What agents are
Drawsy Agents are task-oriented AI runs that operate inside your workspace. Unlike a generic chat thread, an agent starts from a specific canvas region, linked objects, and whatever connectors you have authorized. The output is meant to change or extend the work on the board — not live in a disconnected conversation.
Think of agents as collaborators that can research, structure, and draft — but only with the context you expose. They do not have perfect or infinite memory; they use the current board state and connected sources available at run time.
What agents do
Agents help move work forward when the task needs more than a quick answer: expanding a partial architecture sketch into a fuller diagram, summarizing open questions across a research board, drafting a decision memo from linked sources, or generating next-step structure for a plan.
- Expand diagrams and plans from partial sketches
- Research and synthesize across connected sources
- Summarize decisions, trade-offs, and open questions on the board
- Generate structured outlines, cards, or annotations on the canvas
- Analyze options against constraints visible in the workspace
Context agents use
Each run assembles context from three layers: the canvas objects you select or scope, linked notes and relationships on the board, and snippets from connected sources (GitHub, Notion, Slack, Drive, Gmail, Figma, PostgreSQL, HydraDB).
You control what is in scope. Agents do not silently pull from tools you have not connected. Context quality depends on what you link — better connectors and clearer board structure produce more useful runs.
Canvas interaction
Agents read spatial layout, frames, cards, links, and annotations on the canvas. They can propose additions or edits to those objects rather than only returning text in a side panel. That keeps outcomes visible to the team and preserves the visual memory of how a conclusion was reached.
For best results, use frames to group related work, link cards to sources, and keep decision records on the board instead of only in chat.
Connected sources
Agents become significantly more useful when connectors are active. A Build workflow with GitHub can ground architecture expansion in real modules. A Research workflow with Drive and Notion can synthesize across documents already on the board. See Connectors for the full list and plan access.
Basic connectors are on Free; all connectors unlock on Pro+. Agents respect the same authorization boundaries as Q&A.
Examples
Practical agent runs teams use today:
- Build: expand a service diagram from three boxes into a fuller map with suggested boundaries
- Plan: break a quarterly theme into milestone cards with dependencies
- Research: compare claims across linked sources and list gaps
- Decide: draft a decision record from option columns and linked Slack threads
- Organize: cluster tasks by theme and flag items missing linked context
Limitations
Agents are not autonomous workers with unlimited memory or guaranteed correctness. They can misread ambiguous boards, miss sources you did not connect, and produce drafts that need human review. They work best when the canvas is structured and connectors reflect what the team actually trusts.
Runs are metered by plan: Free includes 3 agent runs per month, Pro includes 15, and Team includes unlimited runs. See https://app.drawsy.tech/pricing for the live matrix.
Agents vs normal chat
Normal chat in Drawsy is for quick questions and back-and-forth clarification — grounded in canvas and sources, but conversational. Agents are for heavier tasks that produce structured output on the board: multi-step research, diagram expansion, or generating a set of cards from partial input.
- Use chat when you need a fast answer or iterative refinement
- Use agents when the task should change canvas structure or synthesize across many sources
- Use agents when the output should be visible to collaborators on the board
- Both are grounded in workspace context — neither replaces judgment or authorization choices
