🧬 Axolotl — Visual AI-Agent Orchestration

"Draw logic, don't write it"

Axolotl is a visual platform for building and executing AI-agent workflows. Design workflows from nodes on an infinite canvas, connect LLM providers, manage memory via MemPalace, and run execution in real-time with full trajectory visibility.

---

✨ Features

Visual Editor

• 🎨 Infinite Canvas — drag-and-drop, zoom, pan via VueFlow
• 🧩 11 Node Types: Source, Agent, Output, Condition, Loop, Memory, Guardrail, Human, Fallback, Subagent, Group
• 🔗 Typed Edges — data, condition true/false, loop
• ↩️ Undo/Redo — Cmd+Z / Cmd+Shift+Z
• 📋 Copy/Paste/Duplicate — Cmd+C / Cmd+V / Cmd+D
• 🔍 Search — Cmd+F by node name or type
• 📦 JSON Import/Export — full schema exchange
• 📊 Mermaid Export — diagrams from schema
• 🐍 Python Export — generate executable .py scripts
• 📷 PNG/SVG Export — canvas screenshots
• 📝 Comments — text notes on canvas
• 📦 Node Grouping — group 2+ nodes (Ctrl+G)

Workflow Execution

• ⚡ Parallel Execution — independent branches via CompletableFuture
• 📡 WebSocket Real-time — progress, logs, tokens, metrics, waves
• 🔄 Token Streaming — character-by-character LLM response
• 🚦 Convergence Monitoring — error counter, threshold 3 → BLOCKED
• 🛑 Cancel Execution — stop on demand
• 📊 Execution History — records of past runs
• 🎭 Execution Modes: EXECUTE (full), ANALYZE (read-only), DRY_RUN (simulate)

Tool-Enabled Agents

• 🔧 8 Built-in Tools: file_read, file_write, directory_read, bash, memory_read, memory_write, web_search, web_fetch
• 🧠 RLM Predict Tool — call sub-LM with DSPy signatures for structured extraction
• 📊 Trajectory Panel — visualize iterations, tool calls, timing
• 🔒 Dangerous Command Blocking — rm -rf, format, mkfs, etc.
• ⚙️ Tool Permissions — per-node allowed paths, blocked commands

Agent Memory & Skills

• 🎯 Agent Types: Assistant, Coder, Researcher, Reviewer, Project Analyzer
• 🧠 Skill Auto-Generation — extract patterns from trajectories → auto-generate skills
• 📈 Skill Tracking — usage count, success rate, version history
• 💾 Pattern Storage — saved to MemPalace (axolotl/patterns, axolotl/skills)

LLM Integration

• 🦙 Ollama — local models, NDJSON streaming
• 🤖 OpenAI — GPT-4o/mini, SSE streaming
• 🧠 Anthropic — Claude Sonnet/Opus/Haiku
• 🔍 DeepSeek — budget-friendly model
• 🔗 Custom Endpoints — add OpenAI-compatible providers
• 🎯 Per-Node Model — each AgentNode selects its own model

MemPalace — Long-term Memory

• 🧠 Memory Node — search memory, filter by wing/room
• 💾 Auto-save — agent results → MemPalace automatically
• 🌐 Graph Visualization — Memory Graph View with wings, rooms, tunnels
• 📊 Graph Context — structured tree + table → injection into systemPrompt
• 🔎 Semantic Search — cosine similarity
• 🗂️ Memory Result Cards — search results as floating cards on canvas

Plan / Workspace

• 📋 Todo List — tasks with statuses and priorities
• ✍️ Batch Add — mass-add via textarea
• ✅ Acceptance Criteria — validation on DONE transition
• 🔗 Node Linking — bind task to canvas node
• 🤖 MCP Server — 7 tools via JSON-RPC 2.0 at /mcp
• 🧩 Skills — auto-learning system with usage tracking and success rate

Observability

• 📊 Prometheus Metrics — /actuator/prometheus
• 📚 OpenAPI/Swagger — /swagger.html
• 📝 Structured JSON Logging — Logstash encoder for ELK/Loki

Remote API & Integrations

• 🔑 API Keys — management for external systems
• 📡 Remote API — /api/remote/* for workflow triggers
• ⚡ Rate Limiting — 60 req/min per key
• 🔗 Webhook Callbacks — completion notifications
• 📤 Share Links — read-only links with expiration

Subagent Workflows

• 🤝 Subagent Node — invoke nested workflows
• 🔄 Input/Output Mapping — data passing between workflows
• 🛡️ Max Depth — recursion protection (5 levels)
• 📜 Nested Logs — indented logs in Execution Panel

