Agentlas terminal

Runtimes & permissions

Two questions: which AI does the work, and what is it allowed to touch? You decide both, and you can change them mid-chat.

Who answers

Agentlas hosts your existing setup. If you use Claude Code or Codex, agentlas runs it for you and shows it inside the Boston Terrier terminal — your normal sign-in, no extra API key. If you would rather use your own API key (BYOK) or a local model with Ollama, agentlas runs its own agent loop instead.

claude-code

Uses your Claude Code sign-in. Best tools. Hosted inside agentlas.

codex

Uses your Codex sign-in. Hosted inside agentlas.

gemini

Uses your Gemini CLI.

anthropic · openai · google

Your own API key (BYOK). Agentlas runs the agent loop and tools.

ollama

A local model on your machine. Private, offline, free.

Switch the runtime

/runtime codex          # inside a chat
/runtime anthropic      # use your own Claude key
/model claude-sonnet-4-5

# or just for one command:
agentlas seo --runtime codex "..."

Permissions

Permissions decide how far the agent can go. The default is write.

read

Look only — read files, list folders. No changes.

write

Read, plus create and edit files. (default)

full

Everything, including running shell commands.

/permission read        # inside a chat

# or just for one command:
agentlas seo --permission full "..."

HEADS UP · full lets it run commands

full allows shell commands (like installing packages or deleting files). Use it when you trust the task. write is enough for most editing work.

Memory comes along

Every chat reads your project memory first, and quietly saves new durable notes after — the same memory the app uses. Type /memory to see what is being fed in. Work in a folder twice and Agentlas starts keeping that folder's memory automatically.

See where agent files live →

How Agentlas lays out Claude Code and Codex projects.