cancel this send. Compaction itself processes the history once; it is not free.

Artifacts, /design and /slides

To publish Claude Artifacts with /design or /slides, turn on Artifacts, /design and /slides in Claude's Runtime section under Settings → Agent providers → Provider tools, then start a new session. It is off by default and needs a claude.ai login on a Pro, Max, Team, or Enterprise plan. Turning it on does not grant account eligibility.

Multiple Claude accounts

Under Provider tools → Claude, Add account creates another Claude account with its own sign-in folder, and Import directory reuses a Claude config folder you already signed in to. Each account has its own sign-in, status, and terminal command.

You can also import local Claude projects and conversations.

Updating

Tejus detects how Claude Code was installed. Settings → Agent providers → Updates offers Update, which runs claude update for the native installer or npm install -g @anthropic-ai/claude-code@latest for an npm install. To update yourself:

claude update

Restart Tejus after replacing the executable.

Troubleshooting

Claude is not detected

Tejus shows “Claude Agent CLI (claude) is not installed or not on PATH.”

1. Run claude --version from a new PowerShell window.
2. Compare the resolved executable (Get-Command claude) with the custom binary path in Tejus.
3. Restart Tejus after changing PATH or installing Claude.
4. Clear an obsolete custom path rather than pointing Tejus at a wrapper that no longer exists.

Claude is detected but not authenticated

Use Sign in in Settings → Agent providers, or run claude directly and complete the provider's own login flow. For enterprise backends, verify the required cloud credentials and environment variables in the same environment Tejus starts with. The status check can take up to 20 seconds; a timeout does not mean you are signed out.

Models or effort options are missing

Update Claude Code, confirm the account exposes the expected model, and refresh provider status. Start a new task after refreshing so the current model catalog and effort options apply. Model availability is account- and CLI-dependent.

Approve for me is unavailable

Update Claude Code to 2.1.111 or newer and pick a model that supports Auto mode.

A session behaves differently from the Claude terminal

Compare the selected model, permission mode, project settings, local instructions, and executable path. A different Claude installation or configuration source can produce different behavior.

Official documentation

• Set up Claude Code
• Claude Code CLI reference

Continue with Your first task or Provider handoffs.