The Autonomous Layer
For AI Coding Agents
Sugar manages your task queue, runs 24/7, and ships working code while you focus on what matters. Works with Claude Code, OpenCode, Aider, and more.
The Problem
AI coding assistants are powerful, but they need constant supervision.
Context Amnesia
Re-explaining project requirements every session. AI forgets everything between conversations.
Manual Breakdown
Splitting tasks into tiny steps yourself. You become the project manager for AI.
Constant Monitoring
Watching every action to ensure quality. You can't walk away and trust it.
Reactive, Not Proactive
AI waits for instructions. It doesn't discover work or identify what needs to be done.
You spend more time managing AI than benefiting from it.
The Solution
Sugar = AI Agent + Persistence + Autonomy
Instead of one-off interactions, Sugar works like a development team.
Core Principles
Built for developers who want AI that actually works autonomously.
Delegation, Not Babysitting
You define what and why. Sugar figures out how and executes. Rich context gives AI everything it needs.
Continuous Execution
Sugar runs 24/7 through a priority queue. It picks up the next task and keeps going until the queue is empty.
Specialized Agents
Orchestrator coordinates. Planner breaks down features. Guardian enforces quality. Like a real team.
Agent-Agnostic
Works with Claude Code, OpenCode, Aider, or any CLI-based AI agent. Use /sugar-task to delegate work mid-conversation.
Smart Discovery
Sugar finds work: error logs, code quality issues, missing tests, GitHub issues. It doesn't just wait.
Rich Task Context
Business context, technical requirements, success criteria. More context = better autonomous execution.
Updated in v3.7 Agent Integrations
OpenCode Integration
UPDATEDOne-command setup with OpenCode, the open-source AI coding agent with 80k+ stars. Run sugar opencode setup and you're done.
- • MCP servers for tasks & memory
- • 11 slash commands including
/sugar-remember - • Persistent memory across sessions
- • Automatic config detection
Claude Code Plugin
Native integration with Claude Code via MCP servers and slash commands. Delegate tasks mid-conversation with /sugar-add.
- • MCP server for memory & tasks
- • Slash command integration
- • SessionStart hooks
- • Context auto-injection
New in v3.5 Memory System
Persistent Memory
Remember decisions, preferences, and error patterns across sessions. No more context loss.
Semantic Search
Find relevant context with AI-powered search. sugar recall "auth" finds authentication memories.
MCP Servers
Expose memory and tasks to any MCP-compatible agent. Works with Claude Code, OpenCode, and more.
Six Memory Types
Decisions, preferences, file context, error patterns, research, outcomes. Organized and searchable.
v3.4 Security & Control
Tool Restrictions
Tier-based security controls which tools are available per task type.
Bash Permissions
Wildcard patterns control command execution. Dangerous commands always blocked.
Task Hooks
Pre/post execution hooks with variable substitution for automation.
Thinking Capture
View Claude's reasoning via sugar thinking for debugging.
How It Works
The Sugar Loop: continuous autonomous development.
↻ Repeat continuously until queue is empty
You Assign
Add tasks with priorities and rich context. Define what needs to be built and why.
Sugar Picks Up
Grabs highest priority work from the queue. No prompting needed.
AI Agent Executes
Runs in background. Uses specialized agents as needed. Tests and validates.
Completes & Repeats
Commits working code, moves to next task. Continuous until queue is empty.
Ralph Wiggum: Self-Correcting AI
Complex tasks need iteration. Ralph keeps going until tests actually pass.
Without Ralph
80% right isn't shipped code.
With Ralph
Iterates until it actually works.
The Iteration Loop
↻ Repeats until <promise>DONE</promise> or max iterations
When to Use Ralph
Complex Bugs
Iterates through debugging until the fix actually works
Refactoring
Self-corrects when changes break existing tests
TDD Features
Keeps implementing until all tests go green
# Enable iterative execution
sugar add "Implement rate limiting" --ralph --max-iterations 10
# Sugar iterates until tests pass
# Iteration 1: implement → test → 2/5 pass
# Iteration 2: fix → test → 4/5 pass
# Iteration 3: fix edge case → test → 5/5 pass
# Output: <promise>DONE</promise>
✓ Task completed after 3 iterationsIntelligent Triage
Not sure if your task needs Ralph? Let Sugar decide.
# Let Sugar analyze complexity and decide
sugar add "Refactor auth to use repository pattern" --triage
# Sugar analyzes:
# - Task complexity (keywords, scope, risk)
# - Codebase capabilities (pytest, eslint, etc.)
# - Generates completion criteria automatically
🔍 Triage: Ralph recommended (78% confidence)
Completion: "All tests pass, no lint errors"
✓ Task added with Ralph mode enabledTriage auto-enables Ralph when confidence is 60%+. Simple tasks run single-pass.
Who It's For
Sugar scales with your needs.
Solo Developers
Run Sugar while you sleep. Wake up to completed features and fixed bugs. Multiply your output without burning out.
Small Teams
Sugar becomes your always-on teammate. Handles the backlog while humans focus on architecture and product.
Enterprises
24/7 autonomous development with audit trails, multi-project isolation, and team collaboration.
Ready to Ship Faster?
Install Sugar in 30 seconds. Start delegating work to AI that actually executes.
pip install sugarai