MORNINGSTAR
Watch it think.
Everything you need. Nothing you don't.
36 features. 28 brand-new capabilities rivaling Claude Code.
7 AI Providers
DeepSeek, OpenAI, Anthropic, Google, Ollama, Groq, OpenRouter — switch mid-session with /model
12 Built-in Tools
read, write, edit, delete, bash, grep, glob, ls, git, web, fetch, gh — with undo, diffs & permissions
Skills System
Reusable prompt-based skills loaded from .md files — auto-trigger on keywords, stack with agents
MCP Integration
Model Context Protocol — connect external tool servers, expose Morningstar as MCP server
8 Permission Modes
auto, ask, strict, bypassPermissions, acceptEdits, plan, dontAsk, delegate — granular access control
Extended Thinking
4 effort levels (low → ultra) with provider-specific thinking params — Anthropic, OpenAI, DeepSeek, Google
File-Based Agents
Define agents as .md files with YAML frontmatter — custom tools, models, temperature per agent
Rules System
Project rules from .md files with @import, path patterns, and priority — auto-injected into system prompt
Enhanced Hooks
14 hook events — preToolExecution, postFileWrite, preBash, sessionStart & more with JSON decision control
Agent Teams
Orchestrate multiple agents with roles (lead, worker, reviewer) — shared task queue & sequential execution
Native Function Calling
API-level tool definitions — OpenAI tools format, Anthropic tool_use blocks, Google functionCall — no regex parsing
Real Token Counting
Actual token usage from provider APIs with reasoning tokens, cache hits — not estimates. Per-model cost breakdown
Task Progress Checklist
Live Claude Code-style checklist — ✔ completed, ◼ in-progress, ◻ pending — with duration, tokens, turn counter
Auto-Test After Changes
Detects vitest, jest, pytest, cargo test, go test — runs relevant tests after file writes and feeds results back to AI
Sub-Agent Delegation
/delegate <agent> <task> — specialized agents execute autonomously with own conversation, tools, and turn limits
Parallel Tool Execution
Independent tool calls execute concurrently — up to 10 agentic turns with auto-fix on errors
Context Radar
Real-time context window monitor — see token usage per segment (System, Chat, Code, Tools) with color-coded bar
Animated Diff
Code morphing animation — old lines fade out in red, new lines type in green with typewriter effect
Dependency Graph
ASCII architecture visualization with /graph — scans imports, builds layered topology with box-drawing
Sandbox Mode
OS-level sandboxing — macOS sandbox-exec, Linux firejail — restrict filesystem access to project directory
Plugin System
Discover, install & manage plugins — registerTool, registerAgent, registerSkill, registerHook API
VS Code Highlighting
Shiki-powered syntax highlighting with the vitesse-dark theme — same engine as VS Code
Memory & History
Persistent notes + input history across sessions, todo system with priorities, save & restore conversations
Smart Auto-Fallback
Cloud API blocked? Auto-fallback to local models. Local too weak? Upgrades to cloud. Seamless multi-tier routing
Git Checkpoints
Create, list & restore checkpoints — pairs git state with conversation history for full rewind
AI Image Generation
Nano Banana Pro (Gemini 3) + RealVisXL local fallback — 6 models, auto-fallback, no content filters, saves to ~/Downloads
File Watcher
Watches project files for external changes — auto-triggers tests, notifies AI of modified files with debouncing
Conversation Branching
Fork conversations at any point, create named branches, switch between them, merge insights back
Streaming Diffs
Real-time diff preview as the AI generates edit operations — see changes character by character
Self-Update
/update check + /update run — pulls latest from GitHub, rebuilds, notifies of new versions
Web Dashboard
Browser-based dashboard at /dashboard — session stats, cost tracking, tool usage, conversation history
Multi-File Context
10 @-mention types with smart chunking — @files:glob, @diff:file, @tree — context budget management with auto-truncation
Smart PR Review
/pr-review — fetch, analyze, and review GitHub Pull Requests via gh CLI with structured severity ratings
Prompt Caching
Provider-aware prompt caching — Anthropic cache_control breakpoints, OpenAI prefix caching, Google context caching with stats
Terminal Multiplexer
Split-pane layouts (default, split, triple, quad) with focus navigation, scrolling, and Unicode box-drawing borders
7 Themes
Default, Ocean, Hacker, Sunset, Nord, Rose, Claude — from bold colors to minimalist monochrome
