OpenAI

OpenAI's GPT models reach Memtro through the Responses API with an OpenAI API key. Add the key once and they answer everywhere Memtro calls a model: chat, agent mode, routing ladders, memory extraction and jobs. The same key can also embed your organisation's memories for retrieval.

Add the key

Create the key on the OpenAI platform at platform.openai.com/api-keys. OpenAI's quickstart says a free test request is included and that you add credits on the billing page to keep going.

On Dashboard → Providers pick OpenAI, paste the key and choose who it is for: Personal (only you), Shared with your organisation, or one team. Which of these you are offered depends on your permissions. Work done for the organisation or a team, such as its jobs and its connections' syncs, spends only shared and team keys, never a member's personal one; your own personal jobs and personal connections can use your personal key.

Saving does not send a test request. Memtro stores the key encrypted, never shows it again, and asks OpenAI which models it can reach, so a good key comes back as Added openai key (personal). with the number of models it found after it. If the listing fails, the key is still saved and the form says why, for a key OpenAI refuses: Saved. Listing models failed: openai 401. Check the key, then use Refresh models. Refresh models beside the key asks again.

The contents of a Codex auth.json are refused here with That looks like a Codex auth file, not an API key. A ChatGPT sign-in belongs under the Codex provider: see the Codex guide.

Leave the optional base URL empty. Hosted Memtro accepts an override only when it is https:// on one of the three hosts Memtro speaks to, api.anthropic.com, api.openai.com or generativelanguage.googleapis.com, so an override cannot point this key at a proxy or gateway; any other host is refused. A host your organisation approves under Custom model hosts does not count here: that list is for OpenAI-compatible endpoints only.

Model ids

Id	What it is
openai/<model>	One model through your key, e.g. openai/gpt-5.4 or openai/gpt-5.4-mini. Bare ids starting gpt-, chatgpt- or o and a digit are read the same way.
auto/openai	Routed by difficulty within OpenAI, on the default ladder gpt-5.4-nano → gpt-5.4-mini → gpt-5.4 → gpt-5.5. See model routing.

The model list beside the key shows the chat models the key can reach; embedding, audio, image and moderation models are left out of it. A named model is used only where an administrator has opened model choice on the Routing page; otherwise it is answered as memtro (see cost control).

What is different about OpenAI in Memtro

• Storage off. The Responses API keeps every prompt and answer for 30 days in your OpenAI account unless asked not to. Memtro sends each request with storage off, so OpenAI keeps no copy of that kind. It cannot switch off OpenAI's abuse-monitoring logs, and embedding requests have no such switch.
• Reasoning. Agent mode, and a reasoning_effort of high, xhigh or max (deep) or low or medium (standard), are passed on as OpenAI's reasoning effort, high or medium.
• Embeddings. An OpenAI key can embed your organisation's memories with text-embedding-3-small, and Memtro uses it for that in place of its own embedding model. It also serves POST /v1/embeddings, which needs a key of your own.
• More than one key. Your own key is tried first, then your team's, then the organisation's; within each, the order set on the Providers page, or Spread the load. A rate-limited or exhausted key hands over to the next before routing moves down the ladder. See several keys for one provider.
• Usage in tokens. Each request is recorded with the input and output tokens OpenAI reports. Allowances are set in tokens, and Memtro does not price them: OpenAI's own dashboard is where the bill is.

OpenAI's Decisions API, used as a decision provider, is added separately in the Decision providers section of the same page.

Your data at OpenAI

As of November 2026, OpenAI says that data sent to its API is not used to train or improve its models unless you explicitly opt in. Abuse-monitoring logs are kept for up to 30 days, separately from the application storage Memtro switches off; zero data retention, where OpenAI has agreed it with you, removes both. Bear in mind that a request through Memtro usually carries your organisation's memory with it, and that the text of each memory reaches your embedding provider when it is written. See privacy for what erasure in Memtro cannot reach.

What this is not

This is the API, billed by OpenAI per token on the account that owns the key. It is not a ChatGPT plan: to run Memtro on one, sign in under the Codex provider instead, as the Codex guide describes.