Providers Reference

JD.AI supports 15 AI providers. This reference covers capabilities, authentication, environment variables, and comparison tables. For setup guides, see AI Providers.

Provider summary

#	Provider	Package / Connector	Auth Type	Requires Internet
1	Claude Code	Claude Code CLI (OAuth)	OAuth	Yes
2	GitHub Copilot	Copilot token exchange	OAuth	Yes
3	OpenAI Codex	Codex CLI (OAuth)	OAuth	Yes
4	Ollama	Local REST API	None	No
5	Local (LLamaSharp)	LLamaSharp (in-process)	File	No
6	OpenAI	`Microsoft.SemanticKernel.Connectors.OpenAI`	API Key	Yes
7	Azure OpenAI	`Microsoft.SemanticKernel.Connectors.AzureOpenAI`	API Key	Yes
8	Anthropic	`Anthropic.SDK` (native Messages API)	API Key	Yes
9	Google Gemini	`Microsoft.SemanticKernel.Connectors.Google`	API Key	Yes
10	Mistral	`Microsoft.SemanticKernel.Connectors.MistralAI`	API Key	Yes
11	AWS Bedrock	`Microsoft.SemanticKernel.Connectors.Amazon`	AWS IAM	Yes
12	HuggingFace	`Microsoft.SemanticKernel.Connectors.HuggingFace`	API Key	Yes
13	OpenRouter	OpenAI connector + custom base URL	API Key	Yes
14	OpenAI-Compatible	OpenAI connector + custom base URL	API Key	Yes

Provider capabilities matrix

Provider	Streaming	Tool Calling	Embeddings	Model Discovery	Cost
Claude Code	✅	✅	❌	✅	Subscription
GitHub Copilot	✅	✅	❌	✅	Subscription
OpenAI Codex	✅	✅	❌	✅	Subscription
Ollama	✅	✅	✅	✅	Free
Local (LLamaSharp)	✅	❌	✅	✅	Free
OpenAI	✅	✅	✅	✅	Pay-per-use
Azure OpenAI	✅	✅	✅	❌	Pay-per-use
Anthropic	✅	✅	❌	❌	Pay-per-use
Google Gemini	✅	✅	❌	❌	Pay-per-use
Mistral	✅	✅	❌	❌	Pay-per-use
AWS Bedrock	✅	✅	✅	❌	Pay-per-use
HuggingFace	✅	❌	❌	❌	Free/Pay
OpenRouter	✅	✅	❌	✅	Pay-per-use
OpenAI-Compatible	✅	Varies	Varies	✅	Varies

Claude prompt caching (API key + OAuth)

Anthropic API-key sessions and Claude Code OAuth sessions both use Anthropic's native Messages API path and support automatic prompt caching controls.

Default behavior:

Prompt caching is enabled by default (prompt_cache=on).
Cache TTL defaults to 5m (prompt_cache_ttl=5m).
Optional extended TTL: 1h.

Policy details:

Auto-enable threshold for Sonnet/Opus models: >=1024 estimated prompt tokens.
Auto-enable threshold for Haiku models: >=2048 estimated prompt tokens.
Caching directives are applied only when thresholds are met and the runtime toggle is enabled.

Configuration commands:

/config get prompt_cache
/config get prompt_cache_ttl
/config set prompt_cache on
/config set prompt_cache_ttl 1h

See Prompt Caching Reference for full behavior and execution coverage.

Provider detection order

On startup, JD.AI checks providers in this order. The first with a valid connection becomes the default:

Claude Code — local CLI session
GitHub Copilot — GitHub authentication
OpenAI Codex — Codex CLI authentication
Ollama — local server at localhost:11434
Local (LLamaSharp) — GGUF files in ~/.jdai/models/
OpenAI — OPENAI_API_KEY or secure store
Azure OpenAI — AZURE_OPENAI_API_KEY + AZURE_OPENAI_ENDPOINT
Anthropic — ANTHROPIC_API_KEY or secure store
Google Gemini — GOOGLE_AI_API_KEY or secure store
Mistral — MISTRAL_API_KEY or secure store
AWS Bedrock — AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY + AWS_REGION
HuggingFace — HUGGINGFACE_API_KEY or secure store
OpenRouter — OPENROUTER_API_KEY or secure store
OpenAI-Compatible — custom endpoints via env vars or secure store

Credential resolution chain

API-key providers resolve credentials in this priority order:

Priority	Source	Example
1	Secure credential store	`~/.jdai/credentials/` — encrypted with DPAPI (Windows) or AES (Linux/macOS)
2	Configuration	`appsettings.json` → `Providers:{name}:{field}`
3	Environment variables	`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, etc.

