# OpenAI

This document explains how to configure OpenAI as the LLM provider for yCrash AI chat.

Note

After you create ycbuddy-config.json, place it in the upload directory and restart yCrash as described in the Introduction.

Use this provider if your organization standardizes on OpenAI instead of Gemini.

# Step 1: Get Your OpenAI API Key

  1. In a web browser, go to platform.openai.com (opens new window)
  2. Sign in with your OpenAI account
  3. Open API keys
  4. Create a new secret key and copy it when it is displayed

Store the key in a secrets manager or restricted server directory. Treat it like a password - do not email it, commit it to Git, or share it in tickets.

# Step 2: Add the OpenAI Settings

Paste the following JSON into ycbuddy-config.json and replace your-openai-api-key with the key you copied from OpenAI:

{
  "llmProvider": "openai",
  "llmApiKey": "your-openai-api-key",
  "llmModelName": "gpt-5.6-terra",
  "llmEmbeddingModelName": "text-embedding-3-small",
  "llmMaxOutputTokens": 4096,
  "llmReasoningEffort": "none"
}
1
2
3
4
5
6
7
8

# What each field means

Field What it is What to put
llmProvider Which AI company yCrash should call "openai"
llmApiKey Your secret key from OpenAI The key from Step 1
llmModelName The model that writes chat answers "gpt-5.6-terra" (recommended)
llmEmbeddingModelName Helper model for finding relevant context "text-embedding-3-small"
llmMaxOutputTokens Soft cap on completion tokens 4096 (recommended default)
llmReasoningEffort Reasoning depth for GPT-5.x / o-series "none" for chat latency

You can omit llmApiKey and set LLM_API_KEY on the yCrash process instead. All other fields are still required in the JSON file.

If llmModelName is omitted, the OpenAI client falls back to gpt-5.6-terra.

# Supported OpenAI Chat Models

yCrash recognizes these OpenAI model IDs (use the exact string in llmModelName). Unknown IDs still work against the API but use conservative token limits.

Model ID Context Best for
gpt-5.6-terra 1.05M Recommended - balanced intelligence and cost for RCA chat
gpt-5.6-sol / gpt-5.6 1.05M Flagship reasoning/coding; use for hardest analyses (gpt-5.6 aliases Sol)
gpt-5.6-luna 1.05M Cost-sensitive / high-volume workloads
gpt-5.5 1.05M Prior frontier flagship
gpt-5.4 1.05M Prior frontier; strong coding/professional work
gpt-5.4-mini 400K Faster/cheaper GPT-5.4-class
gpt-5.4-nano 400K Cheapest GPT-5.4-class for simple tasks
gpt-5.2 / gpt-5.2-pro 400K Earlier GPT-5.2 family
gpt-5.1 400K Earlier GPT-5.1
gpt-5-mini / gpt-5-nano 400K Earlier mini/nano tiers
gpt-4.1 / gpt-4.1-mini ~1.05M GPT-4.1 family
gpt-4o / gpt-4o-mini 128K GPT-4o family

# Optional OpenAI Tuning

Field Default Notes
llmMaxOutputTokens 4096 Soft cap on completion tokens (max_completion_tokens / max_tokens). For GPT-5.x this budget includes reasoning tokens; values much below ~2k often truncate or empty answers. Raise (e.g. 8192) for longer RCAs. Hard ceiling: 128000.
llmReasoningEffort "none" Sent for GPT-5.x / o-series only. Values: none, minimal, low, medium, high, xhigh, max. Use none/minimal for chat latency; medium/high when analysis depth matters more.
maxAllowedTokens 20000 Caps total prompt context for OpenAI/Azure (TPM/account limits). Independent of the model's 1.05M window.

# Step 3: Place the File and Restart

Place ycbuddy-config.json in the yCrash upload directory and restart the server. See the Introduction for file location, restart, and verification steps.

# Troubleshooting

Symptom Likely cause
Log: Unsupported LLM provider Typo in llmProvider (must be exactly openai)
Chat opens but returns auth / 401 errors Invalid, expired, or revoked API key
Empty or truncated answers on GPT-5.x llmMaxOutputTokens too low (raise toward 8192); or llmReasoningEffort consuming the token budget

Also confirm:

  • llmProvider is "openai" (lowercase)
  • Outbound HTTPS from the yCrash server to OpenAI is allowed (api.openai.com)

For shared file-placement and restart checks, see the Introduction troubleshooting checklist.

← Azure OpenAI →