D
DevcoreAI

Getting Started

Tasks

A task is a self-contained agent session — your conversation, code changes, and decisions all captured in one place.

What is a Task?

Every time you send a message to DevcoreAI, you're working inside a task. Each task:

  • Starts with your prompt and builds context as the conversation grows
  • Has a unique ID and dedicated storage directory on disk
  • Contains the full conversation history with all agent messages
  • Tracks token usage, API cost, and execution time per request
  • Can be interrupted and resumed across editor sessions
  • Creates automatic checkpoints via Git snapshots for every file change

Scoping Your Tasks

Each model has a context window — a limit on how much text it can hold in memory at once. As a task grows (more messages, more file content), it consumes more of that window. When the context fills up the agent either stops or needs to compact the history.

Keep tasks focused on a single feature or bug. For large changes, break work into smaller sequential tasks rather than one massive session.

Good task scope

"Add JWT authentication to the Express app"

Too broad

"Build the entire backend from scratch"

Good task scope

"Fix the 404 error on the /users/:id endpoint"

Too broad

"Refactor all files in the project to use TypeScript"

Context Window & Auto Compact

DevcoreAI shows a context-usage bar at the top of each task. When the conversation approaches the model's limit, Auto Compactkicks in automatically: it creates a comprehensive summary of everything that's happened, preserving all technical decisions, code changes, and file context — then continues from where it left off.

You can also trigger compaction manually with /smol before hitting the limit.

New Task vs. Continue

Start a new task

/newtask

The previous goal is done, or you're starting something unrelated. Fresh context = cheaper API calls.

Continue the same task

just keep typing

You're building on what was just done, iterating on the same feature, or debugging the same bug.

Task History

Every task is saved automatically. Open the History panel from the DevcoreAI sidebar to browse, search, and resume any previous task. The history panel shows:

  • Task title (derived from your first message)
  • Date and time started
  • Total token usage and API cost
  • Number of files changed

Resuming Tasks

Click any task in the History panel to resume it. The full conversation is restored and the agent picks up exactly where it left off — including all file context and previous decisions.

Resuming a task that was interrupted mid-edit may require reviewing any open diff views. Use checkpoints to restore the workspace to a known-good state first.