Agent Tool
Launch independent sub-agents with their own conversation and tools to handle parallel or isolated tasks within Yukkai.
What is it?
The Agent Tool lets Yukkai spawn a sub-agent — a separate, independent AI assistant that runs in its own conversation with full access to tools. While the main Yukkai conversation continues as normal, a sub-agent can work on a task in parallel, report back its results, and keep the main conversation clean and focused.
Think of it like delegating a task to a colleague: you give them instructions, they go off and do the work using their own tools and context, and they come back with the results.
How it works
When you ask Yukkai to do something that benefits from a dedicated sub-agent, it uses the Agent Tool to:
- Creates a new conversation — The sub-agent gets its own fresh context, separate from your main chat.
- Gives it instructions — A prompt describes exactly what the sub-agent should accomplish.
- Provides tool access — The sub-agent can use the same tools available to the main agent (files, web, bash, etc.).
- Runs in parallel — The sub-agent works independently while your main conversation continues.
- Returns results — When the sub-agent finishes, its output is delivered back to the main conversation.
Key parameters
| Parameter | Required | Description |
|---|---|---|
prompt |
Yes | The instructions telling the sub-agent what to do |
model |
No | Which AI model to use (defaults to the currently active model) |
How to use it
You don't call the Agent Tool directly — Yukkai decides when it's the right approach based on your request. However, you can encourage its use by describing tasks that are clearly independent or parallel in nature.
When Yukkai uses the Agent Tool
- Parallel tasks — When multiple things can be done at the same time (e.g., "research three different topics simultaneously")
- Isolated analysis — When a task would clutter the main conversation with intermediate steps (e.g., "analyze this large codebase and summarize")
- Specialized work — When a task benefits from a fresh context with no distractions from the current chat history
Examples
Example 1: Parallel research
"Research the latest developments in AI, quantum computing, and biotechnology, then give me a summary of each."
Yukkai can launch three sub-agents — one per topic — each doing its own web research independently, then combine the results.
Example 2: Codebase analysis
"Look through my project folder and tell me which files use deprecated APIs."
A sub-agent can explore the file system, read files, search for patterns, and return a clean summary — all without filling your main conversation with hundreds of file reads.
Example 3: Drafting content
"Write a blog post about renewable energy trends in 2024."
A sub-agent can draft, revise, and polish the content on its own, returning only the final result to your chat.
Tips & best practices
- Be specific in your request — The more detailed your instructions, the better the sub-agent can work independently. Vague prompts lead to vague results.
- Use for complex, multi-step tasks — The Agent Tool shines when a task involves many intermediate steps that you don't need to see in your main conversation.
- Combine with the Swarm Tool for large-scale parallelism — If you need many sub-agents working at once (10 or more), Yukkai's Swarm Tool is better suited — it orchestrates dozens of agents with automatic retry, progress tracking, and result aggregation.
- Trust the delegation — Sub-agents have their own tool access and can make decisions within the scope of their prompt. You don't need to micromanage every step.
- Results quality depends on the prompt — A well-crafted prompt yields excellent results. If the output isn't what you expected, try rephrasing your request with more detail or constraints.
Agent Tool vs. Swarm Tool
| Feature | Agent Tool | Swarm Tool |
|---|---|---|
| Number of sub-agents | 1 per call | 10–100+ |
| Best for | Single isolated task | Large-scale parallel work |
| Progress tracking | No | Yes (status polling) |
| Auto-retry on failure | No | Yes (up to 2 retries) |
| Result aggregation | Manual | Automatic |
For most everyday tasks, the Agent Tool is the right choice. For big jobs — like analyzing an entire repository or scraping dozens of web pages — the Swarm Tool is more appropriate.