[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8},"actionData","errors","root",{"_26":27,"_28":29,"_30":12,"_31":12,"_32":12,"_33":12,"_34":12,"_35":12,"_36":37},"routes/canvas.shared.$sharedTextdocId",{"_9":10,"_11":12},"sharedTextdoc",{"_13":14,"_15":-7,"_16":17,"_18":19,"_20":21,"_22":23,"_24":12,"_25":12},"isAuthenticated",false,"sharedTextdocId","6825f96694a48191af7648cad2996158","versionInt","title","Api Usage Examples","type","document","content","# API Usage Examples with GPT-4.1\n\n## Overview\n\nThis guide provides detailed, real-world examples of using the OpenAI GPT-4.1 API effectively, with a focus on instruction-following, tool integration, agent persistence, and prompt structuring. These examples are designed to help developers and engineers build resilient, production-ready systems using GPT-4.1 across various applications, including customer service, bug fixing, document analysis, and data labeling.\n\nEach example illustrates system prompt construction, tool schema definitions, interaction workflows, and failure mitigation strategies.\n\n---\n\n## Example 1: Customer Support Agent with Tool Use\n\n### Objective\nDeploy a GPT-4.1 assistant to handle user questions about policies and account status.\n\n### System Prompt\n```markdown\nYou are a helpful assistant for NewTelco. Your job is to assist users with account and policy information.\n\n# Instructions\n- Always greet the user.\n- Use tools to retrieve account and policy information.\n- If needed data is missing, ask the user before calling tools.\n- Avoid internal knowledge on restricted topics.\n\n# Workflow\n1. Identify intent\n2. Call tools if sufficient input\n3. Message user before and after tool calls\n4. Cite results\n```\n\n### Tools\n```json\n[\n {\n \"name\": \"lookup_policy_document\",\n \"description\": \"Searches internal docs by topic\",\n \"parameters\": {\n \"type\": \"object\",\n \"properties\": {\n \"topic\": {\"type\": \"string\"}\n },\n \"required\": [\"topic\"]\n }\n },\n {\n \"name\": \"get_user_account_info\",\n \"description\": \"Retrieves account details using phone number\",\n \"parameters\": {\n \"type\": \"object\",\n \"properties\": {\n \"phone_number\": {\"type\": \"string\"}\n },\n \"required\": [\"phone_number\"]\n }\n }\n]\n```\n\n### User Input\n```text\nHow much will it cost for international service in France?\n```\n\n### Assistant Output\n```text\nHi, you've reached NewTelco, how can I help you? 🌍✈️\n\nYou're asking about international service while traveling to France. Let me check that for you — one moment, please. 🕑\n```\n\n### Tool Call\n```json\n{\n \"name\": \"lookup_policy_document\",\n \"arguments\": {\"topic\": \"international service cost France\"}\n}\n```\n\n---\n\n## Example 2: Software Bug Fix Agent with Diff Application\n\n### Objective\nUse GPT-4.1 to investigate and fix a bug in a codebase using the `apply_patch` tool.\n\n### System Prompt Highlights\n```markdown\n# Objective\nFix the issue provided by the user.\n\n# Instructions\n- Plan each step\n- Reflect after each function call\n- Never guess code — read it first using tools\n- Only stop when all tests pass\n\n# Workflow\n1. Understand issue deeply\n2. Investigate codebase\n3. Draft patch\n4. Apply patch\n5. Run tests\n6. Reflect and finalize\n```\n\n### Tool Definition\n```json\n{\n \"name\": \"python\",\n \"description\": \"Execute code or apply a patch\",\n \"parameters\": {\n \"type\": \"object\",\n \"properties\": {\n \"input\": {\"type\": \"string\"}\n },\n \"required\": [\"input\"]\n }\n}\n```\n\n### Tool Call Example\n```bash\n%%bash\napply_patch <<\"EOF\"\n*** Begin Patch\n*** Update File: src/core.py\n@@ def is_valid():\n- return False\n+ return True\n*** End Patch\nEOF\n```\n\n### Test Execution\n```json\n{\n \"name\": \"python\",\n \"arguments\": {\"input\": \"!python3 run_tests.py\"}\n}\n```\n\n---\n\n## Example 3: Long-Context Document Analyzer\n\n### Objective\nSummarize and extract insights from up to 1M tokens of context.\n\n### Prompt Sections\n```markdown\n# Instructions\n- Process documents in 10k token blocks\n- Reflect after each segment\n- Label relevance and extract core ideas\n\n# Strategy\n1. Read → summarize\n2. Score relevance\n3. Synthesize into unified output\n```\n\n### Input Format\n```xml\n\n

