Guides Set it up your way
Agents and models
Each agent runs one of three agent programs, with a model and a thinking level you choose. This guide covers the AI accounts you can use and how to set each agent up.

Agent programs and accounts
| Agent program | Accounts it can use |
|---|---|
| Claude Code | a Claude Pro or Max plan (with a setup-token), an Anthropic API key, or an LLM gateway such as OpenRouter |
| Codex | an OpenAI API key, or a ChatGPT plan |
| OpenCode | an OpenRouter, Ollama Cloud or Synthetic key, an OpenAI or Anthropic key, or another OpenAI-compatible service |
Another OpenAI-compatible service. OpenCode can use any service that speaks the OpenAI API. Add it under Settings > Secrets with the kind Other, named OPENCODE_CONFIG_CONTENT, whose value describes the service. Settings > Funding > What the provider controls shows an example for Ollama on your own machine. (Unverified: this route was not tried end to end for this guide, and the runner's network guard may refuse a service on your home network.)
Through OpenRouter, Claude Code can be pointed at models that are not Claude. Boss Man marks those "(not Claude: may not work)".
Add an account
- Open Settings > Secrets.
- Under Add, choose the kind: Claude setup-token, Anthropic API key, OpenAI API key, OpenRouter API key, Ollama Cloud API key or Synthetic API key.
- Paste the value and save. Boss Man encrypts it and never shows it again.
- Add a funding source under Settings > Funding for it, so Boss Man can track what each agent spends.
For a Claude plan, run claude setup-token on your own computer and paste the token it prints.
When a plan is nearly used up
A plan (a Claude Pro or Max plan, or a Synthetic plan) has a usage allowance. Boss Man calls the pause near the end of it the quota hold. An API key has no hold: its runs just start.
Each plan's card on Settings > Funding has the hold settings, under When the subscription is nearly used up:
- Hold routine work at: a percentage from 80 to 100. The default is 95. At this level Boss Man pauses routine work on the plan and tells you. Chats and reviews keep running.
- I have extra usage (on a Synthetic plan, I have top-up credit): what happens once the plan is at the hold level.
| Choice | Routine work | Urgent tickets |
|---|---|---|
| Hold everything at the limit (the default) | waits | runs until the limit itself, then waits. It also waits if the plan is already using extra usage. |
| Keep urgent work running past the limit (Synthetic: Keep urgent work running on top-up credit) | waits | carries on past the limit, paid from extra usage or top-up credit |
| Keep all work running (Synthetic: Keep all work running on top-up credit) | carries on | carries on |
Chats and reviews always carry on. Use the default if the account has no extra usage or top-up credit. Boss Man can't see your extra-usage balance, so set a spending limit in your Claude account.
Agents on Synthetic's Infinite mode (infinite: models) are never held, because those runs don't spend the plan's credits. Another agent on the same plan with a normal model is still held.
Several Claude plans
If you have more than one Claude plan, Boss Man can move new runs between them.
- Add each plan's setup-token under Settings > Secrets (kind Claude setup-token).
- On Settings > Funding, add a funding source for each plan with the kind Subscription (setup-token) and the provider Anthropic.
- On each plan's card, under Switch between Claude accounts, set Order (1 is used first) and choose its token under Login for this account. Save.
Each new run then goes to:
- the first plan in your order that is under its hold level;
- if none is, the first plan whose extra-usage setting still allows the work. So extra usage is spent only when no plan has room;
- if no plan allows it, nothing: the run waits for the plan that frees up first.
Your first plan is used again as soon as its window resets. Runs already going finish on the plan they started on.
A plan with an empty Order is left out of the list. An agent whose funding source has no order can't switch to the ranked plans, and Boss Man says so when it holds that agent's work. A plan after the first with no login is skipped.
Synthetic
Synthetic (synthetic.new) is a model provider for OpenCode agents. Add its key as Synthetic API key. Its models are named synthetic/<model>. Add a funding source with the kind Subscription (setup-token) and the provider Synthetic.
The Funding page shows how much of the plan's weekly credits are used. Credits come back 2% at a time. Boss Man reads this at most every 5 minutes. You get a notification at 60%, 80% and 92% used.
The quota hold works as in When a plan is nearly used up. If your Synthetic account has paid top-up credit, Synthetic spends it once the weekly credits are used. Choose what happens then under I have top-up credit on the plan's card. With Keep all work running on top-up credit, you get one notification a week when work starts drawing on top-up credit.
Synthetic doesn't tell Boss Man the top-up balance, so Boss Man can't show it or stop before it runs out. Check it at synthetic.new. When it runs out, Synthetic refuses the runs.
Hire an agent
Open Agents and choose Hire. Pick a role under Start from, or Blank to write the job yourself. The roles are:
- most capable model: Senior Developer, Security Reviewer, Adversarial Reviewer, Senior Infrastructure Engineer
- balanced model: Engineer, Researcher, Designer, QA, Writer, Product Manager, Data & Analytics
- fast model: Scrum Master, Assistant, Release Manager, Support
Choose a model
The Model list starts with Default, which uses the agent program's own default. Below that are the models the runner reports for that agent program, and the models your API key offers. Choose Other… to type a model name yourself. If the list is empty, choose Refresh models.
Choose a thinking level
Thinking sets how hard the model thinks before it answers: Default, Low, Medium, High, Extra high or Max. Not every model offers every level. A level the model does not offer is left at the model's default.
Changes the CEO proposes
The CEO can propose new hires and model changes. They come to Decisions and wait for you. For a hire, choose Hire the ticked ones or Not now. For a model change, choose Approve, Edit or Reject.
Something wrong or missing? Tell us on GitHub.