AI Providers

This page covers how to configure AI providers (Claude, OpenAI, Ollama, etc.), create reusable profiles with fallback chains, and assign providers to individual projects.

For authentication between Condrix components (clients, Cores, Maestro), see Authentication.

Overview

Condrix supports multiple AI providers through a Connections + Profiles architecture:

Connection — A single authenticated link to a provider (e.g., “My Claude Plan”, “Work OpenAI Key”)
Profile — A named preset with a primary connection and an ordered fallback chain
Per-project assignment — Each project can use a different profile (or a direct connection)
Workspace-level model selection — Within a workspace, you can pick a specific model from the active connection’s model list

This lets you mix providers freely — for example, use Claude for production code, Ollama for local experiments, and OpenAI as a fallback when Claude hits a rate limit.

Supported Providers

Provider	Auth Method	Use Case
Claude (OAuth)	Browser sign-in via Claude account	Recommended for Claude plan subscribers
Claude (API Key)	Anthropic API key (`sk-ant-api03-...`)	Development, pay-per-use billing
OpenAI	OpenAI API key (`sk-...`)	GPT-4o, GPT-4o-mini, o1, o3, and other OpenAI models
Local	None (no auth required)	Ollama, LM Studio, or any local OpenAI-compatible server
Custom	API key (optional)	Azure OpenAI, Together AI, Groq, Fireworks, or any OpenAI-compatible endpoint

Connections

A Connection represents a single authenticated link to an AI provider. Each connection stores:

Name — A human-readable label (e.g., “My Claude Account”, “Office OpenAI Key”)
Provider — Which provider type (Claude, OpenAI, Local, Custom)
Credentials — API key, OAuth tokens, or nothing for local providers
Base URL — For Local and Custom providers, the endpoint URL
Fallback triggers — Conditions that cause Condrix to skip this connection and try the next one

Creating a Connection

Open Settings (gear icon) in the web client
Navigate to the AI tab
Under Connections, click Add Connection
Select a provider type
Fill in the provider-specific fields
Click Save

Provider-Specific Setup

Claude (OAuth)

The recommended method for Claude. Uses your existing Claude plan allocation.

Select Claude (OAuth) as the provider
Click Sign In with Claude
Approve the request in your browser
Paste the authorization code back into the dialog

The Core handles token refresh automatically. You should rarely need to re-authenticate.

Claude (API Key)

For development or API-billed usage.

Select Claude (API Key) as the provider
Paste your Anthropic API key (sk-ant-api03-...)

You can also set the key via environment variable:

export CONDRIX_CORE_CLAUDE_API_KEY=sk-ant-api03-...

OpenAI

Select OpenAI as the provider
Paste your OpenAI API key (sk-...)
Available models are auto-discovered from the OpenAI API

Local (Ollama / LM Studio)

For free, private, local inference. No API key needed.

Start your local model server:

# Ollama
ollama serve

# Or launch LM Studio and enable the local server

Select Local as the provider
The base URL defaults to http://localhost:11434/v1 — change it if your server uses a different port
Available models are auto-discovered from the server

Compatible local servers:

Ollama — Lightweight, CLI-based, supports thousands of models
LM Studio — GUI-based with a built-in model browser
Any server exposing an OpenAI-compatible /v1/chat/completions endpoint

Custom (OpenAI-Compatible)

For any other provider with an OpenAI-compatible API.

Select Custom as the provider
Enter the Base URL (e.g., https://api.together.xyz/v1, https://api.groq.com/openai/v1)
Enter an API Key if the endpoint requires authentication
Models are auto-discovered from the endpoint’s /models route

Profiles

An AI Profile is a named preset that defines a primary connection and an ordered fallback chain. Profiles decouple routing decisions from individual projects and workspaces.

Creating a Profile

Open Settings → AI tab
Under Profiles, click Add Profile
Enter a name for the profile (e.g., “Production”, “Local Dev”)
Select a primary connection — the default provider
Optionally add fallback connections in priority order
Click Save

Fallback Chains

When the primary connection encounters an error that matches one of its configured fallback triggers, Condrix automatically retries the request with the next connection in the chain.

Available fallback triggers:

Trigger	When It Fires
`rate_limit`	Provider returned HTTP 429
`quota_exceeded`	Account usage quota exhausted
`auth_error`	Credentials invalid or expired
`timeout`	Request timed out
`server_error`	Provider returned HTTP 5xx

Example: resilient production profile

Primary:    Claude (OAuth)         → triggers: rate_limit, quota_exceeded
Fallback 1: OpenAI (API Key)      → triggers: rate_limit, server_error
Fallback 2: Ollama (local)        → no triggers (last resort)

If Claude hits a rate limit, the request automatically retries on OpenAI. If OpenAI also fails, it falls back to local Ollama. The user sees no interruption.

Example Profiles

Profile	Primary	Fallback	Use Case
Cloud First	Claude (OAuth)	OpenAI	Best quality, cloud-billed
Local Dev	Ollama	None	Free, private, offline-capable
Cost-Optimized	Ollama	Claude (API Key)	Local when possible, cloud when needed
High Availability	Claude	OpenAI → Ollama	Maximum uptime with three-tier fallback

Per-Project Assignment

Each project can have its own AI Profile:

Open Settings → Projects tab
Select a project
Choose an AI Profile from the dropdown
All workspaces in that project use the assigned profile by default

This lets you use different providers for different projects — for example, Claude for production work and Ollama for experiments.

Workspace-Level Model Selection

Within a workspace or chat session, you can select a specific model from the active connection’s available models. This is useful when a provider offers multiple tiers:

Claude: claude-sonnet-4-20250514, claude-haiku-4-20250414
OpenAI: gpt-4o, gpt-4o-mini, o1, o3-mini
Ollama: llama3, codellama, mistral, deepseek-coder

Model selection is per-workspace and does not change the profile or connection configuration.

What’s Next?

Authentication — Connect clients, Cores, and Maestro
Monorepo Structure — How the multi-provider system is organized in the codebase