helixmind
v0.8.0
Published
HelixMind – AI Coding CLI with Persistent Spiral Memory that Compounds
Maintainers
Readme
HelixMind
Open Source AI Coding Agent with Persistent Spiral Memory
An autonomous coding agent with memory that compounds — every session makes the next one smarter, with provenance to keep that smartness honest.
Installation -- Memory that Compounds -- Open Source vs Connected -- Features -- Quick Start
Memory that Compounds (v0.8.0)
Other AI coding tools start every session from zero. HelixMind doesn't.
$ helixmind chat
> fix the auth bug we discussed yesterday
[agent recalls: "auth bug = missing rate limit on /api/login (file middleware/auth.ts:42)"]
[runs validation, applies fix, runs tests]
🧠 +3 patterns learned · 1 stale retired · confidence 0.61↑0.64Spiral Memory stores diagnoses, patterns, and architectural notes across
sessions. Provenance keeps the memory honest: when a file changes, every
node tied to that file is automatically marked [STALE] so the agent
re-investigates instead of trusting cached reasoning. The result is a
persistent context that gets sharper over time without becoming a museum
of obsolete claims.
| | Other CLI agents | HelixMind | |---|---|---| | Session 1 | starts cold | starts cold | | Session 5 | starts cold | recalls 247 patterns relevant to your project | | File refactored between sessions | re-investigates from zero | recalls and flags stale references | | Brain growth visible? | n/a | session-exit line shows what was learned |
Multi-provider (Anthropic, OpenAI, Ollama, custom OpenAI-compatible endpoints), self-hostable, AGPL-3.0.
Installation
# Run directly (no install needed)
npx helixmind
# Or install globally
npm install -g helixmind
helixmindRequirements
- Node.js 18+
- An API key for at least one provider: Anthropic, OpenAI, or Ollama (local, no key needed)
Open Source vs Connected
When you start HelixMind for the first time, you choose how to use it:
+----------------------------------------------------------------------------+
| Welcome to HelixMind |
| |
| Login - free, unlock everything Open Source - no account needed |
| |
| [v] Jarvis Task Daemon [v] AI Agent + 22 Tools |
| [v] Validation Matrix [v] Spiral Memory |
| [v] Security Monitor [v] Anthropic/OpenAI/Ollama |
| [v] Autonomous Mode [x] No Jarvis / No Validation |
| [v] 3D Brain Management [x] No Monitor / No Security |
| [v] 3 Brains + Live WS [x] No Brain Management |
| |
| No credit card - Free forever - works offline |
+----------------------------------------------------------------------------+
-> [1] Login / [2] Open Source:| | Open Source | Connected (Free Account) | |---|---|---| | AI Agent with 22 Tools | Yes | Yes | | Spiral Memory (5 levels) | Yes | Yes | | Anthropic / OpenAI / Ollama | Yes | Yes | | Checkpoints and Undo | Yes | Yes | | Bug Journal | Yes | Yes | | Browser Automation | Yes | Yes | | Web Knowledge Enrichment | Yes | Yes | | Swarm (parallel workers) | Yes | Yes | | MCP Server | Yes | Yes | | Jarvis Task Daemon | - | Yes | | Validation Matrix | - | Yes | | Security Monitor | - | Yes | | 3D Brain Management | - | Yes | | Autonomous Mode | - | Yes | | Web Dashboard Remote Control | - | Yes |
Both modes are fully functional coding agents. The open source mode includes everything you need for day-to-day AI-assisted development. Connected mode adds advanced orchestration and monitoring features via a free account at helix-mind.ai.
You can switch anytime by running helixmind login.
Features
22 Agent Tools
HelixMind operates as an autonomous coding agent with built-in tools:
| Tool | Description |
|:-----|:------------|
| read_file / write_file / edit_file | Read, create, and modify files |
| list_dir / search_files / find_files | Navigate and search the codebase |
| run_command | Execute shell commands (sandboxed) |
| git_status / git_commit / git_diff / git_log | Full Git integration |
| spiral_store / spiral_query / spiral_status | Interact with spiral memory |
| web_research | Search the internet and extract knowledge |
| bug_report / bug_list | Track and manage bugs |
| browser_navigate / browser_click / browser_type / browser_screenshot / browser_read / browser_eval | Control a headless browser |
All tools go through a 3-tier permission system: auto (safe operations), ask (confirm first), dangerous (requires explicit approval). Use --yolo to auto-approve everything.
Spiral Memory
A 5-level hierarchical memory that persists across sessions:
Level 1 - Focus Most relevant, recent context
Level 2 - Active Related files, dependencies, web knowledge
Level 3 - Reference Decisions, patterns, code structure
Level 4 - Archive Compressed summaries, old sessions
Level 5 - Deep Long-term knowledge, project historyContext flows between levels based on relevance, recency, and usage. Web knowledge enrichment automatically fetches and stores relevant information from the internet as you work.
