Table of Contents

AI Providers

JD.AI detects and connects to AI providers automatically on startup. No manual configuration is required — install a provider, authenticate if needed, and JD.AI handles the rest.

Provider detection

On startup, JD.AI checks for available providers in this order:

  1. Claude Code — checks for a local CLI session
  2. GitHub Copilot — checks for GitHub authentication
  3. OpenAI Codex — checks for Codex CLI authentication or OPENAI_API_KEY
  4. Ollama — checks for a local server at localhost:11434
  5. Local (LLamaSharp) — scans ~/.jdai/models/ for GGUF model files

The first provider with a valid connection becomes the default. Results are shown on startup:

Provider detection on startup

Claude Code

Setup:

  1. Install Claude Code: npm install -g @anthropic-ai/claude-code
  2. Authenticate: claude auth login
  3. JD.AI detects the session automatically

How it works:

  • Uses the Claude Code CLI's OpenAI-compatible endpoint.
  • Supports Claude Sonnet, Opus, and Haiku models.
  • Auto-refresh: JD.AI attempts to renew expired sessions.
Tip

Best for: Complex reasoning, long-form code generation, and nuanced analysis.

GitHub Copilot

Setup:

  1. Authenticate via GitHub CLI: gh auth login --scopes copilot
  2. Or via VS Code: sign in to the GitHub Copilot extension.
  3. JD.AI detects available Copilot models.

How it works:

  • Uses Copilot token exchange for OpenAI-compatible API access.
  • Available models include GPT-4o, Claude Sonnet, Gemini, and more.
  • Auto-refresh: JD.AI attempts to renew expired tokens.
Tip

Best for: Teams already using GitHub and access to multiple model families.

Ollama

Setup:

  1. Install Ollama from https://ollama.com.
  2. Start the server: ollama serve
  3. Pull models: ollama pull llama3.2 (chat) and ollama pull all-minilm (embeddings).
  4. JD.AI auto-detects all available models.

How it works:

  • Connects to the local Ollama API at http://localhost:11434.
  • Lists all locally available models.
  • No authentication required.
Tip

Best for: Offline usage, privacy, fast iteration, and experimentation.

OpenAI Codex

Setup:

  1. Install the Codex CLI: npm install -g @openai/codex
  2. Authenticate: codex auth login (or set OPENAI_API_KEY environment variable)
  3. JD.AI detects the session automatically.

How it works:

  • Uses the Codex CLI's OAuth token exchange for API access.
  • Credential resolution: API key → access token → OPENAI_API_KEY env → CODEX_TOKEN env → ~/.codex/auth.json → device code login.
  • Supports all OpenAI chat models (GPT-4o, GPT-4.1, o3, etc.).
Tip

Best for: Teams using OpenAI models who want the Codex CLI's authentication flow.

Local models (LLamaSharp)

Setup:

  1. Place .gguf model files in ~/.jdai/models/ (or any directory).
  2. JD.AI detects them automatically on startup.
  3. Or download directly: /local download TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF

How it works:

  • Loads GGUF models in-process via LLamaSharp (C# bindings for llama.cpp).
  • Auto-detects GPU hardware (CUDA, Metal) and falls back to CPU.
  • No external service or internet connection required.
  • Manage models with /local commands: list, add, scan, search, download, remove.
Tip

Best for: Air-gapped environments, privacy-sensitive workloads, and fully standalone operation.

See Local Models for the full guide.

Switching providers and models

Use slash commands to manage providers and models at any time during a session:

/providers          # List all detected providers with status
/provider           # Show current provider and model
/models             # List all available models across providers
/model qwen3:30b    # Switch to a specific model

Provider comparison

Feature Claude Code GitHub Copilot OpenAI Codex Ollama Local (LLamaSharp)
Setup CLI auth GitHub auth CLI auth / API key Local install Drop in .gguf files
Internet required Yes Yes Yes No No
Cost Claude subscription Copilot subscription OpenAI subscription Free (local) Free (local)
Model variety Claude family Multi-family OpenAI family Open source Any GGUF model
Speed Fast Fast Fast Depends on hardware Depends on hardware
Privacy Cloud Cloud Cloud Fully local Fully local
Embedding support Yes Limited Yes Yes No

Environment variables

Variable Description Default
OLLAMA_ENDPOINT Ollama API URL http://localhost:11434
OPENAI_API_KEY OpenAI / Codex API key (if not using CLI auth) —
CODEX_TOKEN Codex CLI access token override —
JDAI_MODELS_DIR Local model storage directory ~/.jdai/models/
HF_HOME HuggingFace cache directory ~/.cache/huggingface/
HF_TOKEN HuggingFace API token —

See also

Edit this page