Define your goal. Launch the dashboard. Watch AI agents plan, execute, and validate your software — with you in control.

Example: "Build a drag-and-drop image compressor" → 3 agents (developer, reviewer, designer) → Working app with tests passing → 45 minutes.

📺 Watch the 4-minute demo →

Sgai turns software development into a goal-driven, multi-agent workflow.

Instead of prompting step-by-step, you:

1. Define the outcome — "Build a music sequencer web app"
2. Agents plan the work — Breaking it into a visual workflow diagram of tasks
3. You supervise — Watch progress, answer questions when agents need guidance
4. Success checks — Tests, linting, or other validation determines "done"

Not autocomplete. Not a chat window. A local AI software factory.

• See what's happening — Visual workflow diagram instead of hidden AI reasoning
• Multiple specialists — Developer writes code, reviewer checks it, safety analyst validates
• Approve before execution — Review the plan and answer questions, then agents work autonomously
• Proof of completion — Tests must pass before work is marked done
• Works locally — Runs in your repository, nothing leaves your machine

opencode update
opencode auth login
opencode --model anthropic/claude-opus-4-6 run "install Sgai using the instructions from https://github.com/sandgardenhq/sgai/blob/main/INSTALLATION.md"

This runs the official installation guide automatically and launches a demo workspace.

Required: Go, Node.js, bun, opencode

Recommended: jj (version control), tmux (session management), ripgrep (code search), Graphviz (diagram rendering)

go install github.com/sandgardenhq/sgai/cmd/sgai@latest

Or build from source:

git clone https://github.com/sandgardenhq/sgai.git
cd sgai
cd cmd/sgai/webapp && bun install && cd ../../..
make build

See INSTALLATION.md for details.

Open: http://localhost:8080

📺 Prefer watching? See the demo → https://youtu.be/NYmjhwLUg8Q

Most users create goals using the built-in wizard.

Goals are stored in GOAL.md and describe outcomes — not implementation steps.

Example GOAL.md:

---
flow: |
  "backend-developer" -> "code-reviewer"
completionGateScript: make test
interactive: yes
---

# Build a REST API

Create endpoints for user registration and login with JWT auth.

- [ ] POST /register validates email, hashes password
- [ ] POST /login returns JWT token
- [ ] Tests pass before completion

See GOAL.example.md for full reference.

Sgai breaks your goal into a workflow diagram of coordinated agents with defined roles.

Dependencies are explicit. Execution is visible.

Before execution begins, agents ask clarifying questions about your goal.

Once you approve the plan, agents work autonomously — executing tasks, running tests, and validating completion.

You can:

• Monitor real-time progress (optional)
• Interrupt execution if needed
• Review diffs and session history
• Fork sessions to try different approaches

Most of the time, you approve the plan and come back when it's done.

Sgai extracts reusable skills and code snippets from completed sessions — your agents get smarter over time.

• Agents operate inside your local repository
• Changes go through your version control (we recommend jj, but Git works)
• Sgai does not automatically push to remote repositories

You stay in control.

Sgai accepts improvements as specifications inside GOALS/.

1. Create GOALS/YYYY_MM_DD_feature_name.md
2. Describe desired behavior and success criteria
3. Submit a PR

See the GOALS directory for examples.

Found a bug or have a feature request? Open an issue →

Want to discuss ideas or share what you built? Start a discussion →

Developer documentation lives in docs/, produced by Doc Holiday, of course!

https://github.com/sandgardenhq/sgai/blob/main/LICENSE

Sgai was created by Ulderico Cirello, and is maintained by Sandgarden.