Core Workflows
Auto Compact
Automatically summarizes conversation history when approaching context limits — so long-running tasks never hit a wall.
How It Works
Every model has a context window limit — the maximum amount of text it can hold at once. As a task grows (more messages, more file content, more tool outputs), the conversation fills that window. When DevcoreAI detects the window is getting full, Auto Compact fires automatically:
Creates a comprehensive summary
Everything that's happened: technical decisions, file changes, patterns discovered, problems solved.
Preserves all technical context
Code changes, file paths, variable names, error messages — nothing important is lost.
Replaces conversation history
The long history is swapped out for the compact summary, freeing up context space.
Continues seamlessly
The agent picks up exactly where it left off, with full awareness of prior work.
Why This Matters
Without Auto Compact, long tasks would simply stop when the context fills — forcing you to start over or manually manage what context to keep. With Auto Compact, you can work on large projects continuously without interruption:
Technical decisions preserved
Architecture choices, naming conventions, and agreed patterns carry forward.
File changes intact
The agent remembers every file it has read or edited in the task.
No restart required
Compaction happens mid-task. You don't need to re-explain anything.
Works on very large projects
Tasks that would otherwise hit limits in minutes can run for hours.
Manual Trigger
You can trigger compaction at any time before hitting the limit using the /smolcommand. This is useful if you've been exploring a lot of files and want to clear noise before starting the main implementation.
PreCompact Hook
If you want to run custom logic before compaction (e.g. save the full conversation to disk, archive decisions to a file), use the PreCompact hook. See Hooks for setup instructions.
Restoring Context After Compaction
Because Auto Compact replaces history with a summary, individual earlier messages are no longer visible after compaction. If you need to review the full history, use Checkpoints to restore the task to a prior state — checkpoints capture the full conversation and file state at each saved point.