Qorven is model-agnostic. Bring OpenAI, Anthropic, Bedrock, DeepSeek, Gemini, Groq, Mistral, Together, Fireworks, Ollama, LM Studio, Perplexity, DashScope, or an OpenRouter aggregator. Each Qor picks its own model.

​
Three layers of configuration

1

Provider keys (install-wide)

You paste an API key per provider in Settings → Provider Keys. Keys are encrypted at rest with your install’s encryption key.
2

Default model (install-wide)

In Models Hub → Models you mark one model as the install-wide default. New Qors start with this model.
3

Per-Qor override

In the Qor’s config page you can override the model — e.g. Prime uses Claude Opus for reasoning, the Coder uses DeepSeek for cheap iteration.

​
Picking a model for your use case

Chief-of-staff / Prime

Needs reasoning + tool use. Claude Opus 4.x, GPT-5, Gemini 2.5 Pro, DeepSeek V3.

Specialists (coding, research)

Specialised mid-size models. DeepSeek Coder, Qwen Coder, GPT-5-mini, Claude Haiku.

Background (dreaming, compaction)

Cheap + good enough. GPT-5-nano, Gemini Flash, Haiku, DeepSeek Chat.

Local / privacy-first

Ollama (Llama 3.3 70B, Qwen 2.5), LM Studio. Zero outbound calls.

Lowest cost

DeepSeek, Gemini Flash, Groq Llama 3.3, OpenRouter’s free tier.

Fastest

Groq (tokens/sec king), Together AI, Fireworks. For real-time chat.

​
The model catalog

Qorven ships with a catalog of 1,700+ models + pricing + context windows + capability flags. It’s refreshed weekly from provider APIs (via an llmstats integration). 
qorven models search --capability vision --max-price 1.00
# Gemini 2.5 Flash  — vision, tools, 1M ctx, $0.075/1M in, $0.3/1M out
# DeepSeek VL       — vision, tools, 128k ctx, $0.20/1M in, $0.40/1M out
# …
Catalog CLI →

​
Failover

When your Prime’s preferred model hits a 429 or 402, Qorven cycles:
1

Next key for the same provider

If Prime is on OpenAI and key sk-abc is rate-limited → try sk-def.
2

Next provider in the priority list

All OpenAI keys exhausted → fall through to Anthropic.
3

Smart scorer tries to match capabilities

Prime needs 200k context → Anthropic Claude Sonnet picked over Gemini Flash.
4

All providers exhausted

The Qor writes an error message to the user: “All LLM providers failed. Check your provider keys.”
Failover deep-dive →

​
Providers

OpenAI

Anthropic

Bedrock

DeepSeek

Gemini

Groq

Mistral

Together

Fireworks

Ollama

LM Studio

Perplexity

DashScope (Qwen)

OpenRouter

​
Where next

Provider keys

Key management UI + encryption.

Default model

Setting the install-wide default.

Per-Qor model

Overriding per agent.

Failover

Automatic rotation on rate-limits.