The /provider add wizard stores keys in the secure credential store. Environment variables are picked up automatically.

Environment variables per provider

Variable	Provider	Description
`OPENAI_API_KEY`	OpenAI / Codex	OpenAI API key
`AZURE_OPENAI_API_KEY`	Azure OpenAI	Azure OpenAI API key
`AZURE_OPENAI_ENDPOINT`	Azure OpenAI	Azure OpenAI endpoint URL
`ANTHROPIC_API_KEY`	Anthropic	Anthropic API key
`GOOGLE_AI_API_KEY`	Google Gemini	Google AI API key
`MISTRAL_API_KEY`	Mistral	Mistral API key
`AWS_ACCESS_KEY_ID`	AWS Bedrock	AWS access key ID
`AWS_SECRET_ACCESS_KEY`	AWS Bedrock	AWS secret access key
`AWS_REGION`	AWS Bedrock	AWS region (default: `us-east-1`)
`HUGGINGFACE_API_KEY`	HuggingFace	HuggingFace Inference API key
`HF_TOKEN`	HuggingFace	HuggingFace API token (legacy)
`OPENROUTER_API_KEY`	OpenRouter	OpenRouter API key
`CODEX_TOKEN`	OpenAI Codex	Codex CLI access token override
`OLLAMA_ENDPOINT`	Ollama	Ollama API URL (default: `http://localhost:11434`)
`JDAI_MODELS_DIR`	Local Models	Model storage directory (default: `~/.jdai/models/`)
`HF_HOME`	HuggingFace	HuggingFace cache directory

OpenAI-Compatible sub-providers

These providers are auto-detected via environment variables and use the OpenAI connector with a custom base URL:

Provider	Env Variable	Base URL	Notable Models
Groq	`GROQ_API_KEY`	`https://api.groq.com/openai/v1`	Llama 3, Mixtral
Together AI	`TOGETHER_API_KEY`	`https://api.together.xyz/v1`	Open-source models
DeepSeek	`DEEPSEEK_API_KEY`	`https://api.deepseek.com/v1`	DeepSeek Coder, Chat
Fireworks AI	`FIREWORKS_API_KEY`	`https://api.fireworks.ai/inference/v1`	Fast inference
Perplexity	`PERPLEXITY_API_KEY`	`https://api.perplexity.ai`	Search-augmented models

Custom endpoints can also be configured interactively:

/provider add openai-compat   # Prompts for alias, base URL, and API key

Supports unlimited named instances — each with its own alias, URL, and key. Works with self-hosted endpoints (vLLM, LocalAI, LiteLLM, etc.).

Codex credential resolution

The OpenAI Codex provider has its own extended credential chain:

API key (direct)
Access token
OPENAI_API_KEY environment variable
CODEX_TOKEN environment variable
~/.codex/auth.json file
Device code login (interactive)

Provider comparison: privacy and cost

Feature	Claude Code	Copilot	Codex	Ollama	Local	OpenAI	Azure	Anthropic	Gemini	Mistral	Bedrock	HuggingFace	OpenRouter	OAI-Compat
Auth	OAuth	OAuth	OAuth	None	File	Key	Key	Key	Key	Key	AWS	Key	Key	Key
Internet	Yes	Yes	Yes	No	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Cost	Sub	Sub	Sub	Free	Free	PPU	PPU	PPU	PPU	PPU	PPU	Free/Pay	PPU	Varies
Privacy	Cloud	Cloud	Cloud	Local	Local	Cloud	Cloud	Cloud	Cloud	Cloud	Cloud	Cloud	Cloud	Varies

Sub = Subscription, PPU = Pay-per-use

Managing providers at runtime

/providers               # List all detected providers with status
/provider                # Interactive picker
/provider list           # Detailed provider list
/provider add openai     # Configure interactively
/provider remove openai  # Remove stored credentials
/provider test           # Test all connections
/provider test openai    # Test specific provider
/models                  # List all available models
/model gpt-4o            # Switch to a specific model

Mid-session model switching

When switching models during a conversation, JD.AI prompts for a transition mode:

Mode	Description
Preserve	Keep full conversation history as-is
Compact	Summarize conversation before switching
Transform	Re-format messages for the new model's style
Fresh	Start clean (history discarded)
Cancel	Abort the switch

Each switch creates a fork point in session history. Use /history to view fork points and double-ESC to roll back.

Table of Contents