Context and tasks

Decopilot organizes work as tasks (your conversations) and subtasks (focused side-conversations that branch off). Understanding how they share memory helps you get more done in a single session.

What is a task?

A task is your conversation with decopilot — where you chat, decopilot uses tools, and work gets done. Every task runs in a scope (organization or agent — see Overview) that determines which instructions and tools are loaded.

Every task has one of four states:

In progress — the agent is working or waiting for your next message
Requires action — paused, waiting for your input (a tool needs approval, or it asked you a question)
Completed — work finished successfully
Failed — something went wrong, or the task timed out

Tasks time out after 30 minutes of inactivity. If you’re working on something longer, send a message periodically to keep it active.

How memory fills up

Every AI model has a limit on how much it can hold in memory at once — your instructions, the tools available, and the full conversation history all count toward that limit. The more you’ve discussed, the more memory it takes up.

When a task starts, decopilot loads:

Its core instructions and behavior
Guidelines from your current scope
The tools available in that scope
Your conversation history so far

As the conversation grows, the history takes up more space. When it gets too full, decopilot automatically summarizes older parts and keeps the summary instead — freeing up space without losing the important context.

The 40/80 rule

Below 40% full — plenty of room, no action needed
40–80% full — still working fine, but worth wrapping up the current phase
Above 80% full — decopilot automatically compacts the conversation history

You can trigger this manually with the /compact command — useful before starting a new phase of complex work.

Use /compact when you’re finishing one major task and starting another. It keeps the conversation lean without losing what matters.

Subtasks

A subtask is a separate conversation that branches off from your main task to handle focused work. The key difference: it starts with a clean slate — no history from your main conversation.

The subtask runs, does the work, and returns only a short summary to your main task. Full details are saved in task history if you need them.

Main task vs. subtask

	Main task	Subtask
Conversation history	Accumulates over time	Starts fresh
Can ask you questions	Yes	No
Can create subtasks	Yes	No
Returns to parent	N/A	Summary only

Subtasks can’t ask you questions and can’t create additional subtasks — they run autonomously and report back.

When subtasks help

Research and analysis — dig into data, explore options, or investigate an issue. You get back the conclusion, not every step.

Complex calculations — anything that generates verbose output belongs in a subtask. You see the result, not the working.

Quality checks — audits, validation, reviews. Run them separately and get back a summary.

Parallel work — if two pieces of work don’t depend on each other, run them as separate subtasks at the same time.

Using specialist agents in subtasks

You can run a subtask with a specific agent — one configured for a particular domain. The agent brings its own focused toolset and instructions without affecting your main conversation.

Example: Your main task is coordinating order fulfillment. You start subtasks using an Inventory Agent to check stock, a Shipping Agent to plan logistics, and a Customer Service Agent to prepare support notes. Each runs independently; your main task gets the summaries and coordinates the outcome.

Practical tips

Watch the usage indicator — when you’re approaching 40%, think about wrapping up the current phase or using /compact .

Use subtasks for heavy work — research, data analysis, code reviews, and anything that generates long outputs should happen in subtasks. Keep summaries in your main conversation.

Be selective with tools — decopilot loads tools into memory. Don’t enable tools you don’t need for the current task.

Run things in parallel — if two subtasks don’t depend on each other, start them at the same time.

Write clear prompts — “Analyze Q4 apparel demand trends” is more useful than “Do analysis” when reviewing task history later.

Next: See Tools for the full tool reference, or Agents for specialist agents you can use in subtasks.