1Z0-1158 Tests Building Governed Agents on 1Z0-1158-26 — Not Chatbot Prompting
For engineers shipping LLM and agent workloads on OCI. Oracle's certification index now lists 1Z0-1158-26 and no longer lists OCI Generative AI Professional (1Z0-1127), so expect the Responses API, hosted agents and guardrails, not just fine-tuning.
Check Your Readiness →Most candidates understand Oracle Cloud Infrastructure Enterprise AI Professional (1Z0-1158-26) concepts — and still fail. This exam tests how you apply knowledge under pressure.
Section names come from a secondary source and Oracle publishes no weights, so prepare evenly. Many items show a goal plus a fragment of a Responses API request or a cluster configuration and ask which tool type, mode, setting or OCID completes it.
Call ApplyGuardrails and log any violations it reports
ApplyGuardrails only reports; the caller decides, and Inform mode lets requests continue. Configure endpoint guardrails in Block mode and pin the version with guardrailVersionConfig. OCI Audit records API calls but not prompt text.
Upload the files and call file_search directly
Uploads are not searchable automatically. Index them through the Vector Stores API, then attach file_search, which OCI executes. For structured database data, use NL2SQL with GenerateSqlFromNl, which only generates SQL; DBTools MCP Server executes it under database permissions.
Raise max output tokens and keep client-side history
Use the Conversations API and pass the conversation ID so OCI manages state, and enable compaction, whose automatic threshold is at least 1000 tokens. Compaction consumes tokens itself and loses some detail.
Use a public endpoint with no platform-managed authentication
Push the image to OCI Container Registry, create a hosted application with a private endpoint, and register an identity domain application for bearer token or browser session auth. Persist state in managed storage, not container memory.
A custom model is not live until it is hosted on a dedicated hosting cluster with an endpoint, and the endpoint OCID goes in the model field. Training data is JSONL prompt and completion pairs, at least 32, split 80/20 automatically.
On OCI, T-Few is the parameter-efficient method for supported Cohere models and LoRA, with frozen base weights, is for supported Llama models. Imported open-weight models run on a dedicated cluster but are not fine-tuned by OCI.
file_search and code_interpreter run inside OCI, and mcp calls a remote MCP server by URL. A function tool only returns a structured call; your application executes it and sends the result back in a follow-up request.
Dedicated clusters bill in unit-hours, and hosting carries a 744 unit-hour minimum commitment whether or not traffic arrives. Fine-tuning clusters bill per job with a one unit-hour minimum. Light or bursty load usually belongs on on-demand models.
Model access is controlled by IAM policies that allow or exclude model IDs per group. Guardrails inspect content: moderation, prompt injection and English-only PII detection. Use API keys for testing and IAM authentication in production.
Strengthen weak areas with exam-style practice questions and detailed explanations.
Simulate the real exam experience and assess your readiness under timed conditions.
Review key concepts, objectives, and exam topics in one place.
Get personalized explanations, learning recommendations, and instant answers.
Follow a structured learning path designed to help you prepare efficiently.
1Z0-1158 rewards knowing which OCI tool, cluster and control fits. Test your Enterprise AI readiness.