Copilot
The in-editor AI assistant — three modes, four depth levels, and context-aware edits with checkpoints.
Copilot is the in-editor assistant for the workflow you have open. For the workspace-wide AI assistant — conversations, knowledge bases, building workflows from scratch by chat — see Chat.
Copilot is the in-editor assistant that helps you build, understand, and improve workflows. It can:
- Explain — answer questions about Scrydon and your current workflow.
- Guide — suggest edits and best practices.
- Plan — draft a multi-step todo list before any edit.
- Edit — apply changes to blocks, connections, and settings when you approve.
Copilot orchestration runs on a Scrydon-managed service (copilot.scrydon.com). For self-hosted deployments:
- Go to scrydon.com → Settings → Copilot and generate a Copilot API key.
- Set
COPILOT_API_KEYin your self-hosted environment. - Make sure
PUBLIC_APP_URLandAUTH_URLresolve from outside your cluster — the Copilot Backend calls back into your cluster during tool execution.
For the architectural detail, see Architecture → Copilot.
Modes
Copilot has three modes. The mode dropdown shows their labels in the UI:
Ask
Q&A mode for explanations, guidance, and suggestions without editing your workflow. Best for "how do I…?" and "why does X happen?" questions.
Build
Build-and-edit mode. Copilot proposes specific edits (add blocks, wire variables, tweak settings) and applies them when you approve. Best for incremental changes to an existing workflow.
Plan
Plan mode drafts a multi-step plan with checklist items before any edit. Useful for larger workflows — you review the plan, then approve execution step by step.
Depth levels
Each mode runs with one of four depth levels — the trade-off is reasoning quality vs. cost and latency.
Fast
Quickest and cheapest. Best for small edits, simple workflows, minor tweaks.
Auto
Balanced speed and reasoning. The recommended default.
Pro
More reasoning for larger workflows and complex edits while staying performant.
Max
Maximum reasoning for deep planning, debugging, and architectural-scale changes.
Context — using @mentions
Copilot can take typed context into the conversation via @mentions. Available kinds:
| Mention | What it attaches |
|---|---|
@workflow | Another workflow's definition |
@current_workflow | The active workflow's state |
@blocks | Specific selected blocks |
@logs | Execution logs |
@knowledge | A knowledge-base entry |
@templates | A workflow template |
@docs | A Scrydon documentation page |
@past_chat | A previous Copilot conversation |
@logs is particularly useful for debugging — attach a failing run's logs and ask "why did this fail?".
Checkpoints
Every Copilot edit creates a checkpoint. From the chat panel you can revert to any previous checkpoint and the canvas snaps back to that state. Useful when an agent edit goes sideways.
Tool confirmations
Build mode tools default to confirmation required — Copilot proposes the edit, you approve. You can opt into auto-allow per session for quick iteration, with the understanding that auto-allowed edits still produce audit events you can review later.
Related
- Architecture → Copilot — how the conversation flows through the cluster.
- Audit logging — every Copilot edit is logged.
- Architecture → Cortex — model routing for in-cluster LLM calls.