nodecode-obscurify

Obscurify.ai: a privacy-focused, OpenAI-compatible API to 200+ models, listed by the models a turn can run. A Nodecode rewrite of Obscurify-ai/hermes-obscurify-provider by Obscurify-ai.

Obscurify.ai (https://obscurify.ai) serves the OpenAI chat wire at https://obscurify.ai/v1 with a bearer key. Accounts are a random account number -- no email, no password -- and text requests use zero-data-retention routes. With this cell, /models offers Obscurify.ai, /models lists its models, and a turn on one of them goes out the way the API wants it:

the model list is the API's own public /v1/models catalog, minus the decision models (they answer structured questions only, and refuse a chat request without their response_format) and minus the models whose context window is under the floor Hermes required, 64,000 tokens; the floor is the section's min_context_tokens here, so it can be lowered
context windows come from that same catalog, so a turn's budget is the API's own number rather than a guess
reasoning effort goes as the top-level reasoning_effort, the API's own spelling of the rungs off through max; a turn that turned reasoning off sends none
before the catalog has answered, the picker lists the five featured models the plugin named, with the windows it checked for them

The catalog is public, so it is read without the key, once at start and hourly after that, and kept in memory: a box with no network keeps the featured list, and a failed refresh keeps the list already read. The key rides only on chat turns.

Get it

Tell the agent "add the obscurify cell", or run nodecode add nc://obscurify.

Set it up

Make an account at obscurify.ai and add a balance: the API works on every plan with a positive balance, frontier models want the Casual plan or higher. On obscurify.ai/account, click Show API Key, then pick Obscurify.ai in /models and paste the key when the picker asks, or set OBSCURIFY_API_KEY in the environment. A key saved with /connect answers first, and no other provider's key is ever sent to Obscurify.ai.

The address and the floor are optional, for a relay or a smaller context budget:

{
  "obscurify": {
    "base_url": "https://obscurify.ai/v1",
    "min_context_tokens": 64000,
    "timeout_seconds": 8
  }
}

Use it

Ask for a model -- /models lists what the catalog currently serves -- then talk as usual. /think offers the rungs; a model the catalog says does not reason is listed without them. Good first picks are deepseek/deepseek-v4-flash, z-ai/glm-5.3, moonshotai/kimi-k2-0905 and qwen/qwen3-coder. A turn costs money from the Obscurify.ai balance at that model's price.

Test

From this repository's root:

.github/test-cell.sh providers/nodecode-obscurify

MIT licensed. See NOTICE.