Skip to main content

​
What is a subagent?

A subagent is a child agent that Ana spawns to accomplish one focused objective. It runs as its own thread, with its own context window and its own Python sandbox, and returns a tight, self-contained answer (plus any files it produces) to the agent that started it. Subagents exist to keep a conversation’s reasoning clean. Deep, verbose work — exploring a connector schema, analyzing one entity among many, reading across several sources — produces a lot of intermediate detail. When that work happens in a subagent, only the conclusion comes back to the parent; the noisy exploration stays in the subagent’s own thread.

​
When Ana uses them

Ana reaches for a subagent when a sub-task is either verbose or independent:
  • Deep dives — investigate a single data source or question in isolation and report back just the findings.
  • Parallel fan-out — split work that has independent pieces (for example, the same analysis across many regions or accounts) and run them at the same time.
  • Specialized helpers — hand a narrow, well-scoped job to a purpose-built agent.
Subagents can themselves spawn subagents, up to a fixed recursion depth, so a large task can be decomposed into a tree of focused work.

​
Following along

Each subagent is a real thread, so you can watch its progress:
  • The subagent step in the transcript expands to show the child’s live activity inline.
  • A subagents indicator collects every subagent in the conversation, with running status and a link to open each one as its own thread.
  • Because each subagent is a normal thread, you can open it directly to see everything it did.

​
Results and files

A subagent returns its final answer to the parent. Any files it produces are handed back as artifacts you can open. If the parent needs to keep working with those files directly — for example, to run further analysis on them in Python — Ana can pull them into its own sandbox. Each subagent has its own sandbox, so subagents do not share in-memory state; they communicate through their returned answer and handed-back files.

​
Agents as subagents

In addition to spinning up ad-hoc subagents, Ana can invoke one of your saved agents as a subagent. An agent is eligible only when it is marked callable as a subagent, and you can restrict which members or roles are allowed to invoke it. These controls live in the agent editor.

​
Availability

Subagents are enabled per organization by TextQL. Because each subagent provisions its own sandbox, they consume compute the same way any other workload does — see Pricing for how sandbox usage is billed.