huanchong-99/SoloDawn
Catches and repairs AI hallucinations before delivery: three quality gates, 31 built-in rules, a self-healing loop and a 90-point acceptance review. Fully automated from requirement to merged code. A primary Agent dynamically spawns and orchestrates child Agents on the AI CLIs on your machine (Claude Code, Codex — 8 supported).
SoloDawn – Automated AI‑driven software development
What it is – SoloDawn is an open‑source web application (Rust backend + React frontend) that turns a natural‑language description of a product into a complete, production‑ready codebase. It does this by orchestrating the real AI‑powered command‑line tools you already have installed (Claude Code, Codex, etc.) rather than generating code itself.
How it works
- A primary “orchestrator” agent watches a Git repository. When you submit a requirement, it turns the request into a technical spec and a scoring rubric.
- The orchestrator spawns child agents on demand. Each child runs an AI CLI in a native PTY (terminal) and performs a concrete task – writing code, reviewing, fixing bugs, merging, etc.
- Work is split into parallel Git branches (one per task). After every commit the code passes three layers of quality gates (terminal → branch → repo). If a gate fails, the child agent receives structured fix instructions and retries automatically.
- When all tasks meet the rubric (≥ 90 points) the changes are merged to the main branch and the product is delivered.
Key highlights
| Feature | Why it matters |
|---|---|
| Two‑layer agent architecture | The primary agent decides when and how many child agents to create, eliminating rigid, pre‑written workflows. |
| Native CLI integration | SoloDawn runs the actual AI CLIs you already use, preserving all plugins, skills, and MCP server settings – no re‑configuration needed. |
| Fully automated or manual mode | You can let the system run end‑to‑end, or intervene with a custom workflow graph (choose terminals, models, slash commands, etc.). |
| Quality system | 31 built‑in rules, three gated checkpoints, self‑healing loops and a 90‑point acceptance rubric keep hallucinations, security holes and merge conflicts in check. |
| Scalable parallelism | Up to ~10 concurrent task branches, each isolated in its own worktree, allowing large projects to be built faster. |
| Deployable with Docker | One‑click installer script sets up a server‑runner‑Redis stack; the UI is reachable at http://localhost:23457. |
Typical workflow
- Describe what you need (plain language or a precise spec).
- SoloDawn creates a technical spec + rubric and starts the orchestrator.
- The orchestrator decomposes the spec into tasks, creates Git branches, and launches child agents that run the appropriate AI CLI.
- After each commit the quality gate engine validates the output; failures trigger automatic repair cycles.
- When all tasks score ≥ 90, SoloDawn auto‑merges the changes and presents the finished product.
What you need to run it
- Rust nightly (2025‑12‑04) toolchain
- Node ≥ 18 with pnpm 10.13.1
- Git
- Any of the supported AI CLIs (Claude Code, Codex, Amp, Cursor Agent, Qwen Code, GitHub Copilot, Droid, Opencode). The system can mix up to eight CLIs in a single workflow.
Current status
- Version 1.0 has passed a 48‑hour fully‑autonomous test covering greenfield builds, migrations, refactoring and security hardening, delivering 7/7 tasks with an average score of 88.85 / A.
- Documentation includes a “Run It For Me” prompt you can paste into any coding AI to get SoloDawn installed and started.
Who it’s for
- Solo developers or one‑person startups who want a “virtual development team.”
- Teams that already use AI‑assisted CLIs and want to automate the end‑to‑end workflow without rewriting their tooling.
Quick‑start snippet (copy‑paste into an AI assistant)
Help me install and RUN SoloDawn locally. I want to USE it, not modify its source.
Repository: https://github.com/huanchong-99/SoloDawn
Clone: git clone https://github.com/huanchong-99/SoloDawn.git
Detect my OS (Windows / Linux / macOS) and do whatever is needed to get the web UI open at http://localhost:23457:
1. Install Rust nightly‑2025‑12‑04, Node ≥ 18, pnpm 10.13.1, Git.
2. Run `./install.sh` (or the Docker installer) and start the server.
3. Open the browser at the URL above.
Related
- Project
- Project
- Project
- Project
- Project