Open Plugins
In ChatGPT Plugins, choose O++ Codex Agent or add the hosted connection in Developer Mode.
Hosted Codex agent + responsive knowledge
Your Codex agent, inside ChatGPT.
O++ turns a request in ChatGPT into a real, durable Codex CLI job—and gives every customer a private knowledge base that saves in seconds, retrieves immediately, and curates reviewed Git knowledge in the background.
Native onboarding
Connect once. Work from ChatGPT.
The customer path uses ChatGPT’s native Plugins interface and hosted OAuth, so it works without a developer terminal. During early access, Developer Mode can add the connection URL above directly.
Click Connect
ChatGPT opens O++ sign-in. Authenticate and approve the requested job, credential, and knowledge permissions.
Add your API key
Ask O++ to configure OpenAI, click Open secure setup, and enter a key from the OpenAI dashboard only on that O++ page.
Connect sources and add knowledge
Authorize O++ read-only Gmail and Drive access from its secure Google consent flow, or hand off a selected excerpt from another app such as Slack. The authenticated capture is saved and available immediately; paid Codex review runs later.
Retrieve it
Ask a later conversation what O++ remembers. It returns grounded matches with sources, trust, curation state, and freshness in seconds.
Keep credentials out of chat. O++ will never ask you to paste an OpenAI API key into a conversation, tool argument, support message, or terminal. ChatGPT currently exposes plugins on supported web and desktop surfaces; mobile plugin availability is controlled by OpenAI.
Designed for real work
Agentic execution with durable state.
Agent-task calls start bounded jobs and return durable handles. Knowledge calls use a separate synchronous path. O++ preserves job progress across disconnects, runs configured Codex work in isolated workers, and keeps permissions under service policy.
Run and inspect agent jobs
Start read-only inspections or controlled edits, follow events, retrieve results and patches, continue an approved session, or cancel explicitly.
Remember now, curate later
Each authenticated write commits private working memory immediately. Bounded rolling batches later use a paid Codex editor and independent reviewer before an exact approved revision can merge.
Retrieve with provenance
A fast private lookup returns grounded captures with source identifiers, trust, lifecycle, curation, and freshness metadata. It does not wait for a new model run.
“Use O++ to securely set up my OpenAI API key.”
“Connect my Google Workspace, find the launch decision in Drive, and remember it with provenance.”
“What do you remember about the launch criteria? Cite sources and distinguish pending from reviewed knowledge.”
Privacy boundaries
Your identity, key, and knowledge stay separated.
O++ is built around tenant isolation, explicit user consent, and least-privilege connections. It does not use a shared platform OpenAI key.
Your API key goes directly from the one-time, same-origin setup page to O++. It is stored in a Google Cloud KMS-encrypted envelope, never displayed again, and released only to the short-lived worker making your model call.
Working memory is scoped to your signed-in identity. Its reviewed Git projection is isolated in a private tenant repository and cannot bypass the exact-head review gate.
You can authorize O++ to search and read Gmail and Drive through Google’s consent screen. OAuth credentials stay encrypted and outside model context; other apps such as Slack still use an explicit bounded-excerpt handoff.
OAuth scopes separate job inspection, editing, cancellation, credential management, and knowledge read/write. Every operation is checked against the signed-in tenant and user.
Ready to connect O++?
Open ChatGPT Plugins, choose O++ Codex Agent, and click Connect.