Models

Vaelor exposes models as three tiers rather than vendor model names. A tier names a speed and depth trade-off; we upgrade the model behind it as better ones ship, and announce changes in advance.

Tiers#

medium is the default wherever a tier is optional, such as OSINT runs. Chat requests always name a tier explicitly in config.model.

Current models#

Modes#

Chat requests can turn on up to two modes in config.modes.

• reasoning — the model thinks before answering, at high reasoning effort. Better on math, code and logic; slower and uses more output tokens.
• deep-research — the model can search the web and read pages before it answers, and cites what it used. Each search and page read is billed as a tool call.

{
  "config": {
    "model": "high",
    "modes": ["reasoning", "deep-research"],
    "uncensored": false
  }
}

Uncensored routing#

Setting config.uncensored to true routes the request to an uncensored model, whatever tier you pick. It's built for mature fiction and frank discussion of difficult subjects. It doesn't permit unlawful content. See Uncensored mode for details.

Limits#