Table of Contents

Configuration

JD.AI is configured through project instruction files and runtime commands. No external config files or environment variables are required to get started — sensible defaults apply out of the box.

Project instructions (JDAI.md)

JDAI.md is a special file that JD.AI reads at the start of every session. Place it in your repository root to declare build commands, code style rules, and project-specific conventions. JD.AI injects its contents into the system prompt so every response respects your project's standards.

File search order

JD.AI searches for instruction files in this priority order:

  1. JDAI.md — JD.AI native format
  2. CLAUDE.md — Claude Code compatibility
  3. AGENTS.md — Codex CLI compatibility
  4. .github/copilot-instructions.md — Copilot compatibility
  5. .jdai/instructions.md — dot-directory variant

All discovered files are merged into the system prompt, with JDAI.md taking the highest priority when directives overlap.

Writing effective instructions

A good instruction file is concise and focused on information the AI cannot infer from the code alone.

Good JDAI.md example:

# Build & Test
- Build: `dotnet build MyProject.slnx`
- Test: `dotnet test --filter "Category!=Integration"`
- Format: `dotnet format MyProject.slnx`
- Lint: build must pass with zero warnings

# Code Style
- File-scoped namespaces
- XML doc comments on all public APIs
- Async/await throughout (no .Result or .Wait())
- ILogger<T> for logging, never Console.WriteLine

# Git Conventions
- Conventional commits (feat:, fix:, chore:, etc.)
- PR branches: feature/, fix/, chore/
- Always rebase on main before merging

# Project Notes
- Authentication module is in src/Auth/ — uses JWT
- Database migrations: `dotnet ef database update`

What to include:

✅ Include ❌ Exclude
Build/test commands Obvious language conventions
Code style rules that differ from defaults Standard patterns the AI already knows
Project-specific conventions Detailed API documentation
Architecture decisions File-by-file code descriptions
Environment quirks Information that changes frequently

View loaded instructions

/instructions    # Show all loaded instruction content

Directory structure

JD.AI stores local state in ~/.jdai/:

~/.jdai/
├── sessions.db          # SQLite session database
├── update-check.json    # NuGet update cache
├── exports/             # Exported session JSON files
└── models/              # Local GGUF models and registry
    └── registry.json    # Model manifest

Data directories

JD.AI stores local state and models in the following directories:

~/.jdai/
├── sessions.db          # SQLite session database
├── update-check.json    # NuGet update cache
├── exports/             # Exported session JSON files
└── models/              # Local GGUF models and registry
    └── registry.json    # Model manifest

Skills, plugins, and hooks

JD.AI loads Claude Code extensions from standard locations:

  • ~/.claude/skills/ — Personal skills
  • ~/.claude/plugins/ — Personal plugins
  • .claude/skills/ — Project skills
  • .claude/plugins/ — Project plugins

These extensions are registered as Semantic Kernel functions and filters at startup.

See Skills and Plugins for details.

Runtime configuration

Auto-approve mode

/autorun        # Toggle auto-approve for tools
/permissions    # Toggle all permission checks

Context management

/compact        # Force context compaction
/clear          # Clear conversation history

Environment variables

Variable Description Default
OLLAMA_ENDPOINT Ollama API URL http://localhost:11434
OLLAMA_CHAT_MODEL Default Ollama chat model llama3.2:latest
OLLAMA_EMBEDDING_MODEL Default embedding model all-minilm:latest
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 and registry directory ~/.jdai/models/
HF_HOME HuggingFace cache directory (for local model scanning) ~/.cache/huggingface/
HF_TOKEN HuggingFace API token for authenticated access —
Edit this page