Swarm Mode
For complex multi-task requests, HelixMind automatically decomposes the work and spawns parallel worker sessions:
# Auto-detected when you send multi-part requests
> "Create a user model, add CRUD endpoints, and write tests for each"
# Or force it manually
/swarm refactor auth module, update all tests, fix the CI pipelineEach worker gets its own session with file-level locking to prevent conflicts. Results are merged and summarized when all workers complete.
Checkpoints
Every tool call and chat message creates an automatic checkpoint. Double-tap ESC to open the checkpoint browser and revert to any previous state (chat history, code changes, or both).
Bug Journal
Bugs are automatically detected from your messages (in English and German) and tracked in .helixmind/bugs.json. Each bug gets evidence, status, and persists across sessions.
Browser Automation
Headless Chrome integration via Puppeteer for web testing, scraping, and visual analysis:
/browser https://example.com
> "Click the login button and fill in the form"
> "Take a screenshot and analyze the layout"Multi-Provider Support
| Provider | Models | Setup |
|:---------|:-------|:------|
| Anthropic | Claude Sonnet, Opus, Haiku | ANTHROPIC_API_KEY=sk-ant-... |
| OpenAI | GPT-4o, GPT-4, o1, o3 | OPENAI_API_KEY=sk-... |
| Ollama | Any local model | No key needed (auto-detected) |
Switch providers at runtime with /model.
MCP Server
HelixMind includes a Model Context Protocol server for integration with other tools:
- Claude Code
- Cursor
- VS Code (Copilot)
- Windsurf
- JetBrains AI
Quick Start
# Start interactive chat
helixmind
# Initialize in a project (creates .helixmind/)
helixmind init
# Send a single message
helixmind chat -m "fix all lint errors in src/"
# YOLO mode - auto-approve everything
helixmind chat --yolo
# Skip permission prompts
helixmind chat -s
# Feed files into spiral memory
helixmind feed src/ --deep
# Watch files and update spiral live
helixmind feed src/ --watchCLI Commands
| Command | Description |
|:--------|:------------|
| helixmind | Start interactive chat (default) |
| helixmind init | Initialize HelixMind in project |
| helixmind chat -m "..." | Send a single message |
| helixmind chat --yolo | Auto-approve all operations |
| helixmind chat -s | Skip permission prompts |
| helixmind feed [paths...] | Feed files into spiral |
| helixmind feed --deep | Deep analysis with intent detection |
| helixmind feed --watch | Watch and auto-update spiral |
| helixmind spiral status | Show spiral metrics |
| helixmind spiral search <query> | Search spiral context |
| helixmind spiral compact | Trigger spiral compaction |
| helixmind config set <k> <v> | Set config value |
| helixmind config list | Show all config |
| helixmind export [dir] | Export spiral to .helixmind.zip |
| helixmind import <zip> | Import spiral from archive |
| helixmind login | Authenticate (switch to Connected mode) |
| helixmind logout | Remove stored auth |
Slash Commands (inside chat)
| Command | Description |
|:--------|:------------|
| /help | Show all commands |
| /model | Switch AI provider/model |
| /spiral | Show spiral status |
| /brain | Open 3D brain visualization |
| /context | Show assembled context |
| /tokens | Token usage stats |
| /undo | Undo last file change |
| /diff | Show current file changes |
| /git | Git operations |
| /sessions | Manage background sessions |
| /browser | Start browser session |
| /bugs | View bug journal |
| /swarm <msg> | Force swarm execution |
| /auto | Start autonomous mode |
| /security | Run security audit |
| /yolo | Toggle YOLO mode |
| /exit | Quit |
Keyboard Shortcuts
| Key | Action |
|:----|:-------|
| ESC ESC | Emergency stop / checkpoint browser |
| Ctrl+C | Exit |
| Ctrl+L | Clear screen |
| Ctrl+PageUp/Down | Switch session tabs |
| Tab | Autocomplete command |
Configuration
Environment Variables
ANTHROPIC_API_KEY=sk-ant-... # For Claude
OPENAI_API_KEY=sk-... # For OpenAI/GPT
# Ollama: no key needed (auto-detected on localhost:11434)Project Config
Create .helixmind/config.json in your project root:
{
"provider": "anthropic",
"model": "claude-sonnet-4-20250514"
}Or use the CLI:
helixmind config set provider anthropic
helixmind config set model claude-sonnet-4-20250514Tech Stack
| Component | Technology | |:----------|:-----------| | Language | TypeScript (strict, ESM) | | AI Providers | Anthropic SDK, OpenAI SDK, Ollama | | Database | better-sqlite3 + sqlite-vec (vector search) | | Embeddings | HuggingFace Transformers (MiniLM-L6-v2) | | Browser | Puppeteer (headless Chrome) | | Testing | Vitest (979 tests) | | MCP | Model Context Protocol SDK |
Development
git clone https://github.com/DancingTedDanson011/HelixMind.git
cd HelixMind
npm install
npm run build
npm run dev # Dev mode with watch
npm test # Run 979 testsLicense
AGPL-3.0 -- Free for open-source use. Commercial licenses available.
