Theia IDE

One entry in ai-features.openAiCustom.customOpenAiModels — model, url and apiKey — puts Kunavo behind Theia Coder, the Architect and inline completion.

{
  "ai-features.openAiCustom.customOpenAiModels": [
    {
      "model": "claude-sonnet-5",
      "url": "https://api.kunavo.com/v1",
      "id": "kunavo-sonnet-5",
      "apiKey": "sk-kn-...",
      "developerMessageSettings": "system"
    },
    {
      "model": "claude-haiku-4-5",
      "url": "https://api.kunavo.com/v1",
      "id": "kunavo-haiku-4-5",
      "apiKey": "sk-kn-...",
      "developerMessageSettings": "system"
    }
  ]
}

Step by step

1. Create a key at /app/keys and copy it — it is shown once.
2. Turn the feature on: Theia's docs say to go to Preferences and enable the setting “AI-features => AI Enable”. Nothing below appears until that is done.
3. Open the AI Configuration view — Alt+A, or AI Configuration in the Manage (gear) menu at the bottom-left, right below Settings. Its categories are General, Providers & Models, Model Aliases, Agents, Prompts & Skills, Variables, Tools, Token Usage and MCP Servers.
4. Add the entries above. The docs describe this as clicking the link in the settings section for OpenAI Compatible Models — the preference is a structured list, and Theia notes that structured settings without a dedicated editor “defer to settings.json”, which is the file you land in. One object per model id; the url and apiKey repeat.
5. Point something at it. Under Agents, each agent has a Language Model selector; many agents instead resolve a model alias, so setting default/code, default/universal, default/code-completion, default/summarize and default/fast under Model Aliases moves several agents at once.
6. Send a chat message to Theia Coder, then ask for something that touches a file. The agents in this IDE lean on tool calls and workspace content, so a first run that reads or edits something tells you more than a greeting does — and Token Usage in the same view shows what the turn cost you in tokens.

Checked against the Theia IDE's AI features page, section OpenAI Compatible Models on September 21, 2026. Third-party settings move; if a field name here no longer matches what you see, that page is the authority, not this one.

Verify before you debug the client

One request settles whether a failure is the endpoint, the key, or the configuration file. If this returns JSON, the same base URL and key work in Theia IDE.

# Settles whether a failure is the endpoint, the key, or the client.
curl -sS https://api.kunavo.com/v1/models \
  -H "Authorization: Bearer sk-kn-..."

Which model id to put in the field

Every text model is reachable as a model id — the live list is GET /v1/models, and the catalog with prices is on the models page. Rates are USD per 1M tokens, input / output.

What Theia claims for this provider

The LLM Providers Overview table on Theia's own page rates each provider on three axes. This is Theia's claim about Theia, copied from that table on the date above — it is not a Kunavo test result, and the column headed “What that needs from the model id” is the only part of this table written here.

Theia adds its own caveat two paragraphs above that table, worth repeating because it is the honest frame for this whole page: not all models “may work out of the box, as they may require specific customizations or optimizations”.

What actually costs money

The Theia IDE is open source and free to download, and its AI features carry no charge of their own — what costs money is the model calls, billed by whoever holds the key in apiKey. Two settings move that bill more than the model choice does, and both are in the docs:

1. Automatic Code Completion is on by default, and the docs describe it as making “continuous requests to the underlying LLM while coding”. It is the agent that runs thousands of times a day. Pin default/code-completion to a cheap id, or switch the agent to manual mode in 'AIFeatures'=>'CodeCompletion' and trigger it with Ctrl+Alt+Space.
2. Max Context Lines, in the same settings group, caps how much surrounding file goes into each completion request. Every line of it is billed as input on every keystroke-driven call.

The chat agents are the opposite shape: fewer calls, much larger context, and the same workspace files resent turn after turn. That is the case prompt caching is for — see /docs/caching — and it is why the two halves of the model table above are split by how often the agent fires rather than by how clever it is.

When it does not connect

1. 404 — the url. Kunavo serves /v1/chat/completions, so the field wants the /v1 root; a bare origin or a full .../chat/completions both miss.
2. 401 — the key. Theia's Readme says the apiKey “will be send as a Bearer Token in the authorization request”, which is exactly what an sk-kn- key expects. Note the documented default: with no apiKey at all, Theia sends no-key, so a missing field looks like a rejected key rather than a missing one. (true means “use the global OpenAI API key” — not what you want here.)
3. The model id is not in the picker — that list comes from your own customOpenAiModels entries, not from the endpoint, so a missing id means a missing object. The id field is what the UI shows; leave it out and the model name is used instead.
4. The first system message is rejected or ignored — that is developerMessageSettings. It defaults to developer, which is an OpenAI-shaped role; Theia's own example for a non-OpenAI vendor sets system, which is why the block above does. user, mergeWithFollowingUserMessage and skip are the documented alternatives.
5. Nothing responds at all, anywhere — check Workspace Trust. Theia gates every AI feature behind it, and an untrusted workspace disables the chat input and inline completion and shows an AI Features are Restricted message.