Profiles
Named agent configurations you can switch between instantly — each with its own soul, skills, and tool set.
What are profiles
A profile is a named snapshot of all agent settings: Soul (persona), enabled skills, memory context, AI provider, tool permissions, and custom system instructions. Switching profiles changes the agent's entire configuration in one action.
This lets you maintain completely different agent personalities for different contexts without maintaining separate asyncat installations. A "code reviewer" profile that is terse and critical, a "research assistant" that is thorough and source-driven, a "DevOps bot" that talks in commands — all running on the same backend, switchable mid-session.
Built-in profiles
| Profile | Description |
|---|---|
| Developer | Technical, code-focused, verbose explanations of implementation details. Prefers dry, precise language. Loads all dev-related skills by default. |
| Researcher | Thorough, source-driven, cautious about claims. Always cites sources and flags uncertainty. Loads research and summarization skills. |
| Assistant | Balanced, helpful, action-oriented. Resolves tasks with minimal back-and-forth. The default profile. |
| Partner | Casual, collaborative, ideates openly. Thinks out loud and builds on your ideas. Good for brainstorming sessions. |
Creating profiles
Create custom profiles in Settings → Profiles → New Profile. Each profile can configure:
- Soul (name, tone, verbosity, proactivity, values, system prompt)
- Enabled skills — which of the 49 bundled skills to load
- Tool permissions — which tool categories to allow
- AI provider and model — use a different model per profile
- Memory context — whether to share memory with the default profile or keep it isolated
Delegation
The agent can delegate tasks to other profiles mid-run using the delegate_to_profile tool. This enables multi-agent workflows within a single asyncat instance: a general-purpose agent can hand off a code review task to the "Developer" profile, wait for the result, and incorporate the feedback — all autonomously.
delegate_to_profile(
profile: "Developer",
task: "Review the diff in sandbox-42 for security issues"
)