Skip to main content
Every workflow in SketricGen starts inside an AgentSpace — a visual environment where you connect agents, tools, and logic to build intelligent, multi-agent systems. The following concepts form the foundation of how SketricGen works.

​
AgentSpace

Your AgentSpace is the visual workspace where you design and test multi-agent workflows. Each workflow is composed of nodes and edges that define the logic of collaboration.
  • Nodes = Agents or tools. Each node represents a role or capability.
  • Edges = Handoffs, connections. Define how one agent passes control or data to another and connects to tools.
  • You can version, duplicate, and promote flows as they evolve.
Guideline: Start small. Build two or three connected agents, validate each handoff, and observe the trace before scaling to complex workflows. AgentSpace Canvas showing connected nodes representing agents and tools with edges showing handoffs between them

​
Agents

Agents are the individual roles in your workflow. Each has its own:
  • Instructions (the prompt or system message defining purpose & tone)
  • Model & parameters (OpenAI supported models)
  • Access to tools (data retrieval or action execution)
  • Constraints (enforced via agent instructions and structured schema validation)
Agents can:
  • Call tools to fetch data or perform actions
  • Handoff to other agents to delegate tasks
  • Produce structured outputs such as JSON
Guideline: Define each agent like a job description - what they know, do, and when to delegate. Over-specifying beats ambiguity. Agent configuration interface showing instructions, model settings, tools access, and guardrails

​
Tools (Capabilities)

Tools give agents the power to act, compute, or retrieve knowledge. You can attach one or many tools to an agent.
ToolPurposeTypical Use Case
Sketric App MarketplaceConnect to 2000+ apps via prebuilt connectors and actions.CRM updates, sending emails, project management tasks.
File SearchCreate a private knowledge base from uploaded files or crawled websites.Product manuals, FAQs, policy docs.
Web SearchRetrieve fresh, web-based context.Market research, news summaries, dynamic data.
Code InterpreterExecute Python code securely in a sandbox.Data transformation, quick computation, CSV analysis.
Image GenerationGenerate images using AI models from text descriptions.Marketing visuals, product mockups, social media graphics.
API RequestCall external REST APIs with headers, auth, and schema mapping.Integrate internal systems or external SaaS APIs.
Custom MCPAdd standardized MCP servers for cross-app interoperability.Advanced enterprise integrations.
Guideline: Attach only the tools that each agent truly needs — fewer tools make reasoning more focused and traces easier to interpret. Tools marketplace interface showing available tools including Sketric App Marketplace, File Search, Web Search, Code Interpreter, API Request, and Custom MCP

​
Handoffs (Orchestration)

A handoff defines when and how one agent transfers the task to another. They are the connective logic of your workflow.
  • AI-routed: Agents decide when to hand off based on their instructions and handoff conditions you define on each edge.
  • Forced handoff: Deterministic, designer-controlled routing that guarantees the next agent always runs after the current one completes.
Handoff configuration interface showing AI-routed and forced handoff options

​
Structured Inputs & Outputs (JSON)

Use structured JSON schemas to keep workflows reliable, traceable, and composable.
  • Structured input: The calling agent must follow the defined schema only when it calls this agent (e.g. as a tool or handoff). It is not required when calling other agents or when returning a response directly to the user.
  • Structured output: The agent always returns the defined schema, whether a handoff agent or a human receives the response.
  • Define required and optional fields in the agent node inspector under Advanced Mode → Structured Data.
  • Keep schemas consistent across agents for smoother data flow.
Guideline: Always validate schema changes in the Trace Explorer before publishing—small mismatches can break downstream logic. Edit Agent Advanced Mode with Structured Data — Input Schema and Output Schema Configure Structured Output (JSON) modal with schema name and fields

​
Conversations, Tool Calls & Traces

Every interaction generates data you can inspect and learn from.
  • A Conversation is a complete user interaction flow.
  • Tool Calls show when and how a tool was used.
  • Traces record the sequence of agents, tool inputs/outputs, reasoning artifacts, duration, and credit consumption.
The Trace Explorer helps you visualize each step and pinpoint inefficiencies. Guideline: Review traces often - optimize agents that over-call tools or generate unnecessary handoffs. Trace Explorer interface showing conversation flow, agent interactions, tool calls, durations, and credit consumption

​
Deployment (At a Glance)

Once your workflow is ready, deploy it through any of these paths:
  • Widget: Configure your branding (logo, colors, greeting message) and embed the snippet into your site or CMS.
  • Messaging platforms: Connect to WhatsApp, Telegram, Instagram, or Slack for direct user interaction.
  • Public API & SDKs: Trigger workflows programmatically (Node or Python SDK) and receive structured JSON responses.
(See the Deploy section for integration-specific steps.)

​
Philosophy: Agentic, Not Hard-Coded

SketricGen empowers you to design agentic systems, not brittle automations. You define roles, tools, and logic - SketricGen handles the orchestration, scaling, and traceability. Every workflow you ship becomes a living, improvable system.