Agents and accounts

Threadrun runs three agents: Claude Code, Codex and OpenCode. This page shows you how to sign in to each one, and how to choose its model and settings for a thread.

Choose an agent

You pick the agent for each thread under Harness in the composer.

Sign in

Open Settings → Accounts. Each agent is listed with its status: No account needed, Connected or Not connected. Press Manage beside an agent to work on its account.

Claude Code and Codex

1. Press Log in with Claude (Claude Code) or Log in with ChatGPT (Codex).
2. Follow the steps in the app. The app opens the sign-in page. Depending on the agent, you either enter a one-time code on that page, or paste a code back into Threadrun.

To use an API key instead, paste it under API key and press Save key.

Press Log out next to the Connected badge to remove that agent's login from this app.

OpenCode

OpenCode works with several providers. Press Connect beside one to sign in:

• GitHub Copilot
• ChatGPT Plus/Pro
• OpenCode Zen
• Anthropic API key

Press Reconnect to sign in again. If you already signed in to OpenCode on this Mac, press Import my OpenCode login from this Mac instead.

Choose model, effort, mode and context

The composer shows these settings for the agent you picked:

• Model: the models the agent offers. A long list opens a searchable picker. Type words in any order, such as claude sonnet, or free to see the free models.
• Effort: how hard the model works on each reply. The levels depend on the model. If the model has none, the composer says Not offered by this model.
• Mode: the agent's own working modes. Your governance rules still apply, and they can forbid some modes. See Governance.
• Context (Claude Code only): Standard context or 1M context. The 1M option uses the agent's larger-context model.

The defaults are:

• the agent's own default model;
• Standard context;
• an effort level of medium when the model offers it. Otherwise Threadrun picks high, or the model's first level. A level you choose stays.

You can change any of these during a thread. The change applies to the running agent straight away. Threadrun saves each thread's settings, so a thread that wakes up later keeps the same model and effort.

Compare models

The model picker helps you choose. Next to each model it shows:

• Intelligence, as one to five dots: how the model ranks against the others on the LMArena leaderboard, where people compare answers side by side. Five dots is the top fifth. No dots means the model isn't ranked yet.
• Cost per million input and output tokens, such as $3 / $15 per 1M, or Free.
• Context: how much the model can read at once, such as 200K or 1M tokens.
• What it can do: Reasoning, Tools, Vision and Open weights.
• For free models, their limits. OpenRouter's free models allow 20 requests a minute and 50 a day, or 1,000 a day after you buy $10 of credits. OpenCode doesn't publish the limits of its free models; when you reach one, the thread says Free usage limit reached. Pick another model or connect a provider.

Filters under the search box narrow the list: Free, Smartest (four or five dots), ≤ $1 / 1M in, 1M+ context, Reasoning, Tools, Vision and Open weights. Each shows how many models it would keep; turn on several to combine them, and click one again to turn it off. For example, Smartest with ≤ $1 / 1M in finds strong models that cost little.

The same facts appear under the composer for the model you picked. The data comes from models.dev and the LMArena leaderboard, refreshed daily. Prices are the providers' list prices; your plan or subscription may differ.

Where your logins are kept

• Logins are stored in your computer's keychain: the macOS Keychain, or Windows Credential Manager.
• A login is sent only to the sandboxes of the threads that use it.
• The sign-in steps run in a throwaway sandbox, not on your computer.

For GitHub, which agents use to clone and push code, see Review and ship.