How the canvas works
Each node represents a component of your agent. The orchestrator receives every user message first, decides what needs to happen, then delegates tasks to its subagents. Results flow back up to the orchestrator, which assembles the final response.Orchestrator settings
The orchestrator is the “brain” of your agent. Click the Orchestrator node to configure it.System prompt
The system prompt tells the orchestrator who it is, what it does, and how it should behave. Write this in plain language — the model interprets it at the start of every conversation. A good system prompt includes:- The agent’s role and purpose
- Which markets, brands, or data domains it covers
- Key metrics and how they’re defined
- Any formatting preferences (e.g. “always include a summary row”, “use tables for comparisons”)
- Escalation rules (e.g. “if ROAS drops below 2, flag it as a concern”)
You are a paid search analyst for the Dutch and Belgian markets. Your job is to analyse Google Ads performance using the BigQuery connector. Always present campaign data with dimensions by market, campaign, and time period. Define ROAS as revenue divided by ad spend. Highlight campaigns where CPA exceeds the target set in the campaign name.
Context
Three checkboxes control what background knowledge the orchestrator has access to:Model override
By default, the orchestrator uses the model selected in the General tab. To use a different model for this component specifically, select one from the Model override dropdown. The orchestrator handles coordination and reasoning, so a capable model like Claude Sonnet works well here.Subagents
Subagents are specialists that the orchestrator calls to perform specific tasks. The subagents available to you depend on your agent type.Database Query Agent
What it does: Generates and runs SQL queries against your BigQuery connector.When to use it: Any time your agent needs to retrieve data from your warehouse.
Data Analysis Agent
What it does: Interprets query results — computing summaries, identifying trends, formatting charts.When to use it: When you want the agent to reason about the data, not just return raw rows.
Configuring a subagent
Click a subagent node on the canvas to open its settings panel.System prompt
Like the orchestrator, each subagent has its own system prompt. For most agents, you can leave this blank — the subagent’s built-in instructions are already optimised. Override only if you have specific requirements, such as:- Restricting which tables the Database Query Agent should query
- Telling the Data Analysis Agent to always format numbers in a specific currency
Context
The same three context checkboxes described above — global context, organization context, and organization events — are available per subagent, each with its own default. Enable them to give the subagent the same background knowledge as the orchestrator, or configure it differently if the subagent needs a narrower scope.Model override
You can set a different model per subagent. Common patterns:Tools
Tools extend what a subagent can produce. Each subagent has a set of tools you can enable or disable. Enabled tools appear as options the agent can use when answering a question.Data Analysis Agent — available tools
Database Query Agent — available tools (Multi-Agent Swarm only)
In the Data Analysis Agent type, the Database Query Agent’s tools are fixed — it always runs metadata lookup, data exploration, and SQL queries internally. Tool selection only applies to the orchestrator in this mode.