Security

• 🔐 JWT Authorization — registration/login
• 👥 Multi-tenancy — schema isolation per user
• 🔑 Settings API — CRUD provider API keys
• 🛡️ Guardrail Node — data validation/transformation
• 👤 Human Node — wait for human confirmation

UI/UX

• 🌙 Dark Theme — #1e1e2e background, #6c63ff accent
• ⌨️ Command Palette — Cmd+K quick access
• 🎓 Onboarding — 2-step wizard on first visit
• 🎬 Animations — pulse running, glow completed, shake failed
• 🔔 Toast Notifications — feedback on actions
• 💾 Auto-save Indicator — visual save status

Desktop App (Electron)

• 🖥️ Native Window — 1400x900, min 1024x700
• 🧭 System Tray — show/hide, new workflow, quit
• 📋 Application Menu — File, Edit, View, Window, Help
• ⌨️ Global Shortcut — Cmd/Ctrl+Shift+A toggle visibility
• 🔔 Native Notifications — execution complete, errors
• 💾 File Dialogs — native open/save for workflows
• 🔄 Auto-update — electron-updater from GitHub releases

---

🚀 Quick Start

Requirements

• Java 21+
• Node.js 18+
• (optional) Ollama for local LLM

Backend

cd backend
mvn spring-boot:run
# http://localhost:8080

Frontend

cd frontend
npm install
npm run dev
# http://localhost:5173

Docker Compose (full stack)

docker-compose up -d

---

🛠️ Tech Stack

Backend

Technology	Version
Java	21
Spring Boot	3.2
SQLite	3.x
PostgreSQL	(Docker)
WebSocket	Spring
Micrometer	Prometheus
springdoc	OpenAPI 3.0

Frontend

Technology	Version
Vue	3 (Composition API)
TypeScript	5.x
VueFlow	1.x
Vite	5.x
Pinia	State management
Playwright	E2E testing
Electron	34.x

---

📐 Architecture

Execution Flow

Source → Agent → Condition → Loop → Output
         ↓
      Memory (MemPalace)
         ↓
      Subagent (nested workflow)

1. Topological Sort (Kahn's algorithm) — compute levels
2. Parallel Execution of nodes at same level
3. WebSocket Events: progress, log, error, nodeTime, token, wave, toolCall, iteration
4. Context Management: collect upstream results + LLM summarization
5. Variable Interpolation: {{input}}, {{prev_result}}, {{node:Name}}
6. Execution Modes: EXECUTE/ANALYZE/DRY_RUN

---

📁 Project Structure

Axolotl/
├── backend/                          # Spring Boot 3.2, Java 21
│   └── src/main/java/.../
│       ├── controller/              # REST endpoints
│       ├── service/                # Business logic
│       ├── llm/                    # LLM providers
│       ├── model/                  # Domain objects
│       └── config/                 # Security, JWT, WebSocket
├── frontend/                         # Vue 3 + TypeScript + Vite
│   └── src/
│       ├── components/
│       │   ├── canvas/             # WorkflowCanvas
│       │   ├── nodes/              # Node types
│       │   └── execution/          # ExecutionPanel
│       └── stores/                  # Pinia
├── electron/                         # Electron desktop app
├── .github/workflows/               # CI/CD
│   ├── ci.yml                     # Compile & build
│   └── release.yml                # Docker → GHCR
├── kubernetes/axolotl/             # Helm chart
├── e2e/                           # Playwright tests
├── CONTRIBUTING.md
├── CODE_OF_CONDUCT.md
├── DEPLOY.md
└── docker-compose.yml

---

🧪 Tests

# Backend tests
cd backend && mvn test

# Frontend unit tests
cd frontend && npm run test:unit

# E2E tests
cd e2e && npx playwright test

---

📊 Implementation Status

Category	Implemented	Total
Visual Editor	✅	14
Workflow Execution	✅	10
LLM Providers	✅	7
Tool-Enabled Agents	✅	6
MemPalace	✅	6
Plan / MCP	✅	7
Skills System	✅	5
Remote API	✅	6
Security	✅	6
UI/UX	✅	8
Desktop App	✅	8
Infrastructure	✅	5
Total		100

---

🎯 Key Differentiators

Feature	Axolotl	n8n	LangFlow
Infinite canvas to chat	✅	❌	❌
Tool-enabled agents	✅	❌	❌
Trajectory visualization	✅	❌	❌
Memory as graph	✅	❌	❌
Auto-learning Skills	✅	❌	❌
Execution modes	✅	❌	❌
Built-in Plan/Todo	✅	❌	❌
Human-in-the-loop core	✅	basic	basic
Local-first privacy	✅	❌	❌
Desktop App	✅	✅	❌

---

🤝 Contributing

See CONTRIBUTING.md for:

• Development setup
• PR process
• Coding standards
• Commit message format

---

📝 License

MIT