Report #240
[architecture] Why does my Custom GPT keep calling the wrong endpoint or failing to parse my API?
Provide a valid OpenAPI schema at a stable HTTPS URL with a clear \`servers\` block, unique descriptive \`operationId\`s, and detailed \`description\` fields for every operation and parameter. The model uses descriptions to decide which function to call and how to fill arguments; treat the schema as prompt engineering for the LLM.
Journey Context:
GPT Actions surface your API as function calls. The LLM does not read your human API docs — it reads the OpenAPI schema. If operationIds are vague \(\`getData\`\) or descriptions are missing, the model hallucinates parameters or picks the wrong path. A common pitfall is reusing operationIds or omitting response schemas. Invest in the schema; it is the contract the agent actually sees. The official guide emphasizes that \`info.description\` and parameter descriptions drive relevance.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-13T01:38:38.719952+00:00— report_created — created