7 Providers. Any Model.
--model deepseek-reasoner--model o3--model claude-opus-4-20250514--model gemini-2.0-flash--model morningstar--model llama-3.3-70b-versatile--model openrouter/autoOur Models on Hugging Face
3 purpose-built open-weight models for code generation, architecture design, and vision. Free to download, fine-tune, and deploy. Run locally with Ollama.
huggingface.co/kurdman991Benchmark Comparison
Morningstar 14B vs popular open-source coding models
Morningstar 14B beats the base Qwen2.5-14B across all benchmarks, and outperforms CodeLlama-34B and DeepSeek-33B by a wide margin despite being smaller.
Built for Developers
40+ Programming Languages
First-class support for every major language. Morningstar understands idioms, conventions, and best practices for each one.
19 Interface Languages
The CLI speaks your language. Full UI translation including prompts, errors, and help text. RTL support included.
Built for Speed
A CLI experience designed for power users. Every interaction is optimized for developer velocity.
Interactive Mode
Full autocomplete, syntax highlighting, and inline suggestions.
Streaming Output
Watch AI responses stream in real-time with live formatting.
Live Diff View
Review every change before it is applied with highlighted diffs.
Keyboard Shortcuts
Power-user shortcuts for maximum velocity.
Extended Thinking
4 effort levels with provider-specific thinking parameters.
Checkpoints & Rewind
Git-based checkpoints paired with conversation state.
Vim Mode
Native Vim keybindings for the input editor.
! Bash Mode
Prefix with ! to execute shell commands directly.
Skills & Rules
Reusable .md-based skills and project rules with auto-trigger.
MCP & Plugins
Connect external tool servers and install community plugins.
Sandbox Mode
OS-level sandboxing restricts filesystem access.
Export & Copy
Export conversations as markdown, copy last response.
AI Image Generation
Nano Banana Pro (Gemini 3) + local RealVisXL with auto-fallback.
Persistent History
Input history persists across sessions. Arrow keys to navigate.
Auto-Test
Detects test runners and runs relevant tests after file changes.
Sub-Agent Delegation
Delegate tasks to specialized agents with own conversation context.
Task Progress
Live checklist tracking agentic loop turns, tokens, and duration.
Native Function Calling
API-level tool definitions instead of regex parsing.
File Watcher
Auto-detects directories, watches for external changes.
Conversation Branching
Fork, switch, and merge conversation branches.
Self-Update
Check for updates and pull latest from GitHub.
Web Dashboard
Browser-based session dashboard with live stats.
Multi-File Context
10 @-mention types with smart chunking and context budget.
Smart PR Review
Fetch, analyze, and review GitHub PRs with severity ratings.
Prompt Caching
Provider-aware caching with hit stats and cost savings.
Terminal Multiplexer
Split-pane layouts with focus navigation and scrolling.
Security First
Your code is your most valuable asset. Morningstar treats it that way.
8 Permission Modes
From strict approval to full bypass — auto, ask, strict, bypassPermissions, acceptEdits, plan, dontAsk, delegate.
OS-Level Sandbox
macOS sandbox-exec and Linux firejail restrict filesystem access to your project directory only.
Encrypted Session Memory
Session data is encrypted at rest. Conversation history never leaves your machine.
Local-Only Mode
Run entirely offline with Ollama. Zero data transmitted to external servers.
Injection Prevention
Built-in prompt injection detection and sanitization for all AI interactions.
Hook-Based Access Control
14 hook events with JSON decision control — pre/post validation on every tool execution.
Per-Project Settings
Isolated allow/deny lists, tool restrictions, and command filters per project via .morningstar/settings.local.json.
Get started in 30 seconds.
Requires Node.js 22+
export DEEPSEEK_API_KEY=sk-...OPENAI_API_KEY=sk-proj-...First launch prompts you