agent-creation

Building a Cognigy AI Agent

Prerequisites

• A project (use list_resources { resourceType: "project" } to find one, or let create_ai_agent create one)
• A working LLM resource in the target project before testing the agent

Steps

1. list_resources { resourceType: "project" } — inspect all projects
2. Choose the target project
   • If it already exists: use its projectId
   • If the user wants a new project: call create_ai_agent without projectId, keep the returned projectId, and if llmStatus is "unknown" continue with the LLM reuse steps below before testing the agent
3. Run list_resources { resourceType: "llm_model", projectId } — check whether the target project already has a reusable LLM with a non-empty connectionId
4. If the target project has no reusable LLM, check the other projects for reusable LLMs
   • Run list_resources { resourceType: "llm_model", projectId } for each other project
   • Choose only candidates whose llm_model has a non-empty connectionId
   • Do not treat an LLM without connectionId as a valid reuse candidate
5. If another project already has a reusable LLM, transfer the required model set together with its connection resource(s):
   • manage_packages { operation: "list_exportable", projectId: "" }
   • If this workflow will use knowledge, identify the source project's exact Knowledge Search model and embedding model before exporting anything
   • manage_packages { operation: "export", projectId: "", resourceIds: ["", "", ""], name: "llm-transfer" }
   • manage_packages { operation: "upload_and_inspect", projectId: "", filePath: "" }
   • manage_packages { operation: "import", projectId: "", packageId: "" }
   • list_resources { resourceType: "llm_model", projectId: "" } — verify the required models are present before testing or continuing setup
6. Only if no reusable LLM with connectionId exists, or the package transfer failed: setup_llm { projectId, provider: "openAI", modelType: "gpt-4o", apiKey }
   • With isDefault: true (the default), agents auto-use this LLM — no extra step needed.
   • Save the referenceId from the response if you need to assign it explicitly later. (See the llm-providers skill for valid provider/modelType values)
7. create_ai_agent { projectId, name, description } — Returns: agent, flow, endpoint, endpointUrl — The default LLM is automatically assigned to the agent's job node. If llmStatus is "configured", no extra step is needed.
8. (Only if llmStatus is "unknown" after create) Do not test yet. First repeat steps 4-6 for the returned projectId, then assign LLM to agent if needed: update_ai_agent { aiAgentId, jobConfig: { llmProviderReferenceId: "<referenceId from step 4 or 5>" } }
9. Only after the project has a confirmed working LLM: talk_to_agent { endpointUrl, message }
10. Refine the agent using update_ai_agent — see "All configuration fields" below
11. Repeat 9-10 until satisfied