Agents & parallel work
Syntic Code lets you go beyond a single conversation with Amara. When a task is large, repetitive, or naturally splits into independent pieces, you can spin up specialized agents, watch them work, and coordinate several of them at once. This section covers everything from defining a focused subagent to orchestrating whole teams that collaborate on a shared goal.
An agent in Syntic Code is an instance of Amara with its own context window, its own tool permissions, and a scoped assignment. Because each agent keeps its own conversation history, agents avoid the context bloat that comes from cramming an entire project into one session. You hand off a task, the agent works it, and only the result comes back to you.
What you can build
- Custom subagents — reusable specialists defined in
.syntic/agents/, each with a tailored system prompt, tool allowlist, and model. - Agent view — a live monitor for every agent currently running, so you can inspect progress, pause, or stop work.
- Agent teams — multiple agents that share results and divide labor on a single objective.
- Dynamic workflows — multi-step orchestration where one agent’s output feeds the next.
- Worktrees — isolated git checkouts that let parallel agents edit code without stepping on each other.
When to reach for agents
Use a single Amara session for exploratory or conversational work. Reach for agents when the task is well-defined and separable: auditing a codebase, migrating many files, running research in parallel, or applying a repeatable review across pull requests. The rule of thumb is simple — if you can describe the job as a clear instruction with a clear deliverable, an agent can carry it.
Read next
Start with custom subagents to define your first specialist, then explore agent teams and dynamic workflows to combine them.