Skip to content
Olympusx AIOlympusx AIOlympusx AI documentation

Code Assistant

Use cases › Code Assistant connects the editor tools your developers already use to models served in your tenant. Chat, edits, coding agents and autocomplete call the tenant’s gateway with a key that belongs to one developer, so every request is charged to a person and to a team. A tenant admin turns it on and chooses the models and the teams; developers create their own keys and paste the configuration Olympusx AI writes for their tool.

It does four jobs:

  • Editor access. Continue, Cline, Aider, Zed and any client that speaks the OpenAI API call the gateway with a per-developer key. See Connect your editor and Developer keys.
  • Codebase knowledge. Repositories, or whole groups, from GitLab, GitHub or Bitbucket Cloud are indexed after a secrets scan, so a developer’s editor can search the team’s code and ask what uses a function. See Code sources and The repository view.
  • Merge request review. A webhook from your code host starts a review of each merge request, and with a review token the comments are posted on their lines. See Merge request review.
  • Spend by developer and team. See Usage and spend.

Who can open it

Section titled “Who can open it”

The page has five tabs. Each tab is shown only to the roles that can use it.

Role Connect an IDE Usage by developer Codebase knowledge Reviews Settings
Admin Own keys and usage Read Read, add, edit, sync and delete sources; search Read; review a merge request Read and change
Builder Own keys and usage Not shown Read; search Read; review a merge request Read
Auditor Not shown Read Read Read Read
End user, Reviewer Cannot open the page
Platform admin As an admin, in any tenant they pick with the tenant picker

A platform admin’s own keys belong to the tenant their account names; a platform admin with no tenant holds no keys.

What you see

Section titled “What you see”

The page header names the Code Assistant and carries a tenant picker. Below it, the tabs in this order:

  1. Connect an IDE — create your key, copy the configuration for your tool, list your devices, and read your own usage this month.
  2. Usage by developer — the month’s cost and tokens, by team and by developer.
  3. Codebase knowledge — the code sources this tenant indexes, their repositories, and how each last synced.
  4. Reviews — every merge request a webhook delivered or someone asked to review.
  5. Settings — the models, the teams, the key rules and the gateway address.

The page opens on the first tab your role can see.

Set it up in five minutes

Section titled “Set it up in five minutes”

You need a chat model the tenant can call (a deployed model, or one from a connected provider; see Deploy a model and External models) and at least one project to serve as a team (see Projects).

As a tenant admin:

  1. Open Use cases › Code Assistant and select the Settings tab.
  2. Under Models, choose the model for Chat, edit and agents. Choose an Autocomplete model if you want completions as developers type; leave it at None otherwise.
  3. Under Access, choose one or more Teams. Each team is a project: a developer picks one when creating a key, and its requests are charged there.
  4. Turn on Developers can connect their IDE and select Save settings.

As a developer (an admin or a builder):

  1. Select the Connect an IDE tab.
  2. Under 1. Create your key, choose your Team, enter a Device name such as Work laptop, and select Create key. Copy the key now: it is shown only once.
  3. Under 2. Configure your tool, select your tool and paste the configuration into the file it names.
  4. Ask your editor a question. The request appears under Your usage this month on the same tab.

To check the key from a terminal first, select Any OpenAI client and run the curl command it shows. A JSON answer with a choices array means the address, the key and the model all work.

Settings

Section titled “Settings”

The Settings tab is read by everyone who reaches the page and changed by tenant admins. Other roles see a Read only notice.

Section Field What it does
Access Developers can connect their IDE Turns the Code Assistant on. Turning it on needs a chat model and at least one team.
Access Teams The projects developers charge their requests to. Each key belongs to one team.
Access Key lifetime, in days A developer’s key stops working this many days after it is created. 1 to 365; 90 by default.
Access Keys per developer The most active keys one developer may hold, one per device. 1 to 50; 5 by default.
Models Chat, edit and agents Answers questions, edits code and runs coding agents. It also writes merge request reviews and answers questions on a repository’s Ask tab.
Models Autocomplete Completes code as it is typed. A small model is best: it is called on every pause. None offers no autocomplete.
Models Codebase search The embedding model that indexes code sources and the questions asked of them. It cannot change while a source exists.
Gateway Address editors use The gateway’s public OpenAI base URL, ending in /v1. Read only; a platform admin sets it at install time with the gateway.publicUrl chart value, or it is derived from the gateway’s host. See Configuration.

When the Code Assistant is on, every model you choose must be one the tenant can call: granted, and deployed and ready or served by a provider. Olympusx AI refuses to save a model every request to which would fail.

Turning the Code Assistant off stops developers from creating keys and stops the codebase tools in their editors. Keys already created keep calling their models until they expire or are revoked; revoke them on Keys and budgets to stop them at once.

What protects the code

Section titled “What protects the code”
  • A developer’s prompt is checked for secrets and for prompt injection before it reaches the model, whatever the tenant’s guardrail policy says about those two checks. Personal-data detection, toxicity and banned topics are off for editor traffic, so ordinary code reaches the model unchanged. The answer streams back as it is written. See Developer keys.
  • Prompts and code sent from editors are not stored. The spend figures come from the usage ledger, which records tokens and cost, not content.
  • Code is indexed only after a secrets scan. A file the scan could not check is left out of the index. See Code sources.
Section titled “Related”