Featherless, Venice AI and Perplexity Agent in kRouter

kRouter has supported Featherless, Venice AI and Perplexity Agent since v0.5.98, in July 2026. They are three of its 95+ providers. All three take an API key and speak OpenAI's API, so once you paste a key, kRouter asks the provider for its current model list. When they add a model you do not wait for a kRouter release; the live model catalog explains how that works.

If you set one of them up on an early build: chat requests to all three failed until v0.5.103, and Featherless's model list did not load until v0.5.108. Anything newer works.

Here is what each one is good for.

Featherless: open models, one endpoint

Featherless serves open-weight models behind one OpenAI-compatible endpoint, so you do not rent GPUs or manage deployments. Its API listed more than 22,000 models in October 2026, from the big open releases down to community fine-tunes.

• Alias: fl
• Auth: API key
• Good for: one key for many open models, a backup tier in a combo, or a specific open-weight model that the big labs don't host.

A model name is kRouter's alias followed by Featherless's own ID, which is usually the Hugging Face repository name:

fl/deepseek-ai/DeepSeek-V4-Pro
fl/zai-org/GLM-5.3
fl/moonshotai/Kimi-K3

For a model that also has its own API, compare before you commit. Featherless's catalog lists all three of these with a 262,144-token context window, while DeepSeek's own API gives V4 Pro 1M. Prices differ too: Featherless lists DeepSeek V4 Pro at $1.60 in and $3.20 out per million tokens, and DeepSeek charges $0.66 to $1.32 in and $1.98 to $3.96 out, depending on the hour. DeepSeek V4 vs GPT has the full price table.

Venice AI: private inference

Venice AI is built around privacy, and its API is a drop-in replacement for OpenAI's chat endpoint.

• Alias: venice
• Auth: API key
• Good for: work where prompt privacy matters, as long as you pick a Private model, or work that a mainstream provider's filters block for no good reason.

Venice says it does not store or log prompt and response content for normal inference. It also labels every model with one of two privacy modes, and the difference matters:

• Private: the prompt and reply are processed for inference only and are not kept after the request. Most open-weight models run this way, including Venice's own uncensored models, GLM, DeepSeek and Llama.
• Anonymized: Venice forwards the request to the model's maker without your identity, but that provider still sees the prompt. Claude, Gemini and OpenAI's hosted GPT models are in this group.

Perplexity Agent: many labs through one Responses API

This is separate from kRouter's search-focused Perplexity provider (pplx). The Perplexity Agent API serves models from OpenAI, Anthropic, Google and xAI, plus Perplexity's Sonar and hosted GLM, Kimi and Nemotron models, through one endpoint compatible with OpenAI's Responses API.

• Alias: perplexity-agent
• Auth: API key
• Good for: several labs' models on one bill, with Sonar in the same list.

Perplexity's model IDs already contain a slash, so the kRouter name has two:

perplexity-agent/openai/gpt-6-sol
perplexity-agent/anthropic/claude-sonnet-4-6
perplexity-agent/perplexity/sonar

kRouter sends these requests to Perplexity's /v1/responses endpoint and converts the format both ways, so Chat Completions and Claude clients can use them too. Responses API vs Chat Completions covers what that conversion involves.

Connecting any of them

The steps are the same for all three:

1. Open the dashboard at http://localhost:20128/dashboard, go to Providers and pick the provider.
2. Click Add API Key, name the connection and paste the key. As you type, kRouter fetches the provider's model list and shows how many models it found, or the provider's error if the key is wrong.
3. Click Save.

Once saved, they work like any other key provider. Add several keys to one provider and Zenith scoring picks between them. A key that fails is cooled down, and in a combo they take over when an earlier entry fails. The Quota Tracker does not cover these three, so check your remaining credit on each provider's own site.

Try it

npm install -g @sifxprime/krouter
krouter

Browse the full provider list, or read the Providers guide for connection details.