At a glance
Agent Studio lets you create, configure, and evaluate AI agents connected to your ontology and knowledge base. The agent answers your questions in natural language by querying your data, executing SQL or graph queries, and synthesizing a sourced response.
Example: you ask "Which suppliers delivered more than 100 orders this quarter?". The agent identifies the right query, executes it on your data, and responds with the results and their sources.
What you will do
- Create an agent from a business template
- Configure its tools and ontology context
- Run successful and failure test scenarios
- Adjust behavior before production rollout
- Monitor executions in real time via the Hub
- Observe the reasoning path in the Visualizer
Key vocabulary
| Term | Meaning | Example |
|---|---|---|
| Agent | AI program that receives a question, plans actions, and synthesizes a response. | "Supply Chain Analysis" agent |
| Session | A complete agent execution (question → answer). | Session #42 - "Top suppliers" |
| Tool | A capability the agent can use to access data. | SQL query, vector search, graph traversal |
| Iteration | One reasoning cycle (plan → execute → evaluate). | 2 iterations for a complex question |
| Fast Path | Accelerated mode that skips evaluation for simple questions. | Count, direct lookup |
| Hub | Kanban dashboard showing all current and completed sessions. | Columns: In Progress, Action Required, Done |
Recommended paths
Create an agent
Pick a template and define the agent role.
Configure tools
Enable only the capabilities required for your use case.
Test and evaluate
Validate response quality, limits, and failure cases.
Monitoring
Track executions in real time via the Hub.
Visualizer
Observe the agent's reasoning step by step.
Links to other modules
- Ontology: the agent explores your data model to understand your data structure.
- Knowledge Base: the agent searches your imported documents to enrich its answers.
- Workflows: the AI Agent block integrates the agent into automated workflows.
- MCP: the agent is accessible from external AI clients via the MCP protocol.
Expected result
You have a testable, traceable agent aligned with your priority business use case. Verify returned evidence and observed tool calls for every answer.
Start with a minimal toolset, then expand gradually after test validation.
Need help?
Write to us: Support and contact.
What's new
Available capabilities and answer evidence
The Console distinguishes Capability profile, Tools and API, Context and sources, Skill and Verification. The profile limits capabilities; selection chooses useful ones; the platform still enforces permissions. Skills supply reusable instructions without granting rights.
Start with a reading question about CLI-001 and compare the answer with instances. Cited evidence must support the conclusion; a trace does not replace verification. The agent should state missing data and incomplete searches.
Available modules and tools depend on the workspace and environment. See Configure tools and Answer using records and documents.
