Bring your own OpenAI key and power your LunoGen agent with GPT models
OpenAI provides the GPT family of large language models, including GPT-5.4, o3-mini, and the DALL-E image generation suite. The LunoGen OpenAI integration lets you bring your own OpenAI API key, pick which model the agent reasons with, generate embeddings for memory recall, run DALL-E image jobs, and monitor your OpenAI spend — all from WhatsApp.
LunoGen ships with a default model, but teams with an OpenAI contract often want to use their own key and models. Bringing your own key gives you cost control, access to fine-tuned models, and the option to route different tasks to different models — GPT-5.4 for reasoning, o3-mini for quick replies, embeddings for memory. This integration puts all of that under your control and adds cost tracking so you never get a surprise bill.
In the OpenAI dashboard, go to API keys → Create new secret key. Name it 'LunoGen Agent' and copy the key.
In LunoGen, open Credentials → New → OpenAI and paste your API key. Choose an organization ID if you use multiple orgs.
Choose the default reasoning model (GPT-5.4 recommended) and a fast model for quick replies (o3-mini recommended).
Turn on embeddings and pick the embedding model (text-embedding-3-large is the default). Memory recall will now use your OpenAI account.
Configure daily and monthly spend caps. LunoGen will pause new requests and ping you on WhatsApp if a cap is reached.
Connect your OpenAI API key and the LunoGen agent uses your account for all its reasoning calls. You keep billing control, access to your fine-tuned models, and your own rate limits.
Switch my agent to use my OpenAI key for everything. I want full control of billing.
Validating your OpenAI API key and checking available models...
Configure different OpenAI models for different kinds of tasks. Use GPT-5.4 for long-form reasoning, o3-mini for quick replies, and a fine-tuned model for classification — all from WhatsApp.
Route fast replies to o3-mini, long reasoning to gpt-5.4, and classification to the fine-tuned classifier
Updating model routing config...
The agent uses OpenAI embeddings to remember things across conversations. Ask it anything you've mentioned before — it finds the most relevant snippet and brings it back with context.
What did I say about the Contoso deal terms back in March?
Searching memory with embeddings...
The agent pulls your OpenAI usage every morning, breaks it down by model and task, and posts a cost dashboard to WhatsApp. You always know what the agent is spending.
Pulling usage and billing from OpenAI...
Upload training examples, kick off a fine-tune, and deploy the resulting model to your agent — all from WhatsApp. The agent handles file upload, job monitoring, and deployment.
Start a fine-tune on o3-mini using the support-examples.jsonl file I uploaded yesterday
Uploading file and creating the fine-tune job...
Ask the agent to generate an image and it calls DALL-E 3 with your OpenAI key, uploads the result to Drive, and sends it to the WhatsApp chat in seconds.
Make me a clean product hero image: a friendly AI assistant on a phone screen, minimal style, warm background colors
Generating with DALL-E 3...
Bring your Anthropic key and power your WhatsApp agent with Claude models
Ship code faster — PR reviews, triage, and release notes run by an AI agent
Run your Notion workspace from WhatsApp — create pages, query databases, and answer questions from your wiki
Turn any WhatsApp message into studio-quality voice with ElevenLabs AI
Spin up an AI agent in under 2 minutes, plug in your OpenAI credentials, and start running your workflows from WhatsApp.