Roo Code with uttapen (discontinued extension)

Setup in 3.54

1. Roo Code panel → gear (Settings).
2. Set API Provider to OpenAI Compatible — not OpenAI, which uses the Responses API that uttapen doesn't serve.
3. Fill in:

Field	Value
Base URL	https://api.uttapen.ir/v1
API Key	your sk-up-… key from Dashboard → API keys
Model	pick from the list it fetches from GET /v1/models, or choose "Use custom" and type the exact id

The output-token setting matters most

Roo Code sends the output ceiling as max_completion_tokens, and there are three cases:

• Include max output tokens unchecked: no ceiling is sent. uttapen holds a default estimate, and if the model writes more, the real charge is still deducted in full and the balance can dip negative (pricing).
• Checked, but Max Output Tokens left at its -1 default: uttapen no longer rejects this. It treats a zero or negative value as "no ceiling given" and holds on its own default estimate, same as the case above — so you get the same negative-balance risk, not a 400.
• Checked with a positive number (we suggest 8192): that value is sent, the hold stays predictable, and you avoid both risks above. Use this one.

Set Context Window Size to the model's real context as well; it defaults to 128,000 and every model's number is on the model list.

Choosing a model

Roo Code only works with models that support tool calling — it has no XML fallback. Good coding choices: anthropic/claude-sonnet-5, openai/gpt-5, qwen/qwen3-coder, deepseek/deepseek-v3.2.

There is no separate model for titles or summaries; the active profile does everything. If you keep several profiles, give each its own key with a monthly cap so their spend is separated in Dashboard → Usage.

Troubleshooting

Error	Cause in Roo Code	Fix
401 invalid_api_key	Wrong or revoked key	Create a new key
404 model_not_found	A wrong hand-typed id	Pick from the list, or copy from /en/models (the 404 also lists the closest ids it guessed, in uttapen.suggestions)
404 not_found	Provider still on OpenAI, hitting /v1/responses	Choose OpenAI Compatible
402 insufficient_balance	Balance below the required hold	Top up, or lower the output ceiling
429	The key's rate or monthly limit (rate limits)	Wait, or raise the limit

When the balance can't cover the whole output ceiling, uttapen lowers it to what the balance allows rather than refusing (floor: 1,024 tokens) and returns the granted value in the X-Uttapen-Max-Tokens header; the reply may then stop with finish_reason: "length". 402 is reserved for the case where even that floor doesn't fit.

The full tool list is in integrations, and error codes are in errors.