Changes to international billing rules

\n...\n\n```\n\n### Assistant Behavior\n- Chunk input into 10k token sections\n- After each, provide a summary and document scores\n- Compile findings at end\n\n---\n\n## Example 4: Data Labeling Assistant\n\n### Objective\nAssist with structured classification tasks.\n\n### Prompt Template\n```markdown\n# Instructions\n- Label each entry using the provided schema\n- Do not guess; if unsure, flag for human\n\n# Labeling Categories\n- Urgent\n- Normal\n- Spam\n\n# Output Format\n{\"text\": ..., \"label\": ...}\n\n# Example\n{\"text\": \"Win money now!\", \"label\": \"Spam\"}\n```\n\n### User Input\n```json\n[\n \"New system update available\",\n \"Limited time offer! Click now\",\n \"Server crashed, need help ASAP\"\n]\n```\n\n### Assistant Output\n```json\n[\n {\"text\": \"New system update available\", \"label\": \"Normal\"},\n {\"text\": \"Limited time offer! Click now\", \"label\": \"Spam\"},\n {\"text\": \"Server crashed, need help ASAP\", \"label\": \"Urgent\"}\n]\n```\n\n---\n\n## Example 5: Chain-of-Thought for Multi-Hop Reasoning\n\n### Objective\nSupport a planning task by explicitly breaking down the steps.\n\n### Prompt Template\n```markdown\n# Instructions\nFirst, think carefully step by step. Then output the result.\n\n# Reasoning Strategy\n1. Identify user question\n2. Extract context\n3. Connect information across documents\n4. Output answer\n```\n\n### Example Input\n```markdown\n# User Question\nHow did the billing policy change after 2022?\n\n# Context\n...\n...\n```\n\n### Model Output\n```text\nStep 1: Identify relevant documents → IDs 10, 12\nStep 2: Compare clauses\nStep 3: 2022 had flat rates, 2023 added time-of-use billing\nAnswer: Billing policy changed to time-based pricing in 2023.\n```\n\n---\n\n## General Prompt Formatting Guidelines\n\n### Preferred Structure\n```markdown\n# Role\n# Instructions\n# Workflow (optional)\n# Reasoning Strategy (optional)\n# Output Format\n# Examples (optional)\n```\n\n### Tool Use Reminders\n- Only call tools when sufficient information is available\n- Always notify the user before and after calls\n- Use example-triggered calls for teaching tool behavior\n\n### Output Patterns\n- JSON or markdown preferred\n- Cite source documents if used\n- Include fallback responses if uncertain (e.g., \"Insufficient context\")\n\n---\n\n## Best Practices Summary\n\n| Element | Best Practice |\n|--------|---------------|\n| Tool Calls | Always define schema with strong param names |\n| Planning | Enforce pre- and post-action reflection |\n| Output | Enforce format, validate JSON before response |\n| Long Context | Use structured delimiters (Markdown, XML) |\n| Labeling | Use few-shot examples and explicit categories |\n| Diff Format | Use V4A patch format for code updates |\n\n---\n\n## Final Note\n\nThese examples are starting templates. Each system will benefit from iterative refinements, structured logging, and real-world user testing. Maintain modular prompts and tool schemas, and adopt evaluation frameworks to monitor performance over time.\n\n**Clarity, structure, and instruction adherence are the cornerstones of production-grade GPT-4.1 API design.**\n\n","access","public","isModerationBlocked","isAnonifyApiKeyDetected","dd",{"_38":39,"_40":41},"disablePrefetch",true,"shouldPrefetchAccount","shouldPrefetchUser","shouldPrefetchSystemHints","promoteCss","disableSSR","stripModulepreloadImports","statsigGateEvaluationsPromise",["P",37],"traceId","3329080036259818906","traceTime",1779976745272]