CLI Agents as AI Providers

Intent Architect can use a CLI coding agent you already have as the AI harness behind its chat. The agent runs locally under your own subscription and sign-in, and Intent Architect talks to it over the Agent Client Protocol (ACP), supplying its designers, Software Factory, agents and skills as tools and context.

Tip

To drive Intent Architect from your own terminal or IDE instead, see Intent MCP Server.

Claude Code, Codex and GitHub Copilot CLI connect through an ACP adapter (the connector) that Intent Architect installs with npm on first use, so they also need Node.js.

Installing an agent

Install the agent, sign in from a new terminal and check that it responds before enabling it in Intent Architect.

Claude Code

Setup guide · Requires Node.js

curl -fsSL https://claude.ai/install.sh | bash     # macOS, Linux, WSL
irm https://claude.ai/install.ps1 | iex            # Windows PowerShell

Sign in: run claude and follow the prompts.

Note

Claude Code can also sign in with Amazon Bedrock: choose 3rd-party platform > Amazon Bedrock at the login prompt, or run /setup-bedrock. Keep the Bedrock settings in ~/.claude/settings.json rather than exporting them in your shell, as Intent Architect won't see variables exported in a terminal. See Claude Code on Amazon Bedrock.

Codex

Setup guide · Requires Node.js

npm install -g @openai/codex

Sign in: run codex and choose Sign in with ChatGPT, or set OPENAI_API_KEY.

GitHub Copilot CLI

Setup guide · Requires Node.js and a GitHub Copilot subscription

npm install -g @github/copilot

Sign in: run copilot, then /login.

Note

GitHub Copilot CLI is separate from the GitHub Copilot API provider, which you sign in to from the AI Providers tab.

Kiro

Setup guide

curl -fsSL https://cli.kiro.dev/install | bash     # macOS, Linux
irm 'https://cli.kiro.dev/install.ps1' | iex       # Windows PowerShell

Sign in: the installer opens a browser to sign in. To sign in again, run kiro-cli login.

Cursor CLI

Setup guide

curl https://cursor.com/install -fsS | bash                 # macOS, Linux, WSL
irm 'https://cursor.com/install?win32=true' | iex           # Windows PowerShell

Sign in: run cursor-agent login, or set CURSOR_API_KEY.

OpenCode

Setup guide

curl -fsSL https://opencode.ai/install | bash      # macOS, Linux
npm install -g opencode-ai                         # any platform

Sign in: run opencode auth login and choose the model provider OpenCode should use.

Enabling an agent

Restart Intent Architect - installed agents are only detected at startup.
Open AI Configuration (type /config in the chat) and go to AI Providers.
Check the agent's card shows Installed. Use its power button to enable or disable it.

No credentials are entered here - sign-in is handled by the agent itself. The card's Models section controls which of the agent's models appear in the chat.

Selecting an agent in the chat

Choose a model under the agent's group in the chat's model picker. The first chat with Claude Code, Codex or GitHub Copilot CLI installs its connector, which can take a minute.

Agent settings

The Agent settings chip in the chat composer (or /settings) sets the permission mode and reasoning effort for the conversation.

Agent	Permission modes	Reasoning effort
Claude Code	Ask before edits, Edit automatically, Auto, Bypass	None, Low, Medium, High, Extra High, Max
Codex	Ask before edits, Edit automatically, Bypass	Low, Medium, High, Max
GitHub Copilot CLI	Edit automatically, Bypass	None, Low, Medium, High
Kiro	Ask before edits, Edit automatically, Bypass	Low, Medium, High
Cursor CLI	Ask before edits, Edit automatically	-
OpenCode	Ask before edits, Edit automatically, Bypass	-

Ask before edits - asks for approval before editing files or running commands.
Edit automatically - applies changes without asking.
Auto - Claude Code chooses the mode for each task.
Bypass all permissions - runs everything without approval. Best used in a separate worktree.

When approval is needed, the chat shows Allow, Allow all (switches the conversation to Bypass) or Deny (stops the turn).

What the agent receives

Intent Architect's agent tools, through an MCP server named intent_architect_acp.
The instructions of the selected built-in or custom agent, plus applicable instruction files.
Enabled stdio servers from MCP Servers that are available to Coding.

The first time an agent is used in a repository, the chat offers to Add Intent Architect's built-in skills and rules to that agent's native folders (.claude/skills, .github/skills, .kiro/steering, etc.). Commit them to share them with your team.

Advanced

Connector package - the Claude Code, Codex and GitHub Copilot CLI cards let you pin a different connector version: untick Use default and edit the npm package spec. Takes effect on your next chat.

Executable override - to use a specific executable instead of the one Intent Architect finds, set one of these environment variables to its full path before starting Intent Architect: INTENT_CLAUDE_CODE_ACP, INTENT_CODEX_ACP, INTENT_COPILOT_ACP, INTENT_KIRO_ACP, INTENT_CURSOR_ACP, INTENT_OPENCODE_ACP.

Troubleshooting

Problem	Solution
Card shows Not installed	Restart Intent Architect. Check the executable runs from a new terminal, or set the executable override.
"…needs Node.js, which isn't installed"	Install Node.js and restart Intent Architect.
" is not signed in…"	Sign in again from a terminal, then click Retry. The conversation isn't lost.
"ACP agent process exited unexpectedly…"	Choose Restart ACP host from the conversation menu. History is kept.

The agent's error output is written to Intent Architect's agent log. Use Open Logs Folder on the Home screen, or /diagnostics in the chat to gather a bundle for support.