Agents

vemlor runs coding agents inside ephemeral runner pods (or local processes for dev). Connect an agent in Settings → Models. Each provider offers one or both connect methods: sign in (OAuth with your subscription) or paste an API key.

Claude Code

Anthropic's official CLI. OAuth flow:

1. Click Open Claude Code OAuth in onboarding (or Settings → Agents).
2. Approve on console.anthropic.com.
3. Anthropic shows a code in the form code#state. Paste it back.

Behind the scenes vemlor stores an encrypted refresh token bound to your workspace. Every agent run mints a fresh access token; tokens never touch user-managed env files.

Requirements

• An Anthropic account with Claude Code access.
• A subscription that covers the Claude model your runs use.
• No API key needed with OAuth. The sign-in flow handles it.

Codex CLI

OpenAI's agentic CLI. Same shape: OAuth via chatgpt.com, paste the full callback URL back into vemlor.

Codex's redirect URI is fixed to http://localhost:1455/auth/callback, so the round-trip works transparently when vemlor and your browser run on the same machine. For hosted vemlor users, paste the URL Chrome landed on.

Connect with an API key

Prefer a raw key over OAuth, or using a provider that doesn't offer sign-in? Paste an API key in Settings → Models. The key is stored as a workspace credential, encrypted at rest, and used only inside runner pods.

vemlor never resells tokens. Whichever method you pick, runs bill to your own account.

Choosing per workspace

Each workspace stores its own agent credentials. Mix-and-match across workspaces: one team can run Claude, another Codex, with no shared state. When more than one agent is connected, submit a task with the provider set to auto and vemlor routes it for you.

Roadmap

• Aider (open-source, BYO API key)
• Custom agents via Skills