Configure providers, hosts, and model assignments.
Cloud Providers
A
Anthropic
Claude models — two auth paths with different trade-offs
CLI (OAuth) — Runs requests through the
claude --print
subprocess using your Claude.ai subscription (Pro or Max plan). No API key needed; authenticate
with claude auth login on the Cortex host.
⚠ Anthropic now meters agentic tool use separately on paid plans. Each orchestrator pass
(tool loop + Claude response) counts against a separate monthly credit bucket, not just your
general subscription. CLI is fine for direct chat; for heavy automated workflows, an API key
avoids hitting the agentic cap.
Checking…
API Keys — Uses the Anthropic SDK directly, billed per token at
api.anthropic.com
rates. No agentic usage cap; the better choice for orchestrator-heavy or high-volume workloads.
{{ anthropic_key_rows }}
+ Add API key
G
Google
Gemini models via Gemini API — native SDK, not OpenAI-compatible
Gemini API keys serve two roles in Cortex: the orchestrator reasoning loop
(Gemini's tool-calling ReAct pass that runs before handing off to Claude) and any
Gemini chat models you assign in the Models tab. Multiple accounts let you
separate personal and work API usage or stay within free-tier rate limits across accounts.
Keys starting with AIza… from
Google AI Studio.
{{ google_account_rows }}
+ Add Google account
Cloud APIs
OpenAI-compatible cloud inference services — OpenRouter, OpenAI, Groq, X.ai, and more.
Add a service here, then go to Models to register individual models from it.
{{ cloud_host_rows }}
+ Add cloud API service
Local Hosts
Self-hosted OpenAI-compatible servers — Open WebUI, Ollama, LM Studio, etc.
{{ local_host_rows }}
+ Add local host
Models
{{ model_rows }}
+ Add model
Role Assignments
Map each task type to a model. Primary is tried first; Backup is the config-resolution fallback
if the primary model is missing from the registry. Backup does not fire on inference failures.