Skip to main content

Connecting your AI

OtoDock doesn't bundle an AI model — you bring your own. You connect the subscription you already pay for (admins can also add API keys and local models), and OtoDock runs your agents on them. This page covers how to connect each; for which engine to choose, see AI Engines.

Where to connect​

Engines can be connected in two places, and both feed the same platform:

  • Your own engines — any user connects their own Claude or ChatGPT subscription under User Settings → AI Engines. That's the one kind of connection a regular user adds.
  • The platform's engines — an admin sets these up under Setup → AI Engines (the admin page): subscriptions, API keys, and local models.

An admin can connect from either screen. On each of their own connections they also get a checkbox to let it power the platform's shared work — so an admin who connects their Claude or ChatGPT subscription in their personal settings can, with one box, have it run the platform's agents too. More on that in Your own vs. the shared pool below.

OtoDock does its real work with Claude Code and Codex, so the best starting point is the subscription you may already pay for:

  • Claude (Claude Pro / Max). Click connect, sign in to Anthropic, and paste the code back — OtoDock stores the connection securely and keeps it refreshed.
  • ChatGPT (Codex). Click connect and approve the one-time device-code sign-in with OpenAI. (If the sign-in is refused, enable device code login in your ChatGPT security settings first — some accounts have it off by default.)

You can connect more than one account per provider — each shows up as its own entry with the account's email, so two Claude plans (say, a personal Pro and a work Max) live side by side. Reconnecting an account you've already added just refreshes it, never duplicates it.

OtoDock keeps idle connections fresh in the background and warns you before a login expires (72 hours and 24 hours ahead), then tells you the moment one dies — with a per-account Reconnect button in User Settings → AI Engines that revives the same account in place. Starting a chat on an expired account says exactly that, instead of a generic error.

Use an API key​

Prefer pay-as-you-go API billing? Add your own Anthropic or OpenAI key under User Settings → AI Engines — one key per engine, with a weekly or monthly spending cap — and your own chats and tasks run on it. An admin adds Anthropic, OpenAI, or Groq keys to the platform under Setup → AI Engines: these run agent work, and an admin can let an individual user draw on them for their own chats too (see Your own vs. the shared pool). Keys are stored encrypted and kept out of agents' reach — a session only ever carries the key its own work runs on.

Use local models (admins)​

Running your own models? An admin adds a local endpoint under Setup → AI Engines → Local models: pick Ollama or OpenAI-compatible endpoint (llama.cpp, LM Studio, vLLM, and similar), paste its address, add an API key only if the server wants one, and tick which engines may use it. One endpoint serves both Codex and the direct engine, and Discover lists the server's models to add. On a Docker install, use the host's LAN IP or host.docker.internal rather than localhost, and give Ollama its OpenAI-compatible address (http://<host>:11434/v1).

Your own vs. the shared pool​

A subscription you connect powers your own chats and tasks. Every connection shows a Use for my own work switch (on by default) — handy when you've connected more than one subscription and want to rest one for a while without disconnecting it.

Admins get one more switch on each of their own connections, because an admin's subscription can also serve the whole platform:

  • Contribute to the shared pool — let work that belongs to an agent rather than a person draw on it. When an admin connects their own Claude or ChatGPT subscription, this is on by default (they can untick it).

This mirrors how OtoDock decides which connection a session uses:

  • Your own chats and tasks run on your own connections first — you bring your own subscription for your personal work.
  • An agent's own work — a scheduled task, a shared agent, a meeting, a trigger: anything tied to an agent rather than to you personally — runs on the shared platform pool that admins contribute to.

If you haven't connected anything of your own, an admin can switch on platform access for your account — a per-user toggle that's off by default — to let you borrow the platform's API-key or local connections for your own chats. You can never borrow someone else's personal Claude or ChatGPT subscription, though; those always stay with their owner and are reserved for agent work.

Switching engine or model​

Set an agent's default engine and model in its settings, and switch per chat from the model dropdown. While a conversation is live the dropdown offers the active engine's models. Once its session has ended, the agent's other engines appear too: pick one and the chat continues on the new engine with its history carried over. The picker only offers engines you can actually use, and enabling an engine on an agent requires a platform subscription for it.

Next steps​