Table of Contents

Configuration Reference

JD.AI is configured through JSON files, project instruction files, slash commands, and CLI flags. No configuration is required to get started — sensible defaults apply out of the box.

Configuration precedence

Settings are resolved in this priority order (highest wins):

Priority Source Scope Set via
1 CLI flags Process --provider, --model, --system-prompt, etc.
2 Session state Session /model set, /provider, /autorun, /permissions
3 Per-project defaults Project /default project provider, /default project model
4 Global defaults User /default provider, /default model
5 Environment variables System OPENAI_API_KEY, OLLAMA_ENDPOINT, etc.
6 Built-in defaults Application Hard-coded fallbacks

Global configuration: ~/.jdai/config.json

Full schema

{
  "defaults": {
    "provider": "openai",
    "model": "gpt-4o"
  },
  "projectDefaults": {
    "/home/user/projects/my-app": {
      "provider": "ollama",
      "model": "llama3.2:latest"
    },
    "C:\\Projects\\enterprise-api": {
      "provider": "azure-openai",
      "model": "gpt-4o"
    }
  }
}

Fields

Field Type Description
defaults object Global default provider and model
defaults.provider string? Provider identifier (e.g. "openai", "ollama", "anthropic")
defaults.model string? Model identifier (e.g. "gpt-4o", "llama3.2:latest")
projectDefaults object Per-project overrides, keyed by absolute project path
projectDefaults.<path>.provider string? Project-specific provider override
projectDefaults.<path>.model string? Project-specific model override

All fields are optional. Null/missing values fall through to the next priority level.

Setting defaults via commands

/default                           # Show current defaults (global + project)
/default provider openai           # Set global default provider
/default model gpt-4o              # Set global default model
/default project provider ollama   # Set per-project default provider
/default project model llama3.2    # Set per-project default model

Per-project defaults: .jdai/defaults.json

Per-project defaults are stored in .jdai/defaults.json in the project root directory:

{
  "defaultProvider": "ollama",
  "defaultModel": "llama3.2:latest"
}

These override global defaults when JD.AI is launched from that project directory.

Project instructions: JDAI.md

JDAI.md is a Markdown file placed in the repository root. JD.AI reads it at session start and injects its contents into the system prompt.

File search order

JD.AI searches for instruction files in priority order. All discovered files are merged, with JDAI.md taking highest priority:

Priority File Notes
1 JDAI.md JD.AI native
2 CLAUDE.md Recognized instruction filename
3 AGENTS.md Recognized instruction filename
4 .github/copilot-instructions.md Recognized instruction filename
5 .jdai/instructions.md Dot-directory variant

Example JDAI.md

# 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/

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

Content guidelines

✅ 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

Viewing loaded instructions

/instructions    # Show all loaded instruction content

AtomicConfigStore behavior

~/.jdai/config.json is managed by the AtomicConfigStore class, which provides:

File locking

  • Cross-process safety: File-level lock via config.json.lock
  • In-process safety: SemaphoreSlim for concurrent access
  • Retry policy: 5 retries with exponential backoff (50ms → 100ms → 200ms → 400ms → 800ms)

Atomic writes

  1. Current config is backed up to config.json.bak
  2. New config is written to config.json.tmp
  3. Temp file is atomically moved to config.json

Corruption recovery

  • If config.json is empty or contains invalid JSON, a new empty config is returned
  • The .bak file can be manually restored if needed
  • Round-trip validation: serialized JSON is deserialized before writing to catch serialization bugs

Read behavior

File exists + valid JSON  → deserialized config
File exists + empty       → empty config (defaults)
File exists + invalid     → empty config (defaults)
File missing              → empty config (defaults)

Skills, plugins, and hooks

JD.AI uses native .jdai locations for skills/plugins and keeps .claude skills as lower-precedence legacy inputs.

Skills

Path Scope
<install>/skills/ Bundled baseline skills
~/.jdai/skills/ User-managed skills
.jdai/skills/ Workspace skills
~/.claude/skills/ Legacy skills (lower precedence)
.claude/skills/ Legacy skills (lower precedence)

Skill runtime policy/config:

Path Scope
~/.jdai/skills.json User skill lifecycle config
.jdai/skills.json Workspace overrides

Plugins and hooks

Path Scope
~/.jdai/plugins/ Personal plugins
.jdai/plugins/ Project plugins
~/.jdai/hooks.json Hook profiles and toggles

See Skills and Plugins for lifecycle, precedence, and gating details.

Data directory structure

~/.jdai/
├── config.json          # Global defaults (managed by AtomicConfigStore)
├── config.json.bak      # Backup of previous config
├── config.json.lock     # File lock (transient)
├── skills/              # User-managed skills
├── skills.json          # User skills lifecycle config
├── runtime/
│   └── skills/          # Active staged skills for current runtime
├── credentials/         # Encrypted credential store
├── sessions.db          # SQLite session database
├── update-check.json    # NuGet update cache (24h TTL)
├── exports/             # Exported session JSON files
└── models/              # Local GGUF models
    └── registry.json    # Model manifest

Runtime configuration commands

Command Effect
/autorun Toggle auto-approve for tool execution
/permissions Toggle all permission checks
/compact Force context compaction
/clear Clear conversation history
/spinner [style] Change spinner animation style
/config list Show persisted runtime settings
/config get <key> Read a persisted runtime setting
/config set <key> <value> Write a persisted runtime setting

/config keys

Key Meaning Example
theme Terminal theme token /config set theme nord
vim_mode Vim editing mode /config set vim_mode on
output_style Output renderer mode /config set output_style compact
spinner_style Spinner/progress style /config set spinner_style rich
prompt_cache Auto prompt caching for supported providers /config set prompt_cache on
prompt_cache_ttl Prompt cache TTL (5m or 1h) /config set prompt_cache_ttl 1h
autorun Auto-run tool confirmation behavior /config set autorun off
permissions Global permission checks /config set permissions on
plan_mode Plan mode state /config set plan_mode off
welcome_model_summary Show provider/model summary in welcome panel /config set welcome_model_summary on
welcome_services Show daemon/gateway indicators in welcome panel /config set welcome_services on
welcome_cwd Show current working directory in welcome panel /config set welcome_cwd off
welcome_version Show JD.AI version in welcome panel /config set welcome_version on
welcome_motd Enable or disable welcome MoTD line /config set welcome_motd on
welcome_motd_url MoTD source URL (raw text endpoint; use none to clear) /config set welcome_motd_url https://raw.githubusercontent.com/<org>/<repo>/main/docs/motd.txt

Note: output_style=json is session-only and will not persist as startup default.

Update workflow settings (tui-settings.json)

The /update workflow reads these settings from updates:

  • updates.enabled (bool)
  • updates.allowPromptTrigger (bool)
  • updates.requireApproval (bool)
  • updates.components.daemon (bool)
  • updates.components.gateway (bool)
  • updates.components.tui (bool)
  • updates.drainTimeout (timespan)
  • updates.reconnectTimeout (timespan)

These settings control whether update orchestration is enabled, whether prompt-triggered flows are allowed, whether apply needs approval, which components participate, and drain/reconnect timing windows.

Prompt caching defaults

  • prompt_cache=on
  • prompt_cache_ttl=5m
  • Optional extended TTL: /config set prompt_cache_ttl 1h

See Prompt Caching Reference for provider support, thresholds, and behavior.

See also

